perf: reduce redundant media work in director

ltx_director.py

@@ -14,6 +14,7 @@ from PIL import Image
 import os
 import platform
 import subprocess
+from urllib.parse import parse_qs, urlparse
 import folder_paths
 import comfy.model_management
 from server import PromptServer
@@ -73,6 +74,140 @@ MSR_PREFIX_FRAMES = 41
 # trained; Licon MSR V1 uses full-resolution references (1.0). This is INDEPENDENT of any
 # multi-stage scale_by / x2 upscaling, which LTXDirectorGuide handles downstream.
 MSR_LATENT_DOWNSCALE = 1.0
+EMPTY_IMAGE_SHAPE = (1, 512, 512, 3)
+PEAK_BUCKETS = 200
+
+
+def _empty_image_tensor() -> torch.Tensor:
+    return torch.zeros(EMPTY_IMAGE_SHAPE, dtype=torch.float32)
+
+
+def _normalise_b64_payload(payload: str) -> str:
+    if payload and "," in payload:
+        return payload.split(",", 1)[1]
+    return payload or ""
+
+
+def _image_to_tensor(img: Image.Image) -> torch.Tensor:
+    arr = np.asarray(img.convert("RGB"), dtype=np.float32) / 255.0
+    return torch.from_numpy(arr).unsqueeze(0)
+
+
+def _resolve_input_path(path: str, input_dir: str = None) -> str | None:
+    if not path:
+        return None
+
+    input_dir = input_dir or folder_paths.get_input_directory()
+    clean_path = path.replace("\\", "/")
+    candidates = [
+        os.path.join(input_dir, clean_path),
+        os.path.join(input_dir, "whatdreamscost", os.path.basename(clean_path)),
+        os.path.join(input_dir, os.path.basename(clean_path)),
+    ]
+
+    for candidate in candidates:
+        if os.path.exists(candidate):
+            return candidate
+    return None
+
+
+def _load_image_reference(b64_or_url: str = "", filename: str = None, cache: dict | None = None,
+                          input_dir: str = None) -> torch.Tensor:
+    """Load a reference image from a base64 string, a Comfy /view URL, or an input filename."""
+    if not b64_or_url and not filename:
+        return _empty_image_tensor()
+
+    input_dir = input_dir or folder_paths.get_input_directory()
+    cache = cache if cache is not None else {}
+
+    file_path = None
+    if b64_or_url and "view?" in b64_or_url:
+        try:
+            parsed = urlparse(b64_or_url)
+            q = parse_qs(parsed.query)
+            fname = q.get("filename", [None])[0]
+            subfolder = q.get("subfolder", [""])[0]
+            if fname:
+                file_path = _resolve_input_path(os.path.join(subfolder, fname), input_dir)
+        except Exception as e:
+            log.debug(f"[PromptRelay] URL parsing failed for {b64_or_url}: {e}")
+
+    if file_path is None and filename:
+        file_path = _resolve_input_path(filename, input_dir)
+
+    if file_path:
+        cache_key = ("file", file_path)
+        cached = cache.get(cache_key)
+        if cached is not None:
+            return cached
+        try:
+            tensor = _image_to_tensor(Image.open(file_path))
+            cache[cache_key] = tensor
+            return tensor
+        except Exception as e:
+            log.debug(f"[PromptRelay] File image loading failed for {file_path}: {e}")
+
+    if b64_or_url:
+        try:
+            b64_str = _normalise_b64_payload(b64_or_url)
+            cache_key = ("b64", b64_str)
+            cached = cache.get(cache_key)
+            if cached is not None:
+                return cached
+            img_bytes = base64.b64decode(b64_str)
+            tensor = _image_to_tensor(Image.open(_io.BytesIO(img_bytes)))
+            cache[cache_key] = tensor
+            return tensor
+        except Exception as e:
+            log.debug(f"[PromptRelay] Base64 decoding failed: {e}")
+
+    return _empty_image_tensor()
+
+
+def _compute_signal_peaks(samples: np.ndarray, num_peaks: int = PEAK_BUCKETS) -> list[float]:
+    if samples.size == 0:
+        return [0.0] * num_peaks
+
+    step = max(1, int(math.ceil(samples.size / num_peaks)))
+    padded = np.pad(samples, (0, step * num_peaks - samples.size), mode="constant")
+    chunks = padded.reshape(num_peaks, step)
+    return (np.max(np.abs(chunks), axis=1) / 32767.0).astype(np.float32).tolist()
+
+
+def _read_audio_bytes(container, stream, resampler) -> bytes:
+    audio_bytes = bytearray()
+    for frame in container.decode(stream):
+        for resampled_frame in resampler.resample(frame):
+            audio_bytes.extend(resampled_frame.to_ndarray().tobytes())
+    for resampled_frame in resampler.resample(None):
+        audio_bytes.extend(resampled_frame.to_ndarray().tobytes())
+    return bytes(audio_bytes)
+
+
+def _target_resize_dims(src_w: int, src_h: int, custom_width: int, custom_height: int, divisible_by: int) -> tuple[int, int, str]:
+    def snap(val, div):
+        return max(div, (val // div) * div)
+
+    if custom_width > 0 and custom_height > 0:
+        return custom_width, custom_height, None
+    if custom_width > 0:
+        tgt_w = snap(custom_width, divisible_by)
+        tgt_h = snap(int(src_h * tgt_w / src_w), divisible_by)
+        return tgt_w, tgt_h, "stretch to fit"
+    if custom_height > 0:
+        tgt_h = snap(custom_height, divisible_by)
+        tgt_w = snap(int(src_w * tgt_h / src_h), divisible_by)
+        return tgt_w, tgt_h, "stretch to fit"
+    return src_w, src_h, "maintain aspect ratio"
+
+
+def _extract_video_dimensions(file_path: str) -> tuple[int | None, int | None]:
+    try:
+        with av.open(file_path) as container:
+            stream = container.streams.video[0]
+            return stream.width or stream.codec_context.width, stream.height or stream.codec_context.height
+    except Exception:
+        return None, None
 
