ComfyUI-MiniMaxH3-Contex-Loop
Clip chaining for MiniMax H3 in ComfyUI - motion and audio genuinely continue across joins
Nodes (68)
The node that rescales your reference images to match an upscaled H3 render
Layer a named transition onto your H3 chain without nuking the audio plan
Stitch your saved H3 scenes into one MP4
Browse every take of every scene, then upscale just the branch you want
Where your previous scene's motion and audio actually get carried over
Prompt, seed, length and audio window for whatever shot is rendering
Splice the previous HQ tail back in before pass 2 re-times the scene
Stop pass 2 from re-texturing the frames you already locked down
The guard that keeps de-roping from breaking your scene boundaries
Render your whole accepted chain to a lossless PNG sequence
Start an H3 chain from footage you already have
Opening image for scene one, nothing for the rest
Alternate end-frame targets per scene for A/B endpoint chains
Feed your whole finished chain to SeedVR2 as one continuous movie
Finish a shot, hand off, run the next
Start or resume an H3 chain — yes, resuming is the good part
Swap LoRA'd models per scene without ever loading one
Rebuild your manifest from disk and skip the re-render
Turn a video-only upscale into a full H3 pass-2 latent
One definition that drives every scene in your H3 movie
Write your whole MiniMax H3 movie before H3 ever loads a weight
One knob for how your MiniMax H3 scenes connect — transition and audio in a single wire
The node that argues with your MiniMax H3 Plan before the GPU does
Put the de-roped upscale back on the source scene's clock
The pause button that turns 'rerun the whole chain' into 'fix one scene'
The experimental scene editor that colors your MiniMax H3 references and can even rewrite them by AI
Reopen a finished MiniMax H3 production and its reference media, months later, from disk
A proper text editor for your MiniMax H3 scene prompts, because the node-graph text box is not it
Checkpoint your MiniMax H3 scene before you're allowed to be happy with it
Upscale an already-checkpointed MiniMax H3 chain as a separate child run
Lazily load one MiniMax H3 checkpoint for its upscale pass, exactly as saved
The recursive engine that walks your MiniMax H3 upscale chain, one scene at a time
Stitch your upscaled MiniMax H3 scenes into one HQ MP4
Rebuild MiniMax H3's pass-2 conditioning from the checkpoint cache, no original wires needed
Persist one upscaled MiniMax H3 scene so the rest of the chain (and a Drift-Control scene) can use it
Slice your tracked mask to exactly the scene H3 is about to render
Did your H3 clip actually continue the last one?
Copy the real source-video interval into H3's joint target — frame-locked, sound included
Generate the missing middle between two real MiniMax H3 clips
Attach H3's causal video/audio denoise mask to a real AV target — the heart of inpainting
Your H3 inpaint mask is lying to you — snap it to the real grid
Lock a real soundtrack into H3 without letting denoising mangle it
Clip source footage to a length H3 can actually eat
Drift Control across a sigma split, without duplicating the model
A motion-reference loader that doesn't decode the video (until it has to)
See exactly which motion window a scene will pull from
Migrate a 0.4 Plan to 0.5 without re-deciding your audio
Cut the pinned frames off H3 output, picture and sound together
The placeholder node that makes sure patches don't fight
Dial back H3's reference video grip instead of dropping frames
Normalize any source into a clean one-pass H3 performance reference
Give one scene a voiceover without renumbering every reference
Keep one hero face per scene, without rewriting your prompts
The scheduler that compiles your @tags into the right native labels
Schedule one performance video per scene, soundtrack included
Wire every semantic anchor into one socket
The internal stage that rebuilds H3's Qwen presentation around your anchors
Register a story-beat image H3's Qwen can see but native slots never count
One wire to describe your whole source media, decoded only when needed
Check which slice of your footage a scene actually gets
H3 audio references under a stable @tag
H3 motion transfer as a reusable @motion subject
The lazy VIDEO/path variant
File-backed, timeline-typed, and lazy by design
The smallest node in MiniMax H3's prompt-driven reference system
How @tags become native H3 references, scene by scene
Register a whole clip under @performance and let each scene call it when it needs it
Fix the upscale pass without touching your generation graph
ComfyUI MiniMax H3 Contex Loop
Build a multi-scene MiniMax H3 video with one reusable sampling body. Review each scene, retry mistakes, resume interrupted runs, and assemble accepted scenes from disk.
Getting started · Choose a workflow · Troubleshooting · Full wiki
Version 0.5 status:
mainis the supported 0.5 release line. Saved 0.4 workflows and checkpoints remain supported.
Contex is the intentional public repository spelling.
