Normalize MSR slideshow frame sizes

This reverts commit df4c936aae.

README.md

@@ -58,7 +58,7 @@ This node uses Ollama Locally : you will need to install it (very lightweight) a
 
 - You can still enter your description manually if you don't want to install it but it works very well!
 
-- Both references mode work (fixed), Licon MSR and Ghost Mask.
+- Reference mode included: Licon MSR.
 
 - Use their Lora, that is an important step : https://huggingface.co/LiconStudio/LTX-2.3-Multiple-Subject-Reference/tree/main
 

__init__.py

@@ -1,5 +1,7 @@
 from .ltx_director import LTXDirector
 from .ltx_director_guide import LTXDirectorGuide, LTXDirectorCropGuides
+from .msr_character import MSRCharacter, MSRCharacterSet
+from .timeline_nodes import TimelineSection, TimelineSectionSet
 from comfy_api.latest import ComfyExtension, io
 from typing_extensions import override
 from .latent_slice import CleanLatentSlice
@@ -8,6 +10,10 @@ class PromptRelay(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[io.ComfyNode]]:
         return [
+            MSRCharacter,
+            MSRCharacterSet,
+            TimelineSection,
+            TimelineSectionSet,
             LTXDirector,
         ]
 
@@ -16,16 +22,24 @@ async def comfy_entrypoint() -> PromptRelay:
 
 NODE_CLASS_MAPPINGS = {
     "LTXDirectorCS": LTXDirector,
+    "MSRCharacterCS": MSRCharacter,
+    "MSRCharacterSetCS": MSRCharacterSet,
+    "TimelineSectionCS": TimelineSection,
+    "TimelineSectionSetCS": TimelineSectionSet,
     "LTXDirectorGuideCS": LTXDirectorGuide,
     "LTXDirectorCropGuidesCS": LTXDirectorCropGuides,
     "CleanLatentSliceCS": CleanLatentSlice,
 }
 
 NODE_DISPLAY_NAME_MAPPINGS = {
-    "LTXDirectorCS": "LTX Director CS",
-    "LTXDirectorGuideCS": "LTX Director Guide CS",
-    "LTXDirectorCropGuidesCS": "LTX Director Crop Guides CS",
-    "CleanLatentSliceCS": "Clean Latent Slice CS",
+    "LTXDirectorCS": "LTX Director DUMAS",
+    "MSRCharacterCS": "MSR Character DUMAS",
+    "MSRCharacterSetCS": "MSR Character Set DUMAS",
+    "TimelineSectionCS": "Timeline Section DUMAS",
+    "TimelineSectionSetCS": "Timeline Section Set DUMAS",
+    "LTXDirectorGuideCS": "LTX Director Guide DUMAS",
+    "LTXDirectorCropGuidesCS": "LTX Director Crop Guides DUMAS",
+    "CleanLatentSliceCS": "Clean Latent Slice DUMAS",
 }
 
 WEB_DIRECTORY = "./js"

js/ltx_director_guide.js

@@ -3,22 +3,28 @@ import { app } from "../../scripts/app.js";
 // LTX Director Guide is a pure pass-through processor node.
 // All configuration (images, insert frames, strengths) comes from
 // the guide_data output of Prompt Relay Encode (Timeline).
+function hideWidget(node, widgetName) {
+    const widget = node.widgets?.find((entry) => entry.name === widgetName);
+    if (!widget) return;
+
+    widget.hidden = true;
+    widget.options = widget.options || {};
+    widget.options.hidden = true;
+
+    if (!window.LiteGraph || !window.LiteGraph.vueNodesMode) {
+        widget.computeSize = () => [0, -4];
+        widget.draw = () => {};
+    }
+
+    if (widget.element) widget.element.style.display = "none";
+}
+
 app.registerExtension({
     name: "Comfy.LTXDirectorGuideCS",
     async nodeCreated(node) {
         if (node.comfyClass !== "LTXDirectorGuideCS") return;
 
         // Hide retake_mode widget on LiteGraph as it is dynamically auto-detected from the timeline data.
-        const w = node.widgets?.find(x => x.name === "retake_mode");
-        if (w) {
-            w.hidden = true;
-            if (!w.options) w.options = {};
-            w.options.hidden = true;
-            if (!window.LiteGraph || !window.LiteGraph.vueNodesMode) {
-                w.computeSize = () => [0, -4];
-                w.draw = () => { };
-            }
-            if (w.element) w.element.style.display = "none";
-        }
+        hideWidget(node, "retake_mode");
     },
 });