 
 def _preprocess_prompts_with_characters(global_prompt, local_prompts, char1="", char2="", char3=""):
@@ -111,47 +246,9 @@ def _preprocess_prompts_with_characters(global_prompt, local_prompts, char1="",
     return gp, " | ".join(processed_locals)
 
 
-def _load_image_source(b64_or_url: str, filename: str = None) -> torch.Tensor:
-    """Load a reference image from a base64 string, a Comfy /view URL, or an input filename."""
-    if not b64_or_url and not filename:
-        return torch.zeros((1, 512, 512, 3), dtype=torch.float32)
-
-    if b64_or_url and "view?" in b64_or_url:
-        try:
-            from urllib.parse import urlparse, parse_qs
-            parsed = urlparse(b64_or_url)
-            q = parse_qs(parsed.query)
-            fname = q.get("filename", [None])[0]
-            subfolder = q.get("subfolder", [""])[0]
-            if fname:
-                file_path = os.path.join(folder_paths.get_input_directory(), subfolder, fname)
-                if os.path.exists(file_path):
-                    img = Image.open(file_path).convert("RGB")
-                    arr = np.array(img, dtype=np.float32) / 255.0
-                    return torch.from_numpy(arr).unsqueeze(0)
-        except Exception as e:
-            log.debug(f"[PromptRelay] URL parsing failed for {b64_or_url}: {e}")
-
-    if filename:
-        file_path = os.path.join(folder_paths.get_input_directory(), filename)
-        if os.path.exists(file_path):
-            img = Image.open(file_path).convert("RGB")
-            arr = np.array(img, dtype=np.float32) / 255.0
-            return torch.from_numpy(arr).unsqueeze(0)
-
-    if b64_or_url:
-        try:
-            b64_str = b64_or_url
-            if "," in b64_str:
-                b64_str = b64_str.split(",", 1)[1]
-            img_bytes = base64.b64decode(b64_str)
-            img = Image.open(_io.BytesIO(img_bytes)).convert("RGB")
-            arr = np.array(img, dtype=np.float32) / 255.0
-            return torch.from_numpy(arr).unsqueeze(0)
-        except Exception as e:
-            log.debug(f"[PromptRelay] Base64 decoding failed: {e}")
-
-    return torch.zeros((1, 512, 512, 3), dtype=torch.float32)
+def _load_image_source(b64_or_url: str, filename: str = None, cache: dict | None = None,
+                       input_dir: str = None) -> torch.Tensor:
+    return _load_image_reference(b64_or_url=b64_or_url, filename=filename, cache=cache, input_dir=input_dir)
 
