feat(minimax-h3): default workflows; starter-model decision documented (PR 5/5) - #60
Merged
Merged
Conversation
Two default workflows following the Wan video precedent. The i2v variant wires the one First Frame image node and shared Width/Height integer nodes to the prompt, frame-conditioning and denoise nodes so the backend's keyframe-coupling contract holds structurally. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The installer cannot express 'root layout minus the Ref2VA subtrees': a bare repo source pulls ~498 GB and a subfolder-set source drops the root modular_model_index.json the probe requires. Records the unblock options and the MiniMax H3 Community License constraints any future entry must carry. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ult workflow Adds a 'MiniMax H3 Ideal Dimensions' node (mirroring the Wan 2.2 Ideal Dimensions pattern): a thin wrapper over the vendored resolve_canvas_size policy — 768px short edge, soft 768x1344 area cap, both axes snapped to the 32 grid, aspect ratios 1:4 through 4:1. Unlike the Wan nodes there is no resolution preset or rounding mode: H3 was released for exactly one canvas family and the policy is the released pipeline's, verbatim. The 'First Frame to Video - MiniMax H3' default workflow now derives the canvas from the uploaded image (image width/height -> ideal dimensions -> prompt/frame-conditioning/denoise), matching 'Image to Video - Wan 2.2 Lightning'. The manual Width/Height form fields are gone - previously a mismatched aspect ratio stretched the keyframe onto whatever canvas was typed in. openapi.json/schema.ts regenerated for the new node. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
lstein
requested review from
JPPhoto,
Pfannkuchensack and
blessedcoolant
as code owners
August 7, 2026 18:10
github-actions
Bot
added
python
python-tests
backend
services
frontend
invocations
labels
Only conflict was the generated schema.ts, regenerated from the merged source (scripts/generate_openapi_schema.py + pnpm typegen) rather than hand-merged. openapi.json auto-merged and was regenerated alongside it. starter_models.py auto-merged, so it was checked by hand: main's FLUX.2 [dev] region and this branch's MiniMax H3 region are both present in the result. Checked for the two failure modes the earlier syncs in this stack hit: no persisted-state migration collision (this branch touches no redux slice), and the five vendored minimax_h3 files still differ from huggingface/diffusers@abc5e9bf71 by import lines only. Verified: 149 passed / 4 skipped backend, 1738 frontend tests across 144 files, tsc clean, ruff check + format clean. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Note
Port of lstein#144 to the InvokeAI-7 tree. Mechanical replay of the original commits onto v7
main(4a48a37967);openapi.json/schema.tsregenerated and verified drift-free at every level, and thedependencies.pyauto-merge verified by hand. All changes target the backend and the existingfrontend/webapp -webv2is untouched (H3 support in the new UI is future work). Review history and test notes live on the original PR.Note
Supersedes lstein#142, which GitHub auto-marked as merged (and auto-deleted the branch) when a botched force-push briefly made the head ref equal to its base during a stack rebase. No content was ever merged; this PR carries the identical (rebased) commits.
Summary
PR 5 of 5 in the MiniMax H3 stack (on top of lstein#141): default workflows, plus a documented decision on the starter-model entry. Two commits, deliberately separated so the license-sensitive part stays droppable.
Commit 1 — default workflows
video_latents/audio_latentsand the video+audio VAE edges. Form exposes model, prompt, width/height, frames, steps. Notes cover the 17n+5 frame grid, fixed 24 fps, 32-px dimension grid, muxed AAC audio, and the absence of a negative prompt.minimax_h3_frame_conditioning. The backend's keyframe-coupling contract is enforced structurally: a single image node feedsfirst_imageon both the prompt and frame-conditioning nodes, and shared width/height integer nodes feed prompt + frame conditioning + denoise — form edits cannot desynchronize them.Commit 2 — starter model: documented TODO, no entry
Neither expressible install source works, so no entry ships yet:
MiniMaxAI/MiniMax-H3source downloads ~498 GB —filter_fileskeeps everymodel*.safetensorsacross all subtrees, includingFL2VA/,Ref2VA/, andtransformer_ref/.::subfolder+subfoldersyntax keeps only files inside the listed subfolders and silently drops the rootmodular_model_index.jsonthat the H3 probe hard-requires — the install would land as an unknown model.The TODO in
starter_models.pyrecords both unblock paths (a root-config-awarefilter_files, or an upstream FL2VA-only repo) and the MiniMax H3 Community License constraints (territory exclusions extend to outputs; "MiniMax H3" attribution). Until then, users install the model manually by pointing the Model Manager at a locally assembled fl2va folder (~42.5 GB working set).Testing
tests/app/services/model_install+tests/app/services/workflow_records+tests/model_identification: 146 passed, 0 failures; ruff clean.WorkflowValidator.validate_json, and every edge handle was checked against the actual invocation input/output models.🤖 Generated with Claude Code