js/msr_character.js

@@ -0,0 +1,619 @@
+const { app } = window.comfyAPI.app;
+const { api } = window.comfyAPI.api;
+const DEFAULT_ANALYZE_PROMPT = "Describe the character's physical appearance in two concise sentences. Specify their hair color/style, face details, and their clothing type/color. Keep the entire response very brief.";
+
+function findWidget(node, name) {
+  return (node.widgets || []).find((widget) => widget.name === name);
+}
+
+function findInput(node, name) {
+  return (node.inputs || []).find((input) => input.name === name);
+}
+
+function getOriginNodeForInput(node, inputName) {
+  const input = findInput(node, inputName);
+  if (!input || input.link == null) return null;
+  const link = app.graph.links[input.link];
+  if (!link) return null;
+  return app.graph.getNodeById(link.origin_id);
+}
+
+function coercePreviewUrl(value) {
+  if (!value) return "";
+  if (typeof value === "string") return value;
+  if (typeof value === "object") {
+    if (typeof value.src === "string") return value.src;
+    if (typeof value.url === "string") return value.url;
+  }
+  return "";
+}
+
+function resolveFileWidgetUrlFromOrigin(originNode) {
+  if (!originNode) return "";
+
+  const widgets = originNode.widgets || [];
+  const fileWidget = widgets.find((widget) => ["image", "filename", "file"].includes(widget.name));
+  const subfolderWidget = widgets.find((widget) => widget.name === "subfolder");
+  const filename = typeof fileWidget?.value === "string" ? fileWidget.value : "";
+  const subfolder = typeof subfolderWidget?.value === "string" ? subfolderWidget.value : "";
+  if (!filename) return "";
+
+  return api.apiURL(
+    `/view?filename=${encodeURIComponent(filename)}&type=input&subfolder=${encodeURIComponent(subfolder)}`
+  );
+}
+
+function resolvePreviewUrlFromOrigin(originNode) {
+  if (!originNode) return "";
+  const preview = Array.isArray(originNode.imgs) ? coercePreviewUrl(originNode.imgs[0]) : "";
+  return preview || "";
+}
+
+function resolveCandidateImageUrls(originNode) {
+  const urls = [];
+  const fileUrl = resolveFileWidgetUrlFromOrigin(originNode);
+  const previewUrl = resolvePreviewUrlFromOrigin(originNode);
+  if (fileUrl) urls.push(fileUrl);
+  if (previewUrl && previewUrl !== fileUrl) urls.push(previewUrl);
+  return urls;
+}
+
+async function blobToOptimizedDataUrl(blob, maxDim = 1024, quality = 0.82) {
+  const objectUrl = URL.createObjectURL(blob);
+  try {
+    const image = await new Promise((resolve, reject) => {
+      const img = new Image();
+      img.onload = () => resolve(img);
+      img.onerror = () => reject(new Error("Failed to decode image blob"));
+      img.src = objectUrl;
+    });
+
+    const width = image.naturalWidth || image.width || 0;
+    const height = image.naturalHeight || image.height || 0;
+    const scale = width > 0 && height > 0 ? Math.min(1, maxDim / Math.max(width, height)) : 1;
+    const targetWidth = Math.max(1, Math.round(width * scale)) || width || 1;
+    const targetHeight = Math.max(1, Math.round(height * scale)) || height || 1;
+
+    const canvas = document.createElement("canvas");
+    canvas.width = targetWidth;
+    canvas.height = targetHeight;
+    const ctx = canvas.getContext("2d");
+    if (!ctx) {
+      throw new Error("Could not acquire canvas context");
+    }
+    ctx.drawImage(image, 0, 0, targetWidth, targetHeight);
+
+    const optimizedBlob = await new Promise((resolve, reject) => {
+      canvas.toBlob((result) => {
+        if (result) resolve(result);
+        else reject(new Error("Canvas toBlob failed"));
+      }, "image/jpeg", quality);
+    });
+
+    const dataUrl = await new Promise((resolve) => {
+      const reader = new FileReader();
+      reader.onloadend = () => resolve(reader.result);
+      reader.readAsDataURL(optimizedBlob);
+    });
+
+    return {
+      dataUrl,
+      originalWidth: width,
+      originalHeight: height,
+      outputWidth: targetWidth,
+      outputHeight: targetHeight,
+      outputBytes: optimizedBlob.size || 0,
+    };
+  } finally {
+    URL.revokeObjectURL(objectUrl);
+  }
+}
+
+async function imageInputToDataUrl(node, inputName) {
+  const originNode = getOriginNodeForInput(node, inputName);
+  const candidateUrls = resolveCandidateImageUrls(originNode);
+  if (!candidateUrls.length) {
+    return {
+      dataUrl: null,
+      debug: {
+        inputName,
+        originNodeId: originNode?.id ?? null,
+        originNodeType: originNode?.type ?? null,
+        candidateUrls: [],
+        selectedUrl: null,
+        mimeType: null,
+        blobBytes: 0,
+        dataUrlLength: 0,
+        originalWidth: 0,
+        originalHeight: 0,
+        outputWidth: 0,
+        outputHeight: 0,
+        outputBytes: 0,
+        error: "No candidate image URLs found",
+      },
+    };
+  }
+
+  for (const imageUrl of candidateUrls) {
+    try {
+      const response = await fetch(imageUrl);
+      if (!response.ok) {
+        throw new Error(`HTTP ${response.status}`);
+      }
+      const blob = await response.blob();
+      if (!blob.type.startsWith("image/")) {
+        throw new Error(`Unexpected blob type: ${blob.type || "unknown"}`);
+      }
+      const optimized = await blobToOptimizedDataUrl(blob);
+      const dataUrl = optimized.dataUrl;
+      return {
+        dataUrl,
+        debug: {
+          inputName,
+          originNodeId: originNode?.id ?? null,
+          originNodeType: originNode?.type ?? null,
+          candidateUrls,
+          selectedUrl: imageUrl,
+          mimeType: blob.type || null,
+          blobBytes: blob.size || 0,
+          dataUrlLength: typeof dataUrl === "string" ? dataUrl.length : 0,
+          originalWidth: optimized.originalWidth,
+          originalHeight: optimized.originalHeight,
+          outputWidth: optimized.outputWidth,
+          outputHeight: optimized.outputHeight,
+          outputBytes: optimized.outputBytes,
+          error: null,
+        },
+      };
+    } catch (error) {
+      console.warn("[MSRCharacter] Failed candidate image source", imageUrl, error);
+    }
+  }
+
+  return {
+    dataUrl: null,
+    debug: {
+      inputName,
+      originNodeId: originNode?.id ?? null,
+      originNodeType: originNode?.type ?? null,
+      candidateUrls,
+      selectedUrl: null,
+      mimeType: null,
+      blobBytes: 0,
+      dataUrlLength: 0,
+      originalWidth: 0,
+      originalHeight: 0,
+      outputWidth: 0,
+      outputHeight: 0,
+      outputBytes: 0,
+      error: "All candidate image URLs failed",
+    },
+  };
+}
+
+function setWidgetValue(node, widget, value) {
+  if (!widget) return;
+  const previous = widget.value;
+  widget.value = value;
+  if (widget.element) {
+    if (widget.element.type === "checkbox") {
+      widget.element.checked = !!value;
+    } else {
+      widget.element.value = value;
+    }
+  }
+  if (widget.callback) {
+    try {
+      widget.callback(value);
+    } catch (error) {}
+  }
+  if (node.onWidgetChanged) {
+    try {
+      node.onWidgetChanged(widget.name, value, previous, widget);
+    } catch (error) {}
+  }
+  if (node.properties) {
+    node.properties[widget.name] = value;
+  }
+}
+
+function syncFormFromWidgets(node) {
+  const description = findWidget(node, "description")?.value || "";
+  const alias = findWidget(node, "alias")?.value || "";
+  const provider = findWidget(node, "analyze_provider")?.value || "ollama";
+  const model = findWidget(node, "analyze_model")?.value || "";
+  const baseUrl = findWidget(node, "analyze_base_url")?.value || "";
+
+  if (node._msrAliasInput && node._msrAliasInput.value !== alias) {
+    node._msrAliasInput.value = alias;
+  }
+  if (node._msrDescriptionInput && node._msrDescriptionInput.value !== description) {
+    node._msrDescriptionInput.value = description;
+  }
+  if (node._msrMeta) {
+    node._msrMeta.innerHTML = [
+      alias
+        ? `Alias: <span style="color:#e8e8e8">@${String(alias).replace(/^@/, "")}</span>`
+        : "Alias: <span style=\"color:#666\">none</span>",
+      `<div style="margin-top:4px;color:#888">Analyze: ${provider}${model ? ` / ${model}` : ""}${baseUrl ? ` / ${baseUrl}` : ""}</div>`,
+      `<div style="margin-top:4px;color:#666">Description is stored in the field below.</div>`,
+    ].join("");
+  }
+}
+
+function refreshCharacterPreview(node) {
+  if (!node._msrPreviewRow) return;
+
+  const inputNames = ["image_1", "image_2"];
+  const previewUrls = inputNames
+    .map((inputName) => {
+      const originNode = getOriginNodeForInput(node, inputName);
+      const candidates = resolveCandidateImageUrls(originNode);
+      return candidates[0] || "";
+    })
+    .filter(Boolean);
+
+  const previewRow = node._msrPreviewRow;
+  previewRow.innerHTML = "";
+
+  if (previewUrls.length) {
+    previewUrls.forEach((previewUrl) => {
+      const image = document.createElement("img");
+      image.src = previewUrl;
+      Object.assign(image.style, {
+        flex: "1",
+        minWidth: "0",
+        height: "74px",
+        objectFit: "cover",
+        borderRadius: "6px",
+        border: "1px solid #444",
+        background: "#111",
+      });
+      previewRow.appendChild(image);
+    });
+  } else {
+    const placeholder = document.createElement("div");
+    placeholder.textContent = "Connect up to 2 image nodes";
+    Object.assign(placeholder.style, {
+      width: "100%",
+      height: "74px",
+      display: "flex",
+      alignItems: "center",
+      justifyContent: "center",
+      color: "#777",
+      fontSize: "11px",
+      borderRadius: "6px",
+      border: "1px dashed #444",
+      background: "#181818",
+    });
+    previewRow.appendChild(placeholder);
+  }
+
+  syncFormFromWidgets(node);
+}
+
+function buildCharacterUi(node) {
+  const container = document.createElement("div");
+  Object.assign(container.style, {
+    boxSizing: "border-box",
+    width: "100%",
+    padding: "6px 2px 2px 2px",
+  });
+
+  const previewRow = document.createElement("div");
+  Object.assign(previewRow.style, {
+    display: "flex",
+    gap: "6px",
+    minHeight: "74px",
+  });
+
+  const meta = document.createElement("div");
+  Object.assign(meta.style, {
+    marginTop: "8px",
+    fontSize: "11px",
+    color: "#aaa",
+    lineHeight: "1.4",
+  });
+
+  const form = document.createElement("div");
+  Object.assign(form.style, {
+    marginTop: "10px",
+    display: "flex",
+    flexDirection: "column",
+    gap: "6px",
+  });
+
+  const actionsRow = document.createElement("div");
+  Object.assign(actionsRow.style, {
+    display: "flex",
+    gap: "8px",
+    alignItems: "center",
+    flexWrap: "wrap",
+  });
+
+  const analyzeButton = document.createElement("button");
+  analyzeButton.type = "button";
+  analyzeButton.textContent = "Analyze";
+  Object.assign(analyzeButton.style, {
+    alignSelf: "flex-start",
+    background: "#2b4f38",
+    color: "#f3f3f3",
+    border: "1px solid #496d56",
+    borderRadius: "6px",
+    padding: "6px 10px",
+    fontSize: "11px",
+    cursor: "pointer",
+  });
+  analyzeButton.addEventListener("click", () => {
+    const widgetButton = (node.widgets || []).find((widget) => widget.msrAnalyze);
+    if (!widgetButton) {
+      alert("Analyze button widget is missing on this node.");
+      return;
+    }
+    analyzeCharacterNode(node, widgetButton);
+  });
+
+  const settingsButton = document.createElement("button");
+  settingsButton.type = "button";
+  settingsButton.textContent = "Analyze Settings";
+  Object.assign(settingsButton.style, {
+    alignSelf: "flex-start",
+    background: "#252525",
+    color: "#ddd",
+    border: "1px solid #444",
+    borderRadius: "6px",
+    padding: "6px 10px",
+    fontSize: "11px",
+    cursor: "pointer",
+  });
+  settingsButton.addEventListener("click", () => {
+    const providerWidget = findWidget(node, "analyze_provider");
+    const baseUrlWidget = findWidget(node, "analyze_base_url");
+    const modelWidget = findWidget(node, "analyze_model");
+    const promptWidget = findWidget(node, "analyze_prompt");
+    const provider = window.prompt("Analyze provider: ollama, lmstudio, custom, off", providerWidget?.value || "ollama");
+    if (provider == null) return;
+    const baseUrl = window.prompt("Analyze base URL (blank = default)", baseUrlWidget?.value || "");
+    if (baseUrl == null) return;
+    const model = window.prompt("Analyze model (blank = provider default)", modelWidget?.value || "");
+    if (model == null) return;
+    const prompt = window.prompt("Analyze prompt (blank = default)", promptWidget?.value || DEFAULT_ANALYZE_PROMPT);
+    if (prompt == null) return;
+    setWidgetValue(node, providerWidget, provider.trim() || "ollama");
+    setWidgetValue(node, baseUrlWidget, baseUrl.trim());
+    setWidgetValue(node, modelWidget, model.trim());
+    setWidgetValue(node, promptWidget, prompt.trim());
+    syncFormFromWidgets(node);
+    node.setDirtyCanvas?.(true, true);
+  });
+
+  const aliasLabel = document.createElement("label");
+  aliasLabel.textContent = "Alias";
+  Object.assign(aliasLabel.style, {
+    fontSize: "11px",
+    color: "#bbb",
+  });
+
+  const aliasInput = document.createElement("input");
+  aliasInput.type = "text";
+  aliasInput.placeholder = "optional alias, e.g. bob";
+  Object.assign(aliasInput.style, {
+    width: "100%",
+    boxSizing: "border-box",
+    background: "#1c1c1c",
+    color: "#eee",
+    border: "1px solid #444",
+    borderRadius: "6px",
+    padding: "6px 8px",
+    fontSize: "12px",
+  });
+  aliasInput.addEventListener("input", () => {
+    setWidgetValue(node, findWidget(node, "alias"), aliasInput.value);
+    syncFormFromWidgets(node);
+  });
+
+  const descLabel = document.createElement("label");
+  descLabel.textContent = "Description";
+  Object.assign(descLabel.style, {
+    fontSize: "11px",
+    color: "#bbb",
+  });
+
+  const descInput = document.createElement("textarea");
+  descInput.placeholder = "Character description...";
+  Object.assign(descInput.style, {
+    width: "100%",
+    minHeight: "86px",
+    boxSizing: "border-box",
+    resize: "vertical",
+    background: "#1c1c1c",
+    color: "#eee",
+    border: "1px solid #444",
+    borderRadius: "6px",
+    padding: "8px",
+    fontSize: "12px",
+    lineHeight: "1.35",
+  });
+  descInput.addEventListener("input", () => {
+    setWidgetValue(node, findWidget(node, "description"), descInput.value);
+    syncFormFromWidgets(node);
+  });
+
+  actionsRow.appendChild(analyzeButton);
+  actionsRow.appendChild(settingsButton);
+  form.appendChild(aliasLabel);
+  form.appendChild(aliasInput);
+  form.appendChild(descLabel);
+  form.appendChild(descInput);
+  form.appendChild(actionsRow);
+
+  container.appendChild(previewRow);
+  container.appendChild(meta);
+  container.appendChild(form);
+
+  node._msrPreviewContainer = container;
+  node._msrPreviewRow = previewRow;
+  node._msrMeta = meta;
+  node._msrAliasInput = aliasInput;
+  node._msrDescriptionInput = descInput;
+  node._msrAnalyzeButton = analyzeButton;
+  return container;
+}
+
+async function analyzeCharacterNode(node, buttonWidget) {
+  const descriptionWidget = findWidget(node, "description");
+  if (!descriptionWidget) return;
+  const providerWidget = findWidget(node, "analyze_provider");
+  const baseUrlWidget = findWidget(node, "analyze_base_url");
+  const modelWidget = findWidget(node, "analyze_model");
+  const promptWidget = findWidget(node, "analyze_prompt");
+
+  buttonWidget.label = "Analyzing...";
+  if (node._msrAnalyzeButton) {
+    node._msrAnalyzeButton.textContent = "Analyzing...";
+    node._msrAnalyzeButton.disabled = true;
+  }
+  node.setDirtyCanvas?.(true, true);
+
+  try {
+    const imageResults = await Promise.all([
+      imageInputToDataUrl(node, "image_1"),
+      imageInputToDataUrl(node, "image_2"),
+    ]);
+    const imageDebug = imageResults.map((result) => result?.debug || null).filter(Boolean);
+    const images = (
+      imageResults.map((result) => result?.dataUrl || null)
+    ).filter(Boolean);
+
+    if (!images.length) {
+      throw new Error("Connect at least one image input before analyzing.");
+    }
+
+    const response = await api.fetchApi("/ltx_director/analyze_character", {
+      method: "POST",
+      body: JSON.stringify({
+        provider: providerWidget?.value || "ollama",
+        base_url: baseUrlWidget?.value || "",
+        model: modelWidget?.value || "",
+        prompt: promptWidget?.value || "",
+        image_b64: images,
+        image_debug: imageDebug,
+      }),
+    });
+    const result = await response.json();
+    if (result.status !== "success") {
+      console.warn("[MSRCharacter] analyze debug", {
+        sentImages: imageDebug,
+        response: result,
+      });
+      throw new Error(result.message || "Unknown analysis error");
+    }
+
+    const description = result.description || "";
+    setWidgetValue(node, descriptionWidget, description);
+    if (node._msrDescriptionInput) {
+      node._msrDescriptionInput.value = description;
+    }
+    syncFormFromWidgets(node);
+    refreshCharacterPreview(node);
+  } catch (error) {
+    console.error("[MSRCharacter] analyze failed", error);
+    alert(`MSR Character analyze failed: ${error.message || error}`);
+  } finally {
+    buttonWidget.label = "Analyze with Ollama";
+    if (node._msrAnalyzeButton) {
+      node._msrAnalyzeButton.textContent = "Analyze";
+      node._msrAnalyzeButton.disabled = false;
+    }
+    node.setDirtyCanvas?.(true, true);
+  }
+}
+
+function hideNodeWidget(widget) {
+  if (!widget) return;
+  widget.hidden = true;
+  if (!widget.options) widget.options = {};
+  widget.options.hidden = true;
+  if (widget.element) {
+    widget.element.style.display = "none";
+  }
+}
+
+app.registerExtension({
+  name: "Comfy.MSRCharacterCS",
+  async beforeRegisterNodeDef(nodeType, nodeData) {
+    if (nodeData.name === "MSRCharacterCS") {
+      const originalOnNodeCreated = nodeType.prototype.onNodeCreated;
+      nodeType.prototype.onNodeCreated = function () {
+        if (originalOnNodeCreated) originalOnNodeCreated.apply(this, arguments);
+
+        if (!(this.widgets || []).find((widget) => widget.msrAnalyze)) {
+          this.addWidget("button", "Analyze with Ollama", null, () => {
+            analyzeCharacterNode(this, (this.widgets || []).find((widget) => widget.msrAnalyze));
+          }, { serialize: false });
+          const buttonWidget = this.widgets[this.widgets.length - 1];
+          buttonWidget.msrAnalyze = true;
+          buttonWidget.label = "Analyze with Ollama";
+        }
+
+        if (!findWidget(this, "analyze_provider")) {
+          this.addWidget("combo", "analyze_provider", "ollama", null, {
+            values: ["ollama", "lmstudio", "custom", "off"],
+          });
+        }
+        if (!findWidget(this, "analyze_base_url")) {
+          this.addWidget("text", "analyze_base_url", "");
+        }
+        if (!findWidget(this, "analyze_model")) {
+          this.addWidget("text", "analyze_model", "");
+        }
+        if (!findWidget(this, "analyze_prompt")) {
+          this.addWidget("text", "analyze_prompt", DEFAULT_ANALYZE_PROMPT);
+        }
+
+        if (!this._msrPreviewWidget) {
+          const previewContainer = buildCharacterUi(this);
+          this._msrPreviewWidget = this.addDOMWidget("msr_character_ui", "msr_character_ui", previewContainer, {
+            getValue: () => "",
+            setValue: () => {},
+          });
+          this._msrPreviewWidget.computeSize = () => [Math.max(10, (this.size?.[0] || 320) - 30), 260];
+        }
+
+        const originalConnectionsChange = this.onConnectionsChange;
+        this.onConnectionsChange = function () {
+          const result = originalConnectionsChange?.apply(this, arguments);
+          refreshCharacterPreview(this);
+          return result;
+        };
+
+        const originalConfigure = this.onConfigure;
+        this.onConfigure = function () {
+          const result = originalConfigure?.apply(this, arguments);
+          setTimeout(() => {
+            syncFormFromWidgets(this);
+            refreshCharacterPreview(this);
+          }, 0);
+          return result;
+        };
+
+        ["alias", "description", "analyze_provider", "analyze_base_url", "analyze_model", "analyze_prompt"].forEach((widgetName) => {
+          hideNodeWidget(findWidget(this, widgetName));
+        });
+        hideNodeWidget((this.widgets || []).find((widget) => widget.msrAnalyze));
+
+        this.size[0] = Math.max(this.size[0] || 0, 330);
+        syncFormFromWidgets(this);
+        refreshCharacterPreview(this);
+      };
+    }
+
+    if (nodeData.name === "MSRCharacterSetCS") {
+      const originalOnNodeCreated = nodeType.prototype.onNodeCreated;
+      nodeType.prototype.onNodeCreated = function () {
+        if (originalOnNodeCreated) originalOnNodeCreated.apply(this, arguments);
+        this.size[0] = Math.max(this.size[0] || 0, 260);
+      };
+    }
+  },
+});