 
 def _execute_comfy_node(node_class, **kwargs):
@@ -394,24 +491,13 @@ async def unload_ollama_endpoint(request):
 
 def read_wav_peaks(wav_path):
     import wave
-    peaks = []
     with wave.open(wav_path, 'rb') as w:
         n_frames = w.getnframes()
-        if n_frames > 0:
+        if n_frames <= 0:
+            return [0.0] * PEAK_BUCKETS
         frames_bytes = w.readframes(n_frames)
         samples = np.frombuffer(frames_bytes, dtype=np.int16)
-            num_peaks = 200
-            step = max(1, len(samples) // num_peaks)
-            for i in range(num_peaks):
-                chunk = samples[i * step : (i + 1) * step]
-                if len(chunk) > 0:
-                    max_val = np.max(np.abs(chunk)) / 32767.0
-                    peaks.append(float(max_val))
-                else:
-                    peaks.append(0.0)
-        else:
-            peaks = [0.0] * 200
-    return peaks
+    return _compute_signal_peaks(samples)
 
 
 def extract_audio_from_video(video_path):
@@ -445,18 +531,7 @@ def extract_audio_from_video(video_path):
                 rate=44100,
             )
             
-            audio_bytes = bytearray()
-            
-            for frame in container.decode(stream):
-                for resampled_frame in resampler.resample(frame):
-                    arr = resampled_frame.to_ndarray()
-                    audio_bytes.extend(arr.tobytes())
-                    
-            # Flush resampler
-            for resampled_frame in resampler.resample(None):
-                arr = resampled_frame.to_ndarray()
-                audio_bytes.extend(arr.tobytes())
-                
+            audio_bytes = _read_audio_bytes(container, stream, resampler)
             if not audio_bytes:
                 return None, None
                 
@@ -468,17 +543,8 @@ def extract_audio_from_video(video_path):
                 w.writeframes(audio_bytes)
                 
         # Calculate peaks
-        peaks = []
         samples = np.frombuffer(audio_bytes, dtype=np.int16)
-        num_peaks = 200
-        step = max(1, len(samples) // num_peaks)
-        for i in range(num_peaks):
-            chunk = samples[i * step : (i + 1) * step]
-            if len(chunk) > 0:
-                max_val = np.max(np.abs(chunk)) / 32767.0
-                peaks.append(float(max_val))
-            else:
-                peaks.append(0.0)
+        peaks = _compute_signal_peaks(samples)
                 