What you get
| Goal | This pack provides | |---|---| | Make a longer story | A visual scene Plan drives one recursive H3 graph. | | Keep motion and sound connected | Guide and protected AV-prefix transitions. | | Direct each scene | Prompts, seeds, timing, pictures, motion video, and audio references. | | Work from existing footage | Source timelines, clip continuation, inpainting, and two-ended bridges. | | Iterate safely | Review, edit and retry, reroll, stop early, and atomic checkpoints. | | Recover the production | Resume, partial assembly, saved assets, and latent-to-PNG export. |
Quick start
From ComfyUI/custom_nodes:
git clone https://github.com/ethanfel/ComfyUI-MiniMaxH3-Contex-Loop.git
Restart ComfyUI, then:
- Open a maintained workflow from
example_workflows/. - Resolve missing model selections; models are not bundled.
- Give Plan a unique
run_nameand edit its scene prompts. Use{wide shot|close-up}for alternatives. Each scene's Prompt alternatives control can derive a stable scene seed, keep an exact fixed seed, or randomize that scene on every queue. These choices never change sampler seeds, and the exact resolved choice is saved with the checkpoint. - Keep Guide and Generated audio for a simple first run.
- Queue the graph. Preflight checks timing, media, references, compatibility, and resume state before H3 loads.
- At Review Gate, approve, retry, reroll, or approve and stop. To compare
several takes per scene, set its optional
candidate_countabove 1 (or convert it to an input and connect an INT node); Review Gate generates the candidates automatically and continues from the exact take you select. - Assemble the completed or partial manifest.
Version 0.5 expects a current ComfyUI build containing native Add Guide for
MiniMax H3 from ComfyUI PR #15439.
ffmpeg on PATH is preferred; ComfyUI's bundled PyAV can handle review and
assembly when FFmpeg is unavailable.
Some examples need bundled media copied into ComfyUI/input/. See the
asset guide.
Choose a workflow
| I want to… | Start here | |---|---| | Generate from text | T2V Normal | | Animate an opening image | I2V Normal | | Move between first/last images | FL2V Normal | | Use prompt-driven pictures | Ref2V Tagged | | Guide scenes with a source soundtrack | Ref2V Studio Tagged Source Audio | | Inpaint a fixed or tracked region | Masked Video Inpaint | | Inpaint with a picture-defined replacement | Ref2V Masked Video Inpaint | | Continue one existing clip | Masked AV Extension — Single Clip | | Continue several reviewed scenes | Masked AV Extension — Chain | | Generate the gap between two clips | Two-Clip Masked AV Bridge |
Choose Normal for the standard Plan and Scene Prompt Editor. Studio workflows add an optional experimental timeline interface without changing the generation graph. The wiki workflow chooser explains every maintained example and required asset.
How it works
Chain Policy → [Advanced] → [Legacy 0.4] → Plan
Source Timeline ───────────────────────────┘
↓
Preflight → Loop Start → Current Shot
↓
H3 conditioning → sample → decode
↓
trim → checkpoint → review → Loop End ─↺
Loop End manifest → Assemble
Only one scene passes through the sampling body at a time. The accepted predecessor supplies continuity to the next scene; completed media and recovery metadata remain on disk.
Find help by task
| Task | Guide |
|---|---|
| Install and run the first scene | Getting started |
| Choose Cut, Guide, Hard AV, Soft AV, or audio behavior | Continuity and audio |
| Use @tags, motion references, or Source Timeline | References and source media |
| Inpaint, outpaint, extend, or bridge footage | Masked editing |
| Retry, resume, recover, or assemble later | Review, resume, and recovery |
| Upscale a completed checkpoint branch | Runs, review, and recovery |
| Diagnose a problem | Troubleshooting |
| Check where a feature came from | Feature origins |
| Look up Plan fields and implementation details | Advanced reference |
Repository-native references remain available under docs/ and in
the complete Plan format guide.
Checkpoint Manager identifies saved takes by scene and inferred branch, previews
saved media and exact video/audio dependencies, and safely deletes inactive
leaves one revision at a time. Its Plan and Source Timeline pass-throughs can
remain connected in generation workflows, while its selected-manifest output
launches a standalone deferred upscale loop with no source Plan. Each profile
is isolated under upscaled/<profile>, and saving the large HQ latent is optional.
The bundled chain-aware de-rope variant combines LBH 3D with MAINodes in the
same second pass, protects recursive scene boundaries, restores the prior HQ
Drift-Control tail, and returns recovered video/audio to the source clock
before checkpointing.
For whole-video SeedVR2 finishing, Full-Chain Latent Video Adapter instead
re-decodes every selected H3 checkpoint into one cached, lossless, file-backed
movie. It resolves scene overlaps before upscaling and uses a temporary
disk-backed VAE output buffer, so the complete production never becomes one
in-memory IMAGE tensor.
Tagged and Scheduled Ref2VA also cache each active scene's native reference
latents and compact Qwen presentation automatically; the upscale loop restores
them from the checkpoint fingerprint without original reference-media wires.
See
Runs, review, and recovery.
Origins and license
This project began with NikoDemon80's H3 Motion Context and grew into a separate checkpointed production-loop pack. Original, adapted, inspired, integrated, and compatibility work is mapped in Feature traceability; licenses and exact upstream revisions are recorded in Third-party notices.
GPL-3.0. See LICENSE. Contributions are covered by CONTRIBUTING.md.