Add same-day LTX and MiniMax workflow packs

2026-08-11 18:06:40 +00:00
parent ae16868927
commit bd51936b29
6 changed files with 849 additions and 0 deletions
@@ -0,0 +1,48 @@
+# LTX 2.3 INT8 MSR
+
+## Summary
+
+Same-day `LTX 2.3` multiple-subject reference workflow from `javawock7618`. This is the more practical LTX drop if you care about mixing several references into one scene instead of just doing plain text-to-video.
+
+## Model Family
+
+- `ltx`
+
+## Status
+
+- `imported-only`
+
+## Source
+
+- Workflow pack page: <https://huggingface.co/javawock7618/comfy-ltx-2.3-workflows>
+- Workflow file: <https://huggingface.co/javawock7618/comfy-ltx-2.3-workflows/resolve/main/LTX2.3_int8_MSR-javano2608.1.json?download=true>
+- Original publisher: `javawock7618`
+- Date imported: `2026-08-11`
+
+## Developer notes
+
+- The source README says this workflow can blend up to 5 inputs, including 1-4 subjects plus a background.
+- The pack is built around distilled GGUF and INT8 paths for `lightricks/ltx-2.3`.
+- The same pack also calls out an INT8 speedup path, prompt enhancement, and optional voice support in other graphs.
+
+## Our notes
+
+- This is one of the more interesting LTX variants for consistency and scene control.
+- On an RTX 3070 / 8 GB setup it is still tight, but it is a real workflow rather than a placeholder.
+- Good pick when you want a reference-heavy LTX path instead of a generic one-shot generator.
+
+## Required custom nodes
+
+- `ComfyUI-KJNodes`
+- `ComfyUI-LTXVideo`
+- the keyframe/director nodes named in the source README if ComfyUI flags them as missing
+
+## Required models
+
+- `lightricks/ltx-2.3` distilled GGUF or INT8 checkpoints
+- `ltx-2.3-spatial-upscaler-x2-1.1` for the detailer / upscaler lane
+
+## Notes
+
+- Best LTX import from this batch if you want reference blending and multi-subject handling.
+- Expect to trim resolution and keep the graph lean on an 8 GB card.
@@ -0,0 +1,49 @@
+# MiniMax H3 INT8 R2V
+
+## Summary
+
+Same-day `MiniMax-H3` reference-to-video workflow from `javawock7618`. This is the MiniMax drop to watch if you care most about keeping a subject stable from a reference image while still getting motion and camera movement.
+
+## Model Family
+
+- `minimax`
+
+## Status
+
+- `imported-only`
+
+## Source
+
+- Workflow pack page: <https://huggingface.co/javawock7618/comfy-MiniMax-H3-workflows>
+- Workflow file: <https://huggingface.co/javawock7618/comfy-MiniMax-H3-workflows/resolve/main/MiniMax_int8_R2V-javano2608.12.json?download=true>
+- Original publisher: `javawock7618`
+- Date imported: `2026-08-11`
+
+## Developer notes
+
+- The source README positions this as the reference-guided workflow in the pack.
+- It is built around INT8 inference plus the `ComfyUI-H3-Motion-Context` and `ComfyUI-Spectrum-MiniMax-H3` ecosystem.
+- The pack also recommends `ComfyUI-MiniMaxH3_LatentUpscaler` for post-upscale.
+- The R2V lane requires `minimax_h3_ref_lora_rank_256_bf16.safetensors` from `Kijai/MiniMax-H3-experimental`.
+
+## Our notes
+
+- This is the most directly useful MiniMax reference-control item from today.
+- On a 3070 it is still heavy, but it is the kind of workflow that could actually be worth testing if you want stronger subject hold than a generic video model.
+- Better fit than a novelty demo because it is explicit about the reference LoRA and the supporting nodes.
+
+## Required custom nodes
+
+- `ComfyUI-Spectrum-MiniMax-H3`
+- `ComfyUI-H3-Motion-Context`
+- `ComfyUI-MiniMaxH3_LatentUpscaler`
+
+## Required models
+
+- `Comfy-Org/MiniMax-H3`
+- `minimax_h3_ref_lora_rank_256_bf16.safetensors` from `Kijai/MiniMax-H3-experimental`
+
+## Notes
+
+- Lightx2v LoRA and Turbo-LoRA are optional, but the source pack treats them as useful accelerators.
+- This is the MiniMax workflow I would try before the more novelty-leaning releases.
@@ -0,0 +1,47 @@
+# MiniMax H3 Seamless Chain Core
+
+## Summary
+
+Same-day `MiniMax-H3` multishot chain workflow from `joeygambino`. It is built to keep one continuous scene flowing across shot boundaries, with continuous audio instead of obvious cuts.
+
+## Model Family
+
+- `minimax`
+
+## Status
+
+- `imported-only`
+
+## Source
+
+- Workflow pack page: <https://huggingface.co/joeygambino/MiniMax-H3-Multishot-Workflow>
+- Workflow file: <https://huggingface.co/joeygambino/MiniMax-H3-Multishot-Workflow/resolve/main/workflows/H3_Seamless_Chain_CORE.json?download=true>
+- Original publisher: `joeygambino`
+- Date imported: `2026-08-11`
+
+## Developer notes
+
+- The source README says the CORE workflow uses only built-in ComfyUI pieces plus the released H3 node pack logic.
+- The full pack adds Motion Context, RES4LYF, sol-attn, blockcache, custom scripts, and GGUF support, but CORE is the cleanest starting point.
+- The release expects MiniMax-H3 weights plus the matching encoder and VAE files from `Comfy-Org/MiniMax-H3`.
+
+## Our notes
+
+- This is the strongest today-dropped MiniMax workflow for scene continuity.
+- On an RTX 3070 it is still heavy, but the CORE version is the right shape if you want to test the idea without dragging in every extra dependency.
+- Best fit when you care more about continuity across shots than flashy one-off clips.
+
+## Required custom nodes
+
+- current ComfyUI
+- the upstream `ComfyUI-H3-Multishot` pack if you want the shipped node classes
+
+## Required models
+
+- `Comfy-Org/MiniMax-H3`
+- the matching MiniMax-H3 text encoder and VAEs from the same model family
+
+## Notes
+
+- The full release also documents a reference-image lane, but that lives in the non-CORE workflow.
+- This one is the easier starting point for local testing and repo maintenance.