         input_dir = folder_paths.get_input_directory()
         rel_output = os.path.relpath(output_wav, input_dir).replace('\\', '/')
@@ -509,30 +575,11 @@ def get_audio_peaks(audio_path):
                 layout='mono',
                 rate=8000,
             )
-            audio_bytes = bytearray()
-            for frame in container.decode(stream):
-                for resampled_frame in resampler.resample(frame):
-                    arr = resampled_frame.to_ndarray()
-                    audio_bytes.extend(arr.tobytes())
-            for resampled_frame in resampler.resample(None):
-                arr = resampled_frame.to_ndarray()
-                audio_bytes.extend(arr.tobytes())
-                
+            audio_bytes = _read_audio_bytes(container, stream, resampler)
             if not audio_bytes:
                 return None
-                
-            peaks = []
             samples = np.frombuffer(audio_bytes, dtype=np.int16)
-            num_peaks = 200
-            step = max(1, len(samples) // num_peaks)
-            for i in range(num_peaks):
-                chunk = samples[i * step : (i + 1) * step]
-                if len(chunk) > 0:
-                    max_val = np.max(np.abs(chunk)) / 32767.0
-                    peaks.append(float(max_val))
-                else:
-                    peaks.append(0.0)
-            return peaks
+            return _compute_signal_peaks(samples)
     except Exception as e:
         print(f"[LTXDirector] Failed to get audio peaks via PyAV: {e}")
         return None
@@ -657,38 +704,24 @@ async def ltx_director_upload_chunk(request):
 
 
 
-def _load_image_tensor(seg: dict) -> torch.Tensor:
+def _load_image_tensor(seg: dict, cache: dict | None = None, input_dir: str = None) -> torch.Tensor:
     """Decode an image from the ComfyUI input folder (if imageFile provided) or fallback to base64
     to a ComfyUI-style image tensor of shape [1, H, W, 3], float32 in [0, 1]."""
-    if seg.get("imageFile"):
-        file_path = os.path.join(folder_paths.get_input_directory(), seg["imageFile"])
-        if os.path.exists(file_path):
-            img = Image.open(file_path).convert("RGB")
-            arr = np.array(img, dtype=np.float32) / 255.0
-            return torch.from_numpy(arr).unsqueeze(0)
+    return _load_image_reference(
+        b64_or_url=seg.get("imageB64", ""),
+        filename=seg.get("imageFile"),
+        cache=cache,
+        input_dir=input_dir,
+    )
 
-    b64_str = seg.get("imageB64", "")
-    if not b64_str or b64_str.startswith("/view?"):
-        return torch.zeros((1, 512, 512, 3), dtype=torch.float32)
 
-    if "," in b64_str:
-        b64_str = b64_str.split(",", 1)[1]
-    
-    try:
-        img_bytes = base64.b64decode(b64_str)
-        img = Image.open(_io.BytesIO(img_bytes)).convert("RGB")
-        arr = np.array(img, dtype=np.float32) / 255.0
-        return torch.from_numpy(arr).unsqueeze(0)
-    except:
-        return torch.zeros((1, 512, 512, 3), dtype=torch.float32)
-
-def _load_video_tensor(seg: dict, frame_rate: float) -> torch.Tensor:
+def _load_video_tensor(seg: dict, frame_rate: float, input_dir: str = None) -> torch.Tensor:
     """Extracts a sequence of frames from a video file based on the segment's trim parameters,
     and returns them as an [N, H, W, 3] float32 tensor."""
-    file_path = os.path.join(folder_paths.get_input_directory(), seg.get("imageFile", ""))
+    file_path = _resolve_input_path(seg.get("imageFile", ""), input_dir)
     
-    if not os.path.exists(file_path):
-        return torch.zeros((1, 512, 512, 3), dtype=torch.float32)
+    if not file_path:
+        return _empty_image_tensor()
 
     trim_start_frames = float(seg.get("trimStart", 0))
     length_frames = float(seg.get("length", 1))
@@ -727,7 +760,7 @@ def _load_video_tensor(seg: dict, frame_rate: float) -> torch.Tensor:
         log.warning(f"[PromptRelay] Video extract error: {e}")
         
     if not frames:
-        return torch.zeros((1, 512, 512, 3), dtype=torch.float32)
+        return _empty_image_tensor()
         
     frames_np = np.array(frames, dtype=np.float32) / 255.0
     return torch.from_numpy(frames_np)
@@ -848,7 +881,7 @@ def _compress_image(tensor: torch.Tensor, crf: int) -> torch.Tensor:
         return tensor
 
