ComfyUI-MiniMaxH3-Context-Loop
Clip chaining for MiniMax H3 in ComfyUI - motion and audio genuinely continue across joins
Nodes (83)
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
Vocals drive the lip-sync, the full mix stays the soundtrack — one node keeps them straight
Stitch your saved H3 scenes into one MP4
Render a whole episode scene by scene, then ship each chapter on its own schedule
Pull one finished chapter back out of a run without dragging the whole project along
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
The Plan node that runs your whole MiniMax H3 movie — now with its wires in order
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
You'll never add this node by hand — and that's exactly why it matters
Lazily load one MiniMax H3 checkpoint for its upscale pass, exactly as saved
The node that eats the big frames so the next scene doesn't blow up your VRAM
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
Second-pass upscaling conditioning that actually matches your new canvas
This node hands you the frames
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
One node where prompt-driven references meet the scene loop
The (Internal) packer you're only meeting because ComfyUI expanded a node
Drift Control across a sigma split, without duplicating the model
Two dropdowns that decide how every MiniMax H3 scene connects
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
Making H3 lipsync a real song instead of mumbling over it
Cut the pinned frames off H3 output, picture and sound together
The placeholder node that makes sure patches don't fight
The path-backed asset carousel that kills a wall of H3 loader nodes
Dial back H3's reference video grip instead of dropping frames
Normalize any source into a clean one-pass H3 performance reference
The socket that changes type to hand you one scene fact
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
The settings bundle that tunes your tagged references in one place
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 Context Loop
Build a multi-scene MiniMax H3 video with one reusable sampling graph. Review each scene, compare takes, resume from checkpoints, and assemble the result later—without keeping the whole production in memory.
Getting started · Workflows · Changelog · Documentation
Changelog
0.6.9 — Plan controls and workflow loading fixes
- Combine repeated Plan refreshes during workflow loading.
- Fix default steps and fingerprint input visibility; keep deliberate scene overrides, with an explicit option to clear them.
- Add a separate-loader source-audio example without the Carousel.
0.6.8 — Fractional H3 mask correction
- Bring nightly's fractional video/audio denoise-mask correction to main, including feathered AV continuity and native-first compatibility.
- Correct only missing streams; leave native fixes and source latents intact.
0.6.7 — Automatic PNG sequence variants
- Conflicting VIDEO PNG exports automatically use
_2,_3, etc. folders, preserving earlier exports and keeping subsequent scenes in one variant. - Copy verified earlier scenes safely when a rerender changes mid-sequence; retain bounded memory and interrupted-publication recovery.
0.6.6 — Review Gate capture and resizing
- Capture a still from a saved Review Gate preview into the Project Asset Carousel, with safe project selection and numbered tags for new takes.
- Resize the optional prompt editor using its new drag handle; height is saved with the workflow and double-click resets it.
0.6.5 — Upscale anchor overrides and clear checkpoint sources
- Choose semantic-anchor size and mode for cached, rebuilt, or connected upscale references, including the VIDEO conditioning node.
- Checkpoint Manager explicitly names the output branch, available scenes, Original/DeRoPE source and fallback scenes, independently of preview browsing.
0.6.4 — Preserve reference sizing during recovery
- Recover saved Max sizing through linked inputs and Tagged Scene Options.
- Apply Match/Max upscale overrides to automatic caches too, rebuilding native picture conditioning from original masters without changing the source cache.
0.6.3 — Cancellation-safe processing saves
- Preserve saved DeRoPE/upscale scenes when manifest updates or network acknowledgements fail.
- Recover interrupted PNG publication from a scene journal; accept recreated video containers when their delivered pixels match the existing PNGs.
- Keep earlier scenes and conflicting partial files safe. See cancellation and scene-level resume.
0.6.2 — Upscale recovery and saving
-
Recover lost reference caches. Upscale rebuilds missing Ref2VA tensors from verified saved reference media. Per-reference tensor files avoid repeated bundles; legacy caches remain readable and can be converted safely.
-
Reliable deferred sources. Full-sequence upscale respects selected final-cut alternates and saved DeRoPE sources. Checkpoint previews no longer shorten the processing range. Processing tabs expose saved versions and guarded cleanup.
-
Scene-by-scene PNG export. The existing PNG Sequence + Audio node accepts file-backed VIDEO and passes it through after saving each scene, with selectable 8/16-bit PNGs, bounded memory, and network-storage-safe reuse.
-
Exact recovery records. New takes keep their own Plan/workflow snapshots; later edits cannot replace those saved settings. Original media must still exist for a lost reference cache to be rebuilt.
-
Grouped source audio. Assign full mix, vocals and instrumental in the Carousel or Audio Tracks node. Vocals drive lip-sync; the full mix stays the soundtrack. Per-scene Lip-sync On/Off controls do not alter delivery.
Context Loop execution and recovery
- Maintained workflow path. The release docs now describe the proven memory-safe top-level prompt lifecycle: keep the same Plan and creative model stack, let Loop End finish, wait through the cleanup delay, then queue the next heavyweight scene as a new top-level prompt.