load_audio_ui.py

@@ -1,8 +1,16 @@
 import folder_paths
+import numpy as np
 import os
 import torch
 import av
 
+DEFAULT_SAMPLE_RATE = 44100
+DEFAULT_SILENCE_SECONDS = 1
+
+
+def build_silence(sample_rate: int = DEFAULT_SAMPLE_RATE, seconds: int = DEFAULT_SILENCE_SECONDS, channels: int = 2) -> torch.Tensor:
+    return torch.zeros((channels, int(sample_rate * seconds)), dtype=torch.float32)
+
 def f32_pcm(wav: torch.Tensor) -> torch.Tensor:
     """Convert audio to float 32 bits PCM format."""
     if wav.dtype.is_floating_point:
@@ -23,18 +31,17 @@ def load_audio_file(filepath: str) -> tuple[torch.Tensor, int]:
         sr = stream.codec_context.sample_rate
         n_channels = stream.channels
 
-        frames = []
+        frame_arrays = []
         for frame in af.decode(streams=stream.index):
-            buf = torch.from_numpy(frame.to_ndarray())
-            if buf.shape[0] != n_channels:
-                buf = buf.view(-1, n_channels).t()
+            array = frame.to_ndarray()
+            if array.shape[0] != n_channels:
+                array = array.reshape(-1, n_channels).T
+            frame_arrays.append(array)
 
-            frames.append(buf)
-
-        if not frames:
+        if not frame_arrays:
             raise ValueError("No audio frames decoded.")
 
-        wav = torch.cat(frames, dim=1)
+        wav = torch.from_numpy(np.concatenate(frame_arrays, axis=1))
         wav = f32_pcm(wav)
         return wav, sr
 
@@ -99,17 +106,16 @@ class LoadAudioUI:
             missing_info = audio if audio != "none" else "None selected"
             print(f"!!! [LoadAudioUI] Warning: Audio file '{missing_info}' not found. Outputting 1 second of silence.")
 
-            sample_rate = 44100
-            # 1 second of silence (stereo) -> shape [channels, time]
-            waveform = torch.zeros((2, 44100))
+            sample_rate = DEFAULT_SAMPLE_RATE
+            waveform = build_silence(sample_rate)
         else:
             try:
                 waveform, sample_rate = load_audio_file(audio_path)
             except Exception as e:
                 # If decoding fails for any reason, fallback to silence rather than crashing the workflow
                 print(f"!!! [LoadAudioUI] Error decoding {audio}: {e}. Falling back to silence.")
-                sample_rate = 44100
-                waveform = torch.zeros((2, 44100))
+                sample_rate = DEFAULT_SAMPLE_RATE
+                waveform = build_silence(sample_rate)
 
         # Convert seconds to frames
         start_frame = int(start_time * sample_rate)

ltx_director_guide.py

@@ -10,8 +10,7 @@ import comfy.utils
 import folder_paths
 import node_helpers
 from comfy_extras import nodes_lt
-from comfy_api.latest import io
-from .ltx_director import GuideData, MotionGuideData, _resize_image, _execute_comfy_node, _unpack
+from .ltx_director import _resize_image, _execute_comfy_node, _unpack
 
 log = logging.getLogger(__name__)
 
@@ -57,6 +56,18 @@ def _clone_noise_mask(latent, latent_image):
         device=latent_image.device,
     )
 
+def _safe_json_loads(raw_value, default=None):
+    import json
+
+    if isinstance(raw_value, dict):
+        return raw_value
+    if not raw_value:
+        return {} if default is None else default
+    try:
+        return json.loads(raw_value)
+    except Exception:
+        return {} if default is None else default
+
 def _resize_latent_spatial(latent_image, noise_mask, width, height, method):
     b, c, f, h, w = latent_image.shape
     if width == w and height == h:
@@ -78,6 +89,65 @@ def _ceil_to_multiple(value, multiple):
     multiple = max(1, int(multiple))
     return int(math.ceil(value / multiple) * multiple)
 