 
-def _build_combined_audio(timeline_data_str: str, start_frame: int, duration_frames: int, frame_rate: float, override_audio: bool = False) -> dict:
+def _build_combined_audio(timeline_data_value, start_frame: int, duration_frames: int, frame_rate: float, override_audio: bool = False) -> dict:
     """Parses timeline JSON, loads/trims audio directly from memory using PyAV, 
     and aligns to a global timeline yielding ComfyUI's format.
     Output length explicitly mimics the timeline's duration_frames length."""
@@ -856,11 +889,14 @@ def _build_combined_audio(timeline_data_str: str, start_frame: int, duration_fra
     total_samples = max(1, int(math.ceil(duration_frames / frame_rate * target_sr)))
     empty_audio = {"waveform": torch.zeros((1, 2, total_samples), dtype=torch.float32), "sample_rate": target_sr}
 
-    if not timeline_data_str:
+    if not timeline_data_value:
         return empty_audio
 
     try:
-        data = json.loads(timeline_data_str)
+        if isinstance(timeline_data_value, str):
+            data = json.loads(timeline_data_value)
+        else:
+            data = timeline_data_value
         is_retake = data.get("retakeMode", False)
         if is_retake and data.get("retakeVideo"):
             retake_vid = data.get("retakeVideo")
@@ -888,17 +924,9 @@ def _build_combined_audio(timeline_data_str: str, start_frame: int, duration_fra
         buffer = None
         file_key = "videoFile" if override_audio else "audioFile"
         if seg.get(file_key):
-            file_path = os.path.join(folder_paths.get_input_directory(), seg[file_key])
-            if not os.path.exists(file_path):
-                # Try fallback under whatdreamscost subfolder
-                basename = os.path.basename(seg[file_key])
-                fallback_path = os.path.join(folder_paths.get_input_directory(), "whatdreamscost", basename)
-                if os.path.exists(fallback_path):
-                    file_path = fallback_path
-
-            if os.path.exists(file_path):
-                with open(file_path, "rb") as f:
-                    buffer = _io.BytesIO(f.read())
+            file_path = _resolve_input_path(seg[file_key])
+            if file_path:
+                buffer = file_path
         
         if not override_audio and not buffer and seg.get("audioB64"):
             b64 = seg.get("audioB64")
@@ -1266,6 +1294,8 @@ class LTXDirector(io.ComfyNode):
                 use_custom_audio=False, inpaint_audio=True, use_custom_motion=True, override_audio=False,
                 vae=None, reference_strength=1.0, ref_images=None,
                 start=None, end=None, duration=None) -> io.NodeOutput:
+        input_dir = folder_paths.get_input_directory()
+        image_cache = {}
 
         # Parse timeline data
         try:
@@ -1331,7 +1361,7 @@ class LTXDirector(io.ComfyNode):
                 for img_info in images_list:
                     image_b64 = img_info.get("b64", "")
                     file_name = img_info.get("name", "")
-                    tensor = _load_image_source(image_b64, file_name)
+                    tensor = _load_image_source(image_b64, file_name, cache=image_cache, input_dir=input_dir)
                     char_images.append(tensor)
                     slot_tensors.append(tensor)
                 char_slot_images.append(slot_tensors)
@@ -1370,37 +1400,30 @@ class LTXDirector(io.ComfyNode):
             for idx, seg in enumerate(img_segs):
                 seg_start = int(seg.get("start", 0))
                 offset = max(0, start_frame - seg_start)
+                seg_length = int(seg.get("length", 1))
+                seg_for_load = seg
 
                 if seg.get("type") == "video":
                     if offset > 0:
-                        seg["trimStart"] = float(seg.get("trimStart", 0)) + offset
-                        seg["length"] = max(1, int(seg.get("length", 1)) - offset)
-                    tensor = _load_video_tensor(seg, float(frame_rate))
+                        seg_for_load = dict(seg)
+                        seg_for_load["trimStart"] = float(seg.get("trimStart", 0)) + offset
+                        seg_for_load["length"] = max(1, seg_length - offset)
+                    tensor = _load_video_tensor(seg_for_load, float(frame_rate), input_dir=input_dir)
                 else:
-                    tensor = _load_image_tensor(seg)
+                    tensor = _load_image_tensor(seg, cache=image_cache, input_dir=input_dir)
 