- Reference propagation fix. Valid prompt
@tagsagain see the connected Tagged registry during preflight without rewriting prompt text or storing reference data in the Plan. - Crash-safe review and resume. Review snapshots stay visible after a refresh or restart, and durable handoffs/manual resume keep the same Plan semantics while avoiding duplicate queues.
- Release packaging. The changelog, compatibility notes, and workflow docs now call out the WSL2 pinned-memory caveat separately from the architectural fix.
0.6 — Better authoring, recovery, and export
- Production Plan. The familiar scene columns, with settings organized into Project, Canvas, Generation, and Delivery. The original Plan node stays available; migration is optional.
- Context Builder. Compose picture context from multiple saved-scene windows, choose their frame repartition, and adjust boundary controls in Plan Studio. Audio follows the predecessor by default and can be unlocked for independent context selection.
- Plan Studio and review. Non-destructive trims, picture-only final-cut alternates, and scene LoRA routes. Candidate selection now keeps seeds and settings synchronized across the Plan and editors.
- Project Asset Carousel. More reliable project switching and previews, editable semantic-anchor size/mode with downstream inheritance, and video references that keep their own duration.
- Recovery and editing fixes. Safer checkpoint rollback, stale-response protection when switching projects, and fixes for empty-scene navigation, archive restoration, and source audio after scene resizing.
- Faster PNG + WAV export. Parallel PNG saving, progress and timing logs, plus synchronized audio export. Connect the video VAE, audio VAE, or both.
- Rebuilt 0.6 workflows. Clean settings, readable layouts, and wiring checked against the release nodes. Studio examples include the Carousel and Checkpoint Manager; older workflows are kept in the archive.
- Experimental pixel upscaling. A scene-by-scene DLSS5 + USDU example with conditioning matched to the actual upscaled image size and original audio preserved. Full external GPU refinement testing is still pending.
Full changelog · 0.5 → 0.6 visual overview
<details> <summary>Earlier milestones</summary>
- 0.5: generation profiles, candidate review, checkpoint branches, project assets, and deferred upscale workflows.
- 0.4: tagged references, Studio authoring, saved-run recovery, and masked video/audio editing.
- 0.1–0.3: the recursive scene loop, review gate, per-scene checkpoints, prompt editor, and archival PNG export.
Install
From ComfyUI/custom_nodes:
git clone https://github.com/seitanism/ComfyUI-H3-Motion-Context-MultiRef.git
git clone https://github.com/ethanfel/ComfyUI-MiniMaxH3-Context-Loop.git
main is the stable release line. Use nightly for development features.
Restart ComfyUI. Use a build with native MiniMax H3 Add Guide support and
install your H3 model, text encoder, video VAE, and audio VAE; models are not
bundled. ffmpeg on PATH is recommended for review and assembly. Optional
upscale packs and example assets are listed in the workflow catalog.
First run
- Open T2V Normal and select the four model files.
- Give Production Plan a unique
run_name, edit the scene prompts, and keep the default Generation Profile for a first test. - Queue. Preflight checks the plan, then the loop renders and checkpoints one scene at a time.
- At Review Gate, approve, retry, reroll, or approve and stop. After the last approval, Assemble writes the final MP4.
For Studio workflows, set the run name in Project Asset Carousel instead. See Getting started for setup and recovery steps.
To reuse a still as a reference, scrub the saved Review Gate preview, click
Capture frame…, check the destination project and tag, then Save to
Carousel. Reusing a tag creates a numbered take without replacing the
original asset or video. Capture requires ffmpeg.
Choose a workflow
| Starting point | Workflow | |---|---| | Text | T2V Normal | | First image / first and last images | I2V Normal / FL2V Normal | | Tagged references and timeline editing | Ref2V Studio | | References with a source soundtrack | Ref2V Studio Source Audio | | Inpaint, extend, bridge, or upscale video | All examples and requirements |
Normal keeps the scene-column editor. Studio adds an optional experimental timeline interface without changing the sampling graph.
<details> <summary>How the graph is organized</summary>How the graph is organized
Only the current scene enters the sampler. Loop End advances to the next scene or passes a manifest to Assemble. The muted Load Manifest → Assemble branch can assemble saved scenes without rendering again.
</details>Updating from 0.5
Back up your workflows and project folders before updating. Existing Plan nodes remain supported; the new examples have 0.6 in their filenames, and the older examples are preserved under Archive/0.5. Restart ComfyUI after updating and hard-refresh the browser.
Keep the same run_name to resume the same production; use a new one for a
different project. Saved runs live in output/h3_chains/<run_name>/, with final
movies under final/. Carousel media lives in input/h3_projects/<run_name>/
and is mirrored into the run for recovery. These paths follow your configured
ComfyUI input/output directories. Resume checks still reject incompatible
generation changes; see Runs and recovery.
The repository is now spelled Context-Loop. Existing installs named
ComfyUI-MiniMaxH3-Contex-Loop can stay in place—do not install a second copy
just to change the folder name.
Origins and license
Started from NikoDemon80's H3 Motion Context. See Feature origins and Third-party notices for contributors and upstream work.
GPL-3.0 · License · Contributing