+def _normalize_resize_method_for_encode(resize_method):
+    if resize_method == "maintain aspect ratio":
+        return "pad"
+    return resize_method
+
+def _append_latent_frames(latent_image, noise_mask, extra_frames, mask_fill=1.0):
+    extra_frames = int(max(0, extra_frames))
+    if extra_frames == 0:
+        return latent_image, noise_mask
+
+    batch, channels, _, height, width = latent_image.shape
+    latent_tail = torch.zeros(
+        (batch, channels, extra_frames, height, width),
+        dtype=latent_image.dtype,
+        device=latent_image.device,
+    )
+    latent_image = torch.cat([latent_image, latent_tail], dim=2)
+
+    mask_batch, mask_channels, _, mask_height, mask_width = noise_mask.shape
+    mask_tail = torch.full(
+        (mask_batch, mask_channels, extra_frames, mask_height, mask_width),
+        float(mask_fill),
+        dtype=noise_mask.dtype,
+        device=noise_mask.device,
+    )
+    noise_mask = torch.cat([noise_mask, mask_tail], dim=2)
+    return latent_image, noise_mask
+
+def _encode_resized_video_frames(
+    vae,
+    frames,
+    target_width,
+    target_height,
+    resize_method,
+    time_scale_factor,
+    use_tiled_encode,
+    tile_size,
+    tile_overlap,
+):
+    pixels = _resize_image(
+        frames,
+        target_width,
+        target_height,
+        _normalize_resize_method_for_encode(resize_method),
+        divisible_by=1,
+    )
+    num_frames_to_keep = ((pixels.shape[0] - 1) // time_scale_factor) * time_scale_factor + 1
+    encode_src = pixels[:num_frames_to_keep, :, :, :3]
+    if use_tiled_encode:
+        encoded = vae.encode_tiled(
+            encode_src,
+            tile_x=tile_size,
+            tile_y=tile_size,
+            overlap=tile_overlap,
+        )
+    else:
+        encoded = vae.encode(encode_src)
+    return pixels, encoded
+
 def _snap_latent_to_downscale(latent_image, noise_mask, downscale_factor, method):
     factor = int(max(1, round(float(downscale_factor))))
     if factor <= 1:
@@ -307,11 +377,413 @@ class LTXDirectorGuide:
     RETURN_NAMES = ("positive", "negative", "latent", "model", "latent_downscale_factor")
     FUNCTION = "execute"
 
+    @classmethod
+    def _maybe_add_attention(cls, positive, negative, is_lora_active, tokens_added, guide_orig_shape, attention_strength):
+        if not is_lora_active:
+            return positive, negative
+        positive = _append_guide_attention_entry(
+            positive,
+            tokens_added,
+            guide_orig_shape,
+            attention_strength=attention_strength,
+        )
+        negative = _append_guide_attention_entry(
+            negative,
+            tokens_added,
+            guide_orig_shape,
+            attention_strength=attention_strength,
+        )
+        return positive, negative
+
+    @classmethod
+    def _apply_crop_count(cls, positive, negative, crop_frames):
+        crop_frames = max(0, int(crop_frames))
+        values = {"nghtdrp_guide_crop_latent_frames": crop_frames}
+        positive = node_helpers.conditioning_set_values(positive, values)
+        negative = node_helpers.conditioning_set_values(negative, values)
+        return positive, negative
+
+    @classmethod
+    def _load_cached_motion_video_frames(cls, video_cache, video_file, trim_start_frames, length_frames, director_fps, resample_mode):
+        cache_key = (
+            str(video_file),
+            int(trim_start_frames),
+            int(length_frames),
+            float(director_fps),
+            str(resample_mode),
+        )
+        cached = video_cache.get(cache_key)
+        if cached is None:
+            cached = _load_motion_video_frames(
+                video_file,
+                trim_start_frames=trim_start_frames,
+                length_frames=length_frames,
+                director_fps=director_fps,
+                resample_mode=resample_mode,
+            )
+            video_cache[cache_key] = cached
+        return cached
+
+    @classmethod
+    def _encode_image_guide(cls, vae, latent_width, latent_height, img_tensor, scale_factors, target_pix_w, target_pix_h, upscale_method):
+        _, height, width, _ = img_tensor.shape
+        if target_pix_w != width or target_pix_h != height:
+            img_nchw = img_tensor.permute(0, 3, 1, 2)
+            img_resized = comfy.utils.common_upscale(
+                img_nchw,
+                target_pix_w,
+                target_pix_h,
+                upscale_method,
+                "disabled",
+            )
+            img_tensor = img_resized.permute(0, 2, 3, 1)
+        return nodes_lt.LTXVAddGuide.encode(
+            vae,
+            latent_width,
+            latent_height,
+            img_tensor,
+            scale_factors,
+        )
+
+    @classmethod
+    def _apply_image_guides(
+        cls,
+        positive,
+        negative,
+        vae,
+        latent_image,
+        noise_mask,
+        images,
+        insert_frames,
+        strengths,
+        latent_width,
+        latent_height,
+        latent_length,
+        scale_factors,
+        image_attention_strength,
+        is_lora_active,
+        upscale_method,
+    ):
+        target_pix_w = int(latent_width * 32)
+        target_pix_h = int(latent_height * 32)
+        for idx, img_tensor in enumerate(images):
+            frame_value = insert_frames[idx] if idx < len(insert_frames) else 0
+            strength = float(strengths[idx] if idx < len(strengths) else 1.0)
+            if strength <= 0.0:
+                continue
+
+            image_pixels, guide_latent = cls._encode_image_guide(
+                vae,
+                latent_width,
+                latent_height,
+                img_tensor,
+                scale_factors,
+                target_pix_w,
+                target_pix_h,
+                upscale_method,
+            )
+            frame_idx, latent_idx = nodes_lt.LTXVAddGuide.get_latent_index(
+                positive,
+                latent_length,
+                len(image_pixels),
+                int(frame_value),
+                scale_factors,
+            )
+            if latent_idx >= latent_length:
+                continue
+
+            max_frames = latent_length - latent_idx
+            if guide_latent.shape[2] > max_frames:
+                guide_latent = guide_latent[:, :, :max_frames]
+
+            tokens_added = guide_latent.shape[2] * guide_latent.shape[3] * guide_latent.shape[4]
+            guide_orig_shape = list(guide_latent.shape[2:])
+            positive, negative, latent_image, noise_mask = nodes_lt.LTXVAddGuide.append_keyframe(
+                positive,
+                negative,
+                frame_idx,
+                latent_image,
+                noise_mask,
+                guide_latent,
+                strength,
+                scale_factors,
+            )
+            positive, negative = cls._maybe_add_attention(
+                positive,
+                negative,
+                is_lora_active,
+                tokens_added,
+                guide_orig_shape,
+                image_attention_strength,
+            )
+
+        return positive, negative, latent_image, noise_mask
+
+    @classmethod
+    def _apply_motion_segment_guides(
+        cls,
+        positive,
+        negative,
+        vae,
+        latent_image,
+        noise_mask,
+        segments,
+        director_fps,
+        latent_length,
+        latent_width,
+        latent_height,
+        time_scale_factor,
+        scale_factors,
+        latent_downscale_factor,
+        crop,
+        use_tiled_encode,
+        tile_size,
+        tile_overlap,
+        active_resize_method,
+        is_lora_active,
+        video_cache,
+    ):
+        for seg in segments:
+            try:
+                video_file = seg.get("videoFile")
+                if not video_file:
+                    continue
+
+                start_frame = int(seg.get("start", 0))
+                length_frames = int(seg.get("length", 1))
+                trim_start = int(seg.get("trimStart", 0))
+                video_strength = float(seg.get("videoStrength", 1.0))
+                video_attention_strength = float(seg.get("videoAttentionStrength", 0.65))
+                if length_frames <= 0 or video_strength <= 0.0:
+                    continue
+
+                video_frames = cls._load_cached_motion_video_frames(
+                    video_cache,
+                    video_file,
+                    trim_start,
+                    length_frames,
+                    director_fps,
+                    seg.get("resampleMode", "nearest"),
+                )
+                num_frames_to_keep = ((video_frames.shape[0] - 1) // time_scale_factor) * time_scale_factor + 1
+                video_frames = video_frames[:num_frames_to_keep]
+                causal_fix = start_frame == 0 or num_frames_to_keep == 1
+                encode_frames = video_frames if causal_fix else torch.cat([video_frames[:1], video_frames], dim=0)
+
+                _, guide_latent = _encode_video_iclora_guide(
+                    vae,
+                    latent_width,
+                    latent_height,
+                    encode_frames,
+                    scale_factors,
+                    latent_downscale_factor,
+                    crop,
+                    use_tiled_encode,
+                    tile_size,
+                    tile_overlap,
+                    resize_method=active_resize_method,
+                )
+                if not causal_fix:
+                    guide_latent = guide_latent[:, :, 1:, :, :]
+
+                frame_idx = start_frame
+                latent_idx = (frame_idx + time_scale_factor - 1) // time_scale_factor if frame_idx > 0 else 0
+                if latent_idx >= latent_length:
+                    continue
+
+                if start_frame > 0 and guide_latent.shape[2] > 1:
+                    guide_latent = guide_latent[:, :, 1:, :, :]
+                    frame_idx += time_scale_factor
+                    latent_idx += 1
+                    if latent_idx >= latent_length:
+                        continue
+
+                max_frames = latent_length - latent_idx
+                if guide_latent.shape[2] > max_frames:
+                    guide_latent = guide_latent[:, :, :max_frames]
+
+                guide_orig_shape = list(guide_latent.shape[2:])
+                batch, _, frames, height, width = guide_latent.shape
+                guide_mask = torch.ones(
+                    (batch, 1, frames, height, width),
+                    device=guide_latent.device,
+                    dtype=guide_latent.dtype,
+                )
+                if start_frame > 0:
+                    ramp_steps = [0.25, 0.65]
+                    for i, step in enumerate(ramp_steps):
+                        if i < frames:
+                            guide_mask[:, :, i, :, :] = 1.0 + video_strength * (1.0 - step)
+
+                ldf = int(max(1, round(float(latent_downscale_factor))))
+                if ldf > 1:
+                    dilated = _dilate_latent(
+                        {"samples": guide_latent, "noise_mask": guide_mask},
+                        horizontal_scale=ldf,
+                        vertical_scale=ldf,
+                    )
+                    guide_latent = dilated["samples"]
+                    guide_mask = dilated["noise_mask"]
+
+                tokens_added = guide_latent.shape[2] * guide_latent.shape[3] * guide_latent.shape[4]
+                positive, negative, latent_image, noise_mask = nodes_lt.LTXVAddGuide.append_keyframe(
+                    positive,
+                    negative,
+                    frame_idx,
+                    latent_image,
+                    noise_mask,
+                    guide_latent,
+                    video_strength,
+                    scale_factors,
+                    guide_mask=guide_mask,
+                    latent_downscale_factor=float(latent_downscale_factor),
+                    causal_fix=causal_fix,
+                )
+                positive, negative = cls._maybe_add_attention(
+                    positive,
+                    negative,
+                    is_lora_active,
+                    tokens_added,
+                    guide_orig_shape,
+                    video_attention_strength,
+                )
+            except Exception as e:
+                raise RuntimeError(f"LTX Director Guide motion segment failed for {seg}: {e}") from e
+
+        return positive, negative, latent_image, noise_mask
+
+    @classmethod
+    def _execute_retake_mode(
+        cls,
+        positive,
+        negative,
+        vae,
+        latent_image,
+        noise_mask,
+        guide_data,
+        tdata,
+        segments,
+        model,
+        latent_downscale_factor,
+        director_fps,
+        latent_length,
+        latent_width,
+        latent_height,
+        time_scale_factor,
+        active_resize_method,
+        use_tiled_encode,
+        tile_size,
+        tile_overlap,
+        is_empty_latent,
+        initial_latent_length,
+        video_cache,
+    ):
+        print(
+            f"[LTXDirectorGuide] Retake Mode active. Preserving base latent, masking selected regions. "
+            f"is_empty_latent: {is_empty_latent}"
+        )
+        target_width = latent_width * 32
+        target_height = latent_height * 32
+        retake_start = int(tdata.get("retakeStart", 0))
+        retake_len = int(tdata.get("retakeLength", 0))
+        retake_strength = float(tdata.get("retakeStrength", 1.0))
+
+        start_frame = int(guide_data.get("start_frame", 0))
+        relative_start = max(0, retake_start - start_frame)
+        l_start = min(max(0, relative_start // time_scale_factor), latent_length)
+        l_end = min(
+            max(0, int(math.ceil((relative_start + retake_len) / time_scale_factor))),
+            latent_length,
+        )
+
+        need_base_video = is_empty_latent or l_start > 0 or l_end < latent_length
+        if not need_base_video:
+            print(
+                "[LTXDirectorGuide] Stage 2: Retake region covers the entire generated range. "
+                "Skipping base video loading and VAE encoding."
+            )
+
+        retake_vid_info = tdata.get("retakeVideo") or {}
+        video_file = retake_vid_info.get("imageFile", "") if isinstance(retake_vid_info, dict) else ""
+        if not video_file and not retake_vid_info and segments:
+            video_file = segments[0].get("videoFile", "")
+
+        if need_base_video and not video_file:
+            if retake_vid_info and not retake_vid_info.get("imageFile"):
+                raise ValueError(
+                    "Retake Mode is active, but the base video file upload is still in progress (or failed). "
+                    "Please wait for the 'Uploading base video...' overlay on the timeline to disappear before queuing the prompt."
+                )
+            raise ValueError(
+                "Retake Mode is active, but no base video has been selected on the timeline. "
+                "Please drag and drop or upload a base video on the timeline first."
+            )
+
+        if need_base_video and video_file:
+            try:
+                ltxv_length = (latent_length - 1) * time_scale_factor + 1
+                print(
+                    f"[LTXDirectorGuide] Loading and encoding base video file: {video_file} "
+                    f"starting at frame {start_frame} for length {ltxv_length} at resolution "
+                    f"{target_width}x{target_height}"
+                )
+                video_frames = cls._load_cached_motion_video_frames(
+                    video_cache,
+                    video_file,
+                    start_frame,
+                    ltxv_length,
+                    director_fps,
+                    "nearest",
+                )
+                _, base_latent = _encode_resized_video_frames(
+                    vae,
+                    video_frames,
+                    target_width,
+                    target_height,
+                    active_resize_method,
+                    time_scale_factor,
+                    use_tiled_encode,
+                    tile_size,
+                    tile_overlap,
+                )
+                base_latent = base_latent.to(device=latent_image.device, dtype=latent_image.dtype)
+
+                paste_len = min(base_latent.shape[2], latent_length)
+                if is_empty_latent:
+                    latent_image[:, :, :paste_len] = base_latent[:, :, :paste_len]
+                else:
+                    print(
+                        f"[LTXDirectorGuide] Stage 2: Copying high-resolution base latent for preserved "
+                        f"regions (0-{l_start} and {l_end}-{paste_len})"
+                    )
+                    if l_start > 0:
+                        latent_image[:, :, :l_start] = base_latent[:, :, :l_start]
+                    if l_end < paste_len:
+                        latent_image[:, :, l_end:paste_len] = base_latent[:, :, l_end:paste_len]
+            except Exception as e:
+                print(f"[LTXDirectorGuide] Failed to load/encode base video: {e}. Falling back to input latent.")
+
+        noise_mask = torch.zeros_like(noise_mask)
+        if l_end > l_start:
+            noise_mask[:, :, l_start:l_end] = retake_strength
+        print(f"[LTXDirectorGuide] noise_mask slice: {noise_mask[0, 0, :, 0, 0].tolist()}")
+
+        exact_crop_frames = max(0, int(latent_image.shape[2]) - initial_latent_length)
+        positive, negative = cls._apply_crop_count(positive, negative, exact_crop_frames)
+        return (
+            positive,
+            negative,
+            {"samples": latent_image, "noise_mask": noise_mask},
+            model,
+            float(latent_downscale_factor),
+        )
+
     @classmethod
     def execute(cls, positive, negative, vae, latent, guide_data, motion_guide_data=None, model=None, ic_lora_name="None", ic_lora_strength=1.0, scale_by=1.0, upscale_method="bicubic", image_attention_strength=1.0, crop="center", auto_snap_ic_grid=True, use_tiled_encode=False, tile_size=256, tile_overlap=64, retake_mode=False, msr_strength=0.0):
         motion_segments = (motion_guide_data or {}).get("segments", []) if motion_guide_data else []
         image_guides_count = len(guide_data.get("images", [])) if guide_data else 0
         print(f"[LTXDirectorGuide] execute started. motion_segments: {len(motion_segments)}, image_guides: {image_guides_count}, ic_lora_name: {ic_lora_name}, model connected: {model is not None}, retake_mode: {retake_mode}")
+        video_cache = {}
 
         active_resize_method = guide_data.get("resize_method") if guide_data else None
         if not active_resize_method:
@@ -336,24 +808,6 @@ class LTXDirectorGuide:
         if auto_snap_ic_grid and model is not None and ic_lora_name != "None":
             latent_image, noise_mask = _snap_latent_to_downscale(latent_image, noise_mask, latent_downscale_factor, upscale_method)
 
-        # Ghost Mask: the Director kept the latent clean (so it stays length-matched with the audio
-        # latent) and asked us to pre-extend it here by one hidden frame per reference. The tail
-        # sits past the clean region; the reference images (in guide_data["images"], anchored at
-        # (clean + i)) are written into it by the loop below, then dropped downstream by Clean
-        # Latent Slice (length = clean_latent_frames).
-        _ghost_pre_extend = int(guide_data.get("ghost_pre_extend", 0)) if guide_data else 0
-        if _ghost_pre_extend > 0:
-            gb, gc, gf, gh, gw = latent_image.shape
-            latent_image = torch.cat(
-                [latent_image, torch.zeros((gb, gc, _ghost_pre_extend, gh, gw), dtype=latent_image.dtype, device=latent_image.device)],
-                dim=2,
-            )
-            mb, mc, mf, mh, mw = noise_mask.shape
-            noise_mask = torch.cat(
-                [noise_mask, torch.ones((mb, mc, _ghost_pre_extend, mh, mw), dtype=noise_mask.dtype, device=noise_mask.device)],
-                dim=2,
-            )
-
         _, _, latent_length, latent_height, latent_width = latent_image.shape
         initial_latent_length = int(latent_length)
 
@@ -364,13 +818,7 @@ class LTXDirectorGuide:
         if msr is not None:
             return cls._inject_msr(positive, negative, vae, latent_image, noise_mask, msr, model, latent_downscale_factor, msr_strength)
 
-        # Parse timeline JSON to see if retake mode is active in UI
-        import json
-        timeline_data_str = guide_data.get("timeline_data", "{}") if guide_data else "{}"
-        try:
-            tdata = json.loads(timeline_data_str)
-        except Exception:
-            tdata = {}
+        tdata = _safe_json_loads(guide_data.get("timeline_data", "{}") if guide_data else "{}")
         
         is_retake_active = bool(retake_mode) or tdata.get("retakeMode", False)
         is_empty_latent = (latent_image.abs().max().item() < 1e-5)
@@ -392,111 +840,30 @@ class LTXDirectorGuide:
         # Load single base video, encode continuously, apply temporal mask.
         # -----------------------------------------------------------------------
         if is_retake_active:
-            print(f"[LTXDirectorGuide] Retake Mode active. Preserving base latent, masking selected regions. is_empty_latent: {is_empty_latent}")
-            target_width = latent_width * 32
-            target_height = latent_height * 32
-
-            # Calculate retake region latent indices first so we know what to copy/paste
-            retake_start = int(tdata.get("retakeStart", 0))
-            retake_len = int(tdata.get("retakeLength", 0))
-            retake_strength = float(tdata.get("retakeStrength", 1.0))
-
-            start_frame = int(guide_data.get("start_frame", 0))
-            relative_start = max(0, retake_start - start_frame)
-            
-            l_start = relative_start // time_scale_factor
-            l_end = int(math.ceil((relative_start + retake_len) / time_scale_factor))
-            
-            l_start = min(l_start, latent_length)
-            l_end = min(l_end, latent_length)
-
-            # Stage 2 optimization: If the retake region covers the entire generation area,
-            # there are no preserved regions to copy over in Stage 2. We can bypass video loading/encoding.
-            need_base_video = True
-            if not is_empty_latent and l_start == 0 and l_end >= latent_length:
-                need_base_video = False
-                print("[LTXDirectorGuide] Stage 2: Retake region covers the entire generated range. Skipping base video loading and VAE encoding.")
-
-            # 1. Try to load and encode base video from timeline data
-            retake_vid_info = tdata.get("retakeVideo") or {}
-            video_file = retake_vid_info.get("imageFile", "") if isinstance(retake_vid_info, dict) else ""
-            
-            # Fallback to first segment only if it exists and we have no retake video info at all (old workflows)
-            if not video_file and not retake_vid_info and len(segments) > 0:
-                video_file = segments[0].get("videoFile", "")
-
-            if need_base_video:
-                if not video_file:
-                    if retake_vid_info and not retake_vid_info.get("imageFile"):
-                        raise ValueError(
-                            "Retake Mode is active, but the base video file upload is still in progress (or failed). "
-                            "Please wait for the 'Uploading base video...' overlay on the timeline to disappear before queuing the prompt."
-                        )
-                    else:
-                        raise ValueError(
-                            "Retake Mode is active, but no base video has been selected on the timeline. "
-                            "Please drag and drop or upload a base video on the timeline first."
-                        )
-
-            if video_file and need_base_video:
-                try:
-                    print(f"[LTXDirectorGuide] Loading and encoding base video file: {video_file} starting at frame {start_frame} for length {ltxv_length} at resolution {target_width}x{target_height}")
-                    video_frames = _load_motion_video_frames(
-                        video_file, trim_start_frames=start_frame, length_frames=ltxv_length, director_fps=director_fps, resample_mode="nearest"
-                    )
-
-                    # Retake base video must match the exact target latent shape.
-                    # "maintain aspect ratio" doesn't pad, which causes shape mismatch in VAE encode.
-                    # So we fallback "maintain aspect ratio" to "pad" for retake base video.
-                    retake_resize_method = active_resize_method
-                    if retake_resize_method == "maintain aspect ratio":
-                        retake_resize_method = "pad"
-
-                    pixels = _resize_image(video_frames, target_width, target_height, retake_resize_method, divisible_by=1)
-
-                    num_clip_frames = pixels.shape[0]
-                    num_frames_to_keep = ((num_clip_frames - 1) // time_scale_factor) * time_scale_factor + 1
-                    encode_src = pixels[:num_frames_to_keep, :, :, :3]
-
-                    if use_tiled_encode:
-                        base_latent = vae.encode_tiled(encode_src, tile_x=tile_size, tile_y=tile_size, overlap=tile_overlap)
-                    else:
-                        base_latent = vae.encode(encode_src)
-                    
-                    base_latent = base_latent.to(device=latent_image.device, dtype=latent_image.dtype)
-                    
-                    # Copy to latent_image
-                    paste_len = min(base_latent.shape[2], latent_length)
-                    if is_empty_latent:
-                        # Stage 1: Overwrite entire latent with base video VAE encode
-                        latent_image[:, :, :paste_len] = base_latent[:, :, :paste_len]
-                    else:
-                        # Stage 2: Overwrite only preserved regions (before l_start and after l_end)
-                        # leaving the generated retake region from Stage 1 untouched
-                        print(f"[LTXDirectorGuide] Stage 2: Copying high-resolution base latent for preserved regions (0-{l_start} and {l_end}-{paste_len})")
-                        if l_start > 0:
-                            latent_image[:, :, :l_start] = base_latent[:, :, :l_start]
-                        if l_end < paste_len:
-                            latent_image[:, :, l_end:paste_len] = base_latent[:, :, l_end:paste_len]
-                except Exception as e:
-                    print(f"[LTXDirectorGuide] Failed to load/encode base video: {e}. Falling back to input latent.")
-
-            # 2. Build the temporal noise mask (0.0 = frozen, 1.0 = regenerate)
-            noise_mask = torch.zeros_like(noise_mask) # Initialize fully frozen
-            
-            l_start = min(l_start, latent_length)
-            l_end = min(l_end, latent_length)
-
-            if l_end > l_start:
-                noise_mask[:, :, l_start:l_end] = retake_strength
-
-            print(f"[LTXDirectorGuide] noise_mask slice: {noise_mask[0, 0, :, 0, 0].tolist()}")
-
-            # In retake mode, skip normal mode processing entirely and return immediately!
-            exact_crop_frames = max(0, int(latent_image.shape[2]) - initial_latent_length)
-            positive = node_helpers.conditioning_set_values(positive, {"nghtdrp_guide_crop_latent_frames": exact_crop_frames})
-            negative = node_helpers.conditioning_set_values(negative, {"nghtdrp_guide_crop_latent_frames": exact_crop_frames})
-            return (positive, negative, {"samples": latent_image, "noise_mask": noise_mask}, model, float(latent_downscale_factor))
+            return cls._execute_retake_mode(
+                positive,
+                negative,
+                vae,
+                latent_image,
+                noise_mask,
+                guide_data,
+                tdata,
+                segments,
+                model,
+                latent_downscale_factor,
+                director_fps,
+                latent_length,
+                latent_width,
+                latent_height,
+                time_scale_factor,
+                active_resize_method,
+                use_tiled_encode,
+                tile_size,
+                tile_overlap,
+                is_empty_latent,
+                initial_latent_length,
+                video_cache,
+            )
 
         # -----------------------------------------------------------------------
         # Standard Timeline Keyframe Guidance:
@@ -507,121 +874,51 @@ class LTXDirectorGuide:
         if len(images) > 0 or len(segments) > 0:
             print(f"[LTXDirectorGuide] Using Appended Keyframe Guidance. is_lora_active: {is_lora_active}")
 
-            # A. Process Image Guides
-            for idx, img_tensor in enumerate(images):
-                f_idx = insert_frames[idx] if idx < len(insert_frames) else 0
-                strength = float(strengths[idx] if idx < len(strengths) else 1.0)
-                if strength <= 0.0:
-                    continue
-
-                B_img, H_img, W_img, C_img = img_tensor.shape
-                target_pix_w = int(latent_width * 32)
-                target_pix_h = int(latent_height * 32)
-                if target_pix_w != W_img or target_pix_h != H_img:
-                    img_nchw = img_tensor.permute(0, 3, 1, 2)
-                    img_resized = comfy.utils.common_upscale(img_nchw, target_pix_w, target_pix_h, upscale_method, "disabled")
-                    img_tensor = img_resized.permute(0, 2, 3, 1)
-
-                image_pixels, guide_latent = nodes_lt.LTXVAddGuide.encode(vae, latent_width, latent_height, img_tensor, scale_factors)
-                frame_idx, latent_idx = nodes_lt.LTXVAddGuide.get_latent_index(positive, latent_length, len(image_pixels), int(f_idx), scale_factors)
-
-                if latent_idx >= latent_length:
-                    continue
-
-                max_frames = latent_length - latent_idx
-                if guide_latent.shape[2] > max_frames:
-                    guide_latent = guide_latent[:, :, :max_frames]
-
-                tokens_added = guide_latent.shape[2] * guide_latent.shape[3] * guide_latent.shape[4]
-                guide_orig_shape = list(guide_latent.shape[2:])
-
-                positive, negative, latent_image, noise_mask = nodes_lt.LTXVAddGuide.append_keyframe(
-                    positive, negative, frame_idx, latent_image, noise_mask, guide_latent, strength, scale_factors
-                )
-                if is_lora_active:
-                    positive = _append_guide_attention_entry(positive, tokens_added, guide_orig_shape, attention_strength=image_attention_strength)
-                    negative = _append_guide_attention_entry(negative, tokens_added, guide_orig_shape, attention_strength=image_attention_strength)
-
-            # B. Process Motion Video Segments
-            for seg in segments:
-                try:
-                    video_file = seg.get("videoFile")
-                    if not video_file:
-                        continue
-
-                    start_frame = int(seg.get("start", 0))
-                    length_frames = int(seg.get("length", 1))
-                    trim_start = int(seg.get("trimStart", 0))
-                    video_strength = float(seg.get("videoStrength", 1.0))
-                    video_attention_strength = float(seg.get("videoAttentionStrength", 0.65))
-
-                    if length_frames <= 0 or video_strength <= 0.0:
-                        continue
-
-                    start_frame_aligned = start_frame
-                    video_frames = _load_motion_video_frames(video_file, trim_start, length_frames, director_fps, seg.get("resampleMode", "nearest"))
-
-                    num_frames_to_keep = ((video_frames.shape[0] - 1) // time_scale_factor) * time_scale_factor + 1
-                    video_frames = video_frames[:num_frames_to_keep]
-                    causal_fix = int(start_frame_aligned) == 0 or num_frames_to_keep == 1
-                    encode_frames = video_frames if causal_fix else torch.cat([video_frames[:1], video_frames], dim=0)
-
-                    _, guide_latent = _encode_video_iclora_guide(vae, latent_width, latent_height, encode_frames, scale_factors, latent_downscale_factor, crop, use_tiled_encode, tile_size, tile_overlap, resize_method=active_resize_method)
-
-                    if not causal_fix:
-                        guide_latent = guide_latent[:, :, 1:, :, :]
-
-                    frame_idx = start_frame_aligned
-                    latent_idx = (frame_idx + time_scale_factor - 1) // time_scale_factor if frame_idx > 0 else 0
-
-                    if latent_idx >= latent_length:
-                        continue
-
-                    if start_frame > 0 and guide_latent.shape[2] > 1:
-                        guide_latent = guide_latent[:, :, 1:, :, :]
-                        frame_idx += time_scale_factor
-                        latent_idx += 1
-                        if latent_idx >= latent_length:
-                            continue
-
-                    max_frames = latent_length - latent_idx
-                    if guide_latent.shape[2] > max_frames:
-                        guide_latent = guide_latent[:, :, :max_frames]
-
-                    guide_orig_shape = list(guide_latent.shape[2:])
-
-                    B_g, C_g, F_g, H_g, W_g = guide_latent.shape
-                    guide_mask = torch.ones((B_g, 1, F_g, H_g, W_g), device=guide_latent.device, dtype=guide_latent.dtype)
-
-                    if start_frame > 0:
-                        ramp_steps = [0.25, 0.65]
-                        for i, s in enumerate(ramp_steps):
-                            if i < F_g:
-                                guide_mask_val = 1.0 + video_strength * (1.0 - s)
-                                guide_mask[:, :, i, :, :] = guide_mask_val
-
-                    ldf = int(max(1, round(float(latent_downscale_factor))))
-                    if ldf > 1:
-                        dilated = _dilate_latent({"samples": guide_latent, "noise_mask": guide_mask}, horizontal_scale=ldf, vertical_scale=ldf)
-                        guide_mask = dilated["noise_mask"]
-                        guide_latent = dilated["samples"]
-
-                    tokens_added = guide_latent.shape[2] * guide_latent.shape[3] * guide_latent.shape[4]
-                    positive, negative, latent_image, noise_mask = nodes_lt.LTXVAddGuide.append_keyframe(
-                        positive, negative, frame_idx, latent_image, noise_mask, guide_latent, video_strength, scale_factors, guide_mask=guide_mask, latent_downscale_factor=float(latent_downscale_factor), causal_fix=causal_fix
-                    )
-                    if is_lora_active:
-                        positive = _append_guide_attention_entry(positive, tokens_added, guide_orig_shape, attention_strength=video_attention_strength)
-                        negative = _append_guide_attention_entry(negative, tokens_added, guide_orig_shape, attention_strength=video_attention_strength)
-                except Exception as e:
-                    raise RuntimeError(f"LTX Director Guide motion segment failed for {seg}: {e}") from e
+            positive, negative, latent_image, noise_mask = cls._apply_image_guides(
+                positive,
+                negative,
+                vae,
+                latent_image,
+                noise_mask,
+                images,
+                insert_frames,
+                strengths,
+                latent_width,
+                latent_height,
+                latent_length,
+                scale_factors,
+                image_attention_strength,
+                is_lora_active,
+                upscale_method,
+            )
+            positive, negative, latent_image, noise_mask = cls._apply_motion_segment_guides(
+                positive,
+                negative,
+                vae,
+                latent_image,
+                noise_mask,
+                segments,
+                director_fps,
+                latent_length,
+                latent_width,
+                latent_height,
+                time_scale_factor,
+                scale_factors,
+                latent_downscale_factor,
+                crop,
+                use_tiled_encode,
+                tile_size,
+                tile_overlap,
+                active_resize_method,
+                is_lora_active,
+                video_cache,
+            )
 
         else:
             print("[LTXDirectorGuide] No timeline guides present. Passing through.")
 
         exact_crop_frames = max(0, int(latent_image.shape[2]) - initial_latent_length)
-        positive = node_helpers.conditioning_set_values(positive, {"nghtdrp_guide_crop_latent_frames": exact_crop_frames})
-        negative = node_helpers.conditioning_set_values(negative, {"nghtdrp_guide_crop_latent_frames": exact_crop_frames})
+        positive, negative = cls._apply_crop_count(positive, negative, exact_crop_frames)
 
         return (positive, negative, {"samples": latent_image, "noise_mask": noise_mask}, model, float(latent_downscale_factor))
 
@@ -655,16 +952,7 @@ class LTXDirectorGuide:
         # (which trims the slideshow's temporal footprint) lands on the true clean length.
         pad_latents = max(0, prefix_latents - len(keyframes))
         if pad_latents > 0:
-            B, C, F, H, W = latent_image.shape
-            latent_image = torch.cat(
-                [latent_image, torch.zeros((B, C, pad_latents, H, W), dtype=latent_image.dtype, device=latent_image.device)],
-                dim=2,
-            )
-            mb, mc, mf, mh, mw = noise_mask.shape
-            noise_mask = torch.cat(
-                [noise_mask, torch.ones((mb, mc, pad_latents, mh, mw), dtype=noise_mask.dtype, device=noise_mask.device)],
-                dim=2,
-            )
+            latent_image, noise_mask = _append_latent_frames(latent_image, noise_mask, pad_latents, mask_fill=1.0)
 
         latent = {"samples": latent_image, "noise_mask": noise_mask}
 

msr_character.py

@@ -0,0 +1,126 @@
+from __future__ import annotations
+
+from comfy_api.latest import io
+
+
+MSRCharacterData = io.Custom("MSR_CHARACTER")
+MSRCharacterSetData = io.Custom("MSR_CHARACTER_SET")
+
+
+def _compact_alias(alias: str) -> str:
+    if not alias:
+        return ""
+    alias = alias.strip()
+    if alias.startswith("@"):
+        alias = alias[1:]
+    return alias.strip()
+
+
+class MSRCharacter(io.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return io.Schema(
+            node_id="MSRCharacterCS",
+            display_name="MSR Character DUMAS",
+            category="WhatDreamsCost DUMAS",
+            description=(
+                "Reusable MSR character payload. Accepts up to two reference images plus "
+                "a description and optional alias for @alias prompt replacement."
+            ),
+            inputs=[
+                io.Image.Input(
+                    "image_1",
+                    optional=True,
+                    tooltip="Optional first reference image for this character.",
+                ),
+                io.Image.Input(
+                    "image_2",
+                    optional=True,
+                    tooltip="Optional second reference image for this character.",
+                ),
+                io.String.Input(
+                    "description",
+                    multiline=True,
+                    default="",
+                    optional=True,
+                    tooltip="Manual or generated appearance description used for prompt replacement.",
+                ),
+                io.String.Input(
+                    "alias",
+                    default="",
+                    optional=True,
+                    tooltip="Optional alias. Prompts can use @alias alongside @charN/@characterN.",
+                ),
+            ],
+            outputs=[
+                MSRCharacterData.Output(display_name="character"),
+            ],
+        )
+
+    @classmethod
+    def execute(cls, image_1=None, image_2=None, description="", alias="") -> io.NodeOutput:
+        images = []
+        if image_1 is not None:
+            images.append(image_1)
+        if image_2 is not None:
+            images.append(image_2)
+
+        payload = {
+            "images": images,
+            "description": (description or "").strip(),
+            "alias": _compact_alias(alias or ""),
+        }
+        return io.NodeOutput(payload)
+
+
+class MSRCharacterSet(io.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return io.Schema(
+            node_id="MSRCharacterSetCS",
+            display_name="MSR Character Set DUMAS",
+            category="WhatDreamsCost DUMAS",
+            description="Bundles up to six MSR Character nodes into one character set for the director.",
+            inputs=[
+                MSRCharacterData.Input("character_1", optional=True),
+                MSRCharacterData.Input("character_2", optional=True),
+                MSRCharacterData.Input("character_3", optional=True),
+                MSRCharacterData.Input("character_4", optional=True),
+                MSRCharacterData.Input("character_5", optional=True),
+                MSRCharacterData.Input("character_6", optional=True),
+            ],
+            outputs=[
+                MSRCharacterSetData.Output(display_name="character_set"),
+            ],
+        )
+
+    @classmethod
+    def execute(
+        cls,
+        character_1=None,
+        character_2=None,
+        character_3=None,
+        character_4=None,
+        character_5=None,
+        character_6=None,
+    ) -> io.NodeOutput:
+        characters = []
+        for item in (
+            character_1,
+            character_2,
+            character_3,
+            character_4,
+            character_5,
+            character_6,
+        ):
+            if not item:
+                continue
+            characters.append(
+                {
+                    "images": list(item.get("images") or []),
+                    "description": (item.get("description") or "").strip(),
+                    "alias": _compact_alias(item.get("alias") or ""),
+                }
+            )
+
+        return io.NodeOutput({"characters": characters})

patches.py

@@ -1,9 +1,12 @@
 import logging
+import os
 import types
 
 import comfy.ldm.modules.attention
 
 log = logging.getLogger(__name__)
+_DEBUG_PROMPT_RELAY = os.environ.get("PROMPT_RELAY_DEBUG", "").strip().lower() in {"1", "true", "yes", "on"}
+_DEBUG_LOG_PATH = os.path.join(os.path.dirname(__file__), "debug_prompt_relay.log")
 
 
 def _masked_attention(q, k, v, heads, mask, transformer_options={}, **kwargs):
@@ -73,10 +76,10 @@ def _make_masked_override(prev_override):
 
 
 def debug_log(msg):
+    if not _DEBUG_PROMPT_RELAY:
+        return
     try:
-        import os
-        log_path = os.path.join(os.path.dirname(__file__), "debug_prompt_relay.log")
-        with open(log_path, "a", encoding="utf-8") as f:
+        with open(_DEBUG_LOG_PATH, "a", encoding="utf-8") as f:
             f.write(msg + "\n")
     except Exception:
         pass

prompt_relay.py

@@ -1,21 +1,28 @@
 import logging
 import math
+import os
 import torch
 
 log = logging.getLogger(__name__)
+_DEBUG_PROMPT_RELAY = os.environ.get("PROMPT_RELAY_DEBUG", "").strip().lower() in {"1", "true", "yes", "on"}
+_DEBUG_LOG_PATH = os.path.join(os.path.dirname(__file__), "debug_prompt_relay.log")
+
+
+def _build_segment_cost(query_frames, seg, dtype):
+    local = seg["local_token_idx"]
+    distance = (query_frames[:, None] - seg["midpoint"]).abs()
+    cost = seg["strength"] * (torch.relu(distance - seg["window"]) ** 2) / (2 * seg["sigma"] ** 2)
+    return local, cost.to(dtype)
 
 
 def build_temporal_cost(q_token_idx, Lq, Lk, device, dtype, tokens_per_frame):
     """Gaussian penalty matrix [Lq, Lk] for video cross-attention (integer frame indexing)."""
     offset = torch.zeros(Lq, Lk, device=device, dtype=dtype)
-    query_frames = torch.arange(Lq, device=device, dtype=torch.long) // tokens_per_frame
+    query_frames = (torch.arange(Lq, device=device, dtype=torch.float32) / float(tokens_per_frame)).floor()
 
     for seg in q_token_idx:
-        local = seg["local_token_idx"].to(device=device)
-        d = (query_frames.float()[:, None] - seg["midpoint"]).abs()
-        strength = seg.get("strength", 1.0)
-        cost = strength * (torch.relu(d - seg["window"]) ** 2) / (2 * seg["sigma"] ** 2)
-        offset[:, local] = cost.to(offset.dtype)
+        local, cost = _build_segment_cost(query_frames, seg, offset.dtype)
+        offset[:, local.to(device=device)] = cost
 
     return offset
 
@@ -26,27 +33,27 @@ def build_temporal_cost_scaled(q_token_idx, Lq, Lk, device, dtype, latent_frames
     query_frames = torch.arange(Lq, device=device, dtype=torch.float32) * latent_frames / Lq
 
     for seg in q_token_idx:
-        local = seg["local_token_idx"].to(device=device)
-        d = (query_frames[:, None] - seg["midpoint"]).abs()
         if is_audio:
-            sigma_val = seg.get("sigma_audio", seg["sigma"])
-            window_val = seg.get("window_audio", seg["window"])
-            strength_val = seg.get("strength_audio", 1.0)
+            active_seg = {
+                "local_token_idx": seg["local_token_idx"],
+                "midpoint": seg["midpoint"],
+                "window": seg.get("window_audio", seg["window"]),
+                "sigma": seg.get("sigma_audio", seg["sigma"]),
+                "strength": seg.get("strength_audio", 1.0),
+            }
         else:
-            sigma_val = seg["sigma"]
-            window_val = seg["window"]
-            strength_val = seg.get("strength", 1.0)
-        cost = strength_val * (torch.relu(d - window_val) ** 2) / (2 * sigma_val ** 2)
-        offset[:, local] = cost.to(offset.dtype)
+            active_seg = seg
+        local, cost = _build_segment_cost(query_frames, active_seg, offset.dtype)
+        offset[:, local.to(device=device)] = cost
 
     return offset
 
 
 def debug_log(msg):
+    if not _DEBUG_PROMPT_RELAY:
+        return
     try:
-        import os
-        log_path = os.path.join(os.path.dirname(__file__), "debug_prompt_relay.log")
-        with open(log_path, "a", encoding="utf-8") as f:
+        with open(_DEBUG_LOG_PATH, "a", encoding="utf-8") as f:
             f.write(msg + "\n")
     except Exception:
         pass
@@ -61,6 +68,15 @@ def create_mask_fn(q_token_idx, fallback_tokens_per_frame, latent_frames):
     """
     cache = {}
     max_token_idx = max(int(seg["local_token_idx"].max().item()) for seg in q_token_idx) + 1
+    latent_frames = max(1, int(latent_frames))
+    fallback_tokens_per_frame = max(1, int(fallback_tokens_per_frame))
+
+    def _get_video_query_layout(Lq, grid_sizes):
+        if grid_sizes is not None:
+            return int(grid_sizes[1]) * int(grid_sizes[2])
+        if Lq % latent_frames == 0:
+            return Lq // latent_frames
+        return fallback_tokens_per_frame
 
     def mask_fn(Lq, Lk, dtype, device, transformer_options):
         debug_log(f"mask_fn check: Lq={Lq} Lk={Lk} max_token_idx={max_token_idx} cond_or_uncond={transformer_options.get('cond_or_uncond', [])}")
@@ -82,13 +98,7 @@ def create_mask_fn(q_token_idx, fallback_tokens_per_frame, latent_frames):
             mode = "scaled"
             video_lq = -1
         else:
-            if grid_sizes is not None:
-                video_tpf = int(grid_sizes[1]) * int(grid_sizes[2])
-            else:
-                if Lq % latent_frames == 0:
-                    video_tpf = Lq // latent_frames
-                else:
-                    video_tpf = fallback_tokens_per_frame
+            video_tpf = _get_video_query_layout(Lq, grid_sizes)
             video_lq = latent_frames * video_tpf
 
             # Skip cross-modal attention — text keys are padded to a fixed length ≥ max_token_idx and != video_lq
@@ -189,6 +199,34 @@ def get_raw_tokenizer(clip):
     )
 
 
+def _get_tokenizer_input_ids(raw_tokenizer, text):
+    tokenized = raw_tokenizer(text)
+    if isinstance(tokenized, dict) and "input_ids" in tokenized:
+        return tokenized["input_ids"]
+    if hasattr(tokenized, "input_ids"):
+        return tokenized.input_ids
+    if isinstance(tokenized, list):
+        return tokenized
+    return []
+
+
+def _tokenized_length(raw_tokenizer, text, eos_adjustment):
+    return max(0, len(_get_tokenizer_input_ids(raw_tokenizer, text)) - eos_adjustment)
+
+
+def _tokenizer_has_eos(raw_tokenizer):
+    if getattr(raw_tokenizer, "add_eos", False):
+        return True
+    try:
+        ids = _get_tokenizer_input_ids(raw_tokenizer, "test")
+        if not ids:
+            return False
+        eos_id = getattr(raw_tokenizer, "eos_token_id", None)
+        return (eos_id is not None and ids[-1] == eos_id) or ids[-1] == 1
+    except Exception:
+        return False
+
+
 def map_token_indices(raw_tokenizer, global_prompt, local_prompts):
     """Tokenize global + space-prefixed locals; return (full_prompt, per-local token ranges).
 
@@ -197,38 +235,14 @@ def map_token_indices(raw_tokenizer, global_prompt, local_prompts):
     prefixed_locals = [" " + lp for lp in local_prompts]
     full_prompt = global_prompt + "".join(prefixed_locals)
 
-    # Detect if the tokenizer appends EOS dynamically
-    has_eos = getattr(raw_tokenizer, "add_eos", False)
-    if not has_eos:
-        try:
-            test_res = raw_tokenizer("test")
-            if isinstance(test_res, dict) and "input_ids" in test_res:
-                ids = test_res["input_ids"]
-            elif hasattr(test_res, "input_ids"):
-                ids = test_res.input_ids
-            elif isinstance(test_res, list):
-                ids = test_res
-            else:
-                ids = []
-            
-            if ids:
-                eos_id = getattr(raw_tokenizer, "eos_token_id", None)
-                if eos_id is not None and ids[-1] == eos_id:
-                    has_eos = True
-                elif ids[-1] == 1:
-                    has_eos = True
-        except Exception:
-            pass
-
-    eos_adj = 1 if has_eos else 0
-
-    prev_len = len(raw_tokenizer(global_prompt)["input_ids"]) - eos_adj
+    eos_adj = 1 if _tokenizer_has_eos(raw_tokenizer) else 0
+    prev_len = _tokenized_length(raw_tokenizer, global_prompt, eos_adj)
     token_ranges = []
     built = global_prompt
 
     for plp in prefixed_locals:
         built += plp
-        cur_len = len(raw_tokenizer(built)["input_ids"]) - eos_adj
+        cur_len = _tokenized_length(raw_tokenizer, built, eos_adj)
         if cur_len <= prev_len:
             raise ValueError(f"Local prompt produced no tokens: '{plp.strip()}'")
         token_ranges.append((prev_len, cur_len))

timeline_nodes.py

@@ -0,0 +1,117 @@
+from __future__ import annotations
+
+from comfy_api.latest import io
+
+
+TimelineSectionData = io.Custom("TIMELINE_SECTION")
+TimelineSectionSetData = io.Custom("TIMELINE_SECTION_SET")
+class TimelineSection(io.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return io.Schema(
+            node_id="TimelineSectionCS",
+            display_name="Timeline Section DUMAS",
+            category="WhatDreamsCost DUMAS",
+            description="Reusable timeline section with required text, optional image, and a duration in seconds.",
+            inputs=[
+                io.Image.Input(
+                    "image",
+                    optional=True,
+                    tooltip="Optional image for this timeline section. When omitted, the section becomes a text-only block.",
+                ),
+                io.String.Input(
+                    "text",
+                    multiline=True,
+                    default="",
+                    tooltip="Required text/prompt for this timeline section.",
+                ),
+                io.Float.Input(
+                    "duration",
+                    default=1.0,
+                    min=0.1,
+                    max=1000.0,
+                    step=0.01,
+                    tooltip="Section duration in seconds.",
+                ),
+            ],
+            outputs=[
+                TimelineSectionData.Output(display_name="section"),
+            ],
+        )
+
+    @classmethod
+    def execute(cls, image=None, text="", duration=1.0) -> io.NodeOutput:
+        payload = {
+            "image": image,
+            "text": (text or "").strip(),
+            "duration": max(0.1, float(duration or 0.0)),
+        }
+        return io.NodeOutput(payload)
+
+
+class TimelineSectionSet(io.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return io.Schema(
+            node_id="TimelineSectionSetCS",
+            display_name="Timeline Section Set DUMAS",
+            category="WhatDreamsCost DUMAS",
+            description="Bundles up to ten Timeline Section nodes into one reusable section set.",
+            inputs=[
+                TimelineSectionData.Input("section_1", optional=True),
+                TimelineSectionData.Input("section_2", optional=True),
+                TimelineSectionData.Input("section_3", optional=True),
+                TimelineSectionData.Input("section_4", optional=True),
+                TimelineSectionData.Input("section_5", optional=True),
+                TimelineSectionData.Input("section_6", optional=True),
+                TimelineSectionData.Input("section_7", optional=True),
+                TimelineSectionData.Input("section_8", optional=True),
+                TimelineSectionData.Input("section_9", optional=True),
+                TimelineSectionData.Input("section_10", optional=True),
+            ],
+            outputs=[
+                TimelineSectionSetData.Output(display_name="section_set"),
+            ],
+        )
+
+    @classmethod
+    def execute(
+        cls,
+        section_1=None,
+        section_2=None,
+        section_3=None,
+        section_4=None,
+        section_5=None,
+        section_6=None,
+        section_7=None,
+        section_8=None,
+        section_9=None,
+        section_10=None,
+    ) -> io.NodeOutput:
+        sections = []
+        for item in (
+            section_1,
+            section_2,
+            section_3,
+            section_4,
+            section_5,
+            section_6,
+            section_7,
+            section_8,
+            section_9,
+            section_10,
+        ):
+            if not item:
+                continue
+            text = (item.get("text") or "").strip()
+            if not text:
+                continue
+            sections.append(
+                {
+                    "image": item.get("image"),
+                    "text": text,
+                    "duration": max(0.1, float(item.get("duration") or 0.0)),
+                }
+            )
+
+        return io.NodeOutput({"sections": sections})