                 # Apply resize
                 src_h, src_w = tensor.shape[1], tensor.shape[2]
-
-                def snap(val, div):
-                    return max(div, (val // div) * div)
-
-                if custom_width > 0 and custom_height > 0:
-                    # Both dimensions set — apply selected resize_method (pad, crop, stretch, maintain AR)
-                    tensor = _resize_image(tensor, custom_width, custom_height, resize_method, divisible_by)
-                elif custom_width > 0:
-                    # Width only — scale height from AR, snap both, then resize to exact dimensions
-                    tgt_w = snap(custom_width, divisible_by)
-                    tgt_h = snap(int(src_h * tgt_w / src_w), divisible_by)
-                    tensor = _resize_image(tensor, tgt_w, tgt_h, "stretch to fit", divisible_by)
-                elif custom_height > 0:
-                    # Height only — scale width from AR, snap both, then resize to exact dimensions
-                    tgt_h = snap(custom_height, divisible_by)
-                    tgt_w = snap(int(src_w * tgt_h / src_h), divisible_by)
-                    tensor = _resize_image(tensor, tgt_w, tgt_h, "stretch to fit", divisible_by)
-                else:
-                    # Both zero — keep original dimensions, just snap to divisible_by
-                    tensor = _resize_image(tensor, src_w, src_h, "maintain aspect ratio", divisible_by)
+                tgt_w, tgt_h, resize_override = _target_resize_dims(
+                    src_w, src_h, custom_width, custom_height, divisible_by
+                )
+                tensor = _resize_image(
+                    tensor,
+                    tgt_w,
+                    tgt_h,
+                    resize_override or resize_method,
+                    divisible_by,
+                )
 
 
                 # Apply compression
@@ -1413,7 +1436,7 @@ class LTXDirector(io.ComfyNode):
                     derived_w = tensor.shape[2]
 
                 if seg.get("isEndFrame"):
-                    insert_frame = max(0, seg_start + int(seg.get("length", 1)) - 1 - start_frame)
+                    insert_frame = max(0, seg_start + seg_length - 1 - start_frame)
                 else:
                     insert_frame = max(0, seg_start - start_frame)
                 strength = strengths[idx] if idx < len(strengths) else 1.0
@@ -1428,71 +1451,45 @@ class LTXDirector(io.ComfyNode):
                 src_h = derived_h if derived_h > 0 else 512
                 
                 # If there's an IC-LoRA video or retake base video on the timeline, extract its dimensions for accurate aspect ratio scaling
-                tdata_motion = json.loads(timeline_data) if timeline_data else {}
                 found_dims = False
                 
                 # Check for retake base video first
-                is_retake = tdata_motion.get("retakeMode", False)
-                retake_vid = tdata_motion.get("retakeVideo") or {}
+                is_retake = tdata.get("retakeMode", False)
+                retake_vid = tdata.get("retakeVideo") or {}
                 retake_file = retake_vid.get("imageFile", "") if isinstance(retake_vid, dict) else ""
                 if is_retake and retake_file:
-                    r_path = os.path.join(folder_paths.get_input_directory(), retake_file)
-                    if not os.path.exists(r_path):
-                        basename = os.path.basename(retake_file)
-                        fallback_path = os.path.join(folder_paths.get_input_directory(), "whatdreamscost", basename)
-                        if os.path.exists(fallback_path):
-                            r_path = fallback_path
-                    if os.path.exists(r_path):
-                        try:
-                            with av.open(r_path) as container:
-                                stream = container.streams.video[0]
-                                src_w = stream.width or stream.codec_context.width
-                                src_h = stream.height or stream.codec_context.height
+                    r_path = _resolve_input_path(retake_file, input_dir)
+                    if r_path:
+                        src_dims = _extract_video_dimensions(r_path)
+                        if all(src_dims):
+                            src_w, src_h = src_dims
                             found_dims = True
-                        except:
-                            pass
                 
                 # Fallback to normal motion segments
                 if not found_dims:
-                    for mseg in tdata_motion.get("motionSegments", []):
+                    for mseg in tdata.get("motionSegments", []):
                         v_file = mseg.get("videoFile")
                         if v_file:
-                            v_path = os.path.join(folder_paths.get_input_directory(), v_file)
-                            if not os.path.exists(v_path):
-                                basename = os.path.basename(v_file)
-                                fallback_path = os.path.join(folder_paths.get_input_directory(), "whatdreamscost", basename)
-                                if os.path.exists(fallback_path):
-                                    v_path = fallback_path
-                            if os.path.exists(v_path):
-                                try:
-                                    with av.open(v_path) as container:
-                                        stream = container.streams.video[0]
-                                        src_w = stream.width or stream.codec_context.width
-                                        src_h = stream.height or stream.codec_context.height
+                            v_path = _resolve_input_path(v_file, input_dir)
+                            if v_path:
+                                src_dims = _extract_video_dimensions(v_path)
+                                if all(src_dims):
+                                    src_w, src_h = src_dims
                                     found_dims = True
                                     break
-                                except:
-                                    pass
 
                 # Create a dummy tensor of the exact source dimensions
                 tensor = torch.zeros((1, src_h, src_w, 3), dtype=torch.float32)
-
-                def snap(val, div):
-                    return max(div, (val // div) * div)
-
-                # Route the dummy tensor through the exact same resizing pipeline
-                if custom_width > 0 and custom_height > 0:
-                    tensor = _resize_image(tensor, custom_width, custom_height, resize_method, divisible_by)
-                elif custom_width > 0:
-                    tgt_w = snap(custom_width, divisible_by)
-                    tgt_h = snap(int(src_h * tgt_w / src_w), divisible_by)
-                    tensor = _resize_image(tensor, tgt_w, tgt_h, "stretch to fit", divisible_by)
-                elif custom_height > 0:
-                    tgt_h = snap(custom_height, divisible_by)
-                    tgt_w = snap(int(src_w * tgt_h / src_h), divisible_by)
-                    tensor = _resize_image(tensor, tgt_w, tgt_h, "stretch to fit", divisible_by)
-                else:
-                    tensor = _resize_image(tensor, src_w, src_h, "maintain aspect ratio", divisible_by)
+                tgt_w, tgt_h, resize_override = _target_resize_dims(
+                    src_w, src_h, custom_width, custom_height, divisible_by
+                )
+                tensor = _resize_image(
+                    tensor,
+                    tgt_w,
+                    tgt_h,
+                    resize_override or resize_method,
+                    divisible_by,
+                )
                 
                 guide_data["images"].append(tensor)
                 guide_data["insert_frames"].append(0)
@@ -1703,7 +1700,7 @@ class LTXDirector(io.ComfyNode):
                 )
 
         # --- Build Audio Output ---
-        audio_out = _build_combined_audio(timeline_data, start_frame, ltxv_length, float(frame_rate), override_audio=override_audio)
+        audio_out = _build_combined_audio(tdata, start_frame, ltxv_length, float(frame_rate), override_audio=override_audio)
 
         # --- Audio Latent Generation ---
         audio_latent = {}
@@ -1834,7 +1831,6 @@ class LTXDirector(io.ComfyNode):
         # --- Motion guide output from timeline video segments ---
         motion_guide_data = {"segments": [], "frame_rate": float(frame_rate), "duration_frames": int(duration_frames), "resize_method": resize_method}
         try:
-            tdata = json.loads(timeline_data) if timeline_data else {}
             if use_custom_motion:
                 motion_segments = tdata.get("motionSegments", [])
             else: