v1.3.0 – Flexible H3 Conditioning & Qwen Guides

Herrgotts H3 Infinite Continuation Suite v1.3.0

v1.3.0 adds flexible native MiniMax H3 conditioning and Qwen image guides while keeping the proven v1.2 continuation and stitching core intact.

Highlights

• New Flexible Start / Conditioning v1.3 node.
  • First Frame and Last Frame are independent optional inputs.
  • Supports T2VA, I2VA, L2VA and FL2VA from one node.
• New Continue from Latent v1.3 node.
  • Keeps direct phase-aligned video + audio latent continuation.
  • Adds optional Last Frame conditioning for a fresh visual endpoint / quality reset.
• Added auto-growing Qwen Reference inputs.
  • Starts with Qwen Reference 1 and grows up to 9 references as sockets are connected.
  • Qwen References are text/vision-encoder-only images and are not inserted as persistent native Ref2VA/DiT reference latents.
• Added deterministic <Picture N> mapping and a picture_map diagnostic output.
  • Connected First Frame comes first.
  • Connected Last Frame comes next.
  • Qwen References follow in numeric order.
  • Direct continuation latent context is not silently counted as a Qwen Picture.
• Existing v1.2.x nodes remain registered for workflow compatibility.
• Updated all four example workflows and Registry metadata to v1.3.0.

Continuation core

The established continuation path remains unchanged: direct AV-latent handover, freeze-aware Auto Handover, phase-aligned cutoffs, Safe Tail Bridge, rendered seam smoothing and memory-bounded Saved Chain Stitching.

Save / Load clip indices remain manual in v1.3.0 so existing clips can still be selected and regenerated predictably.

Validation

The release source passes the complete automated regression suite:

82 passed

The v1.3 conditioning path was also live-tested with Qwen Reference autogrow, First/Last Picture ordering, multiple Qwen References with separate prompt roles and longer multi-clip continuation.

Updating

Update through ComfyUI Manager, or update the repository in ComfyUI/custom_nodes/Herrgotts-H3-Infinite-Continuation-Suite and restart ComfyUI.

Existing v1.2.x workflow class IDs remain available, so older workflows are not silently reinterpreted by the new v1.3 conditioning nodes.

See README.md, VALIDATION.md and CHANGELOG.md for full details.

v1.2.2 – ComfyUI 0.33 Compatibility

ComfyUI 0.33 compatibility

This release restores H3 Infinite Continuation compatibility with the new MiniMax H3 API introduced in ComfyUI 0.33.

What's changed

• Added automatic support for the new ComfyUI 0.33 MiniMax H3 layout API.
• Continuation video anchors now use ComfyUI's native arbitrary H3 keyframe placement on 0.33+.
• The previous MiniMaxH3.extra_conds payload monkey patch is no longer required on the new API.
• Replaced the old layout patch on ComfyUI 0.33+ with a much smaller marker-gated audio timeline compatibility wrapper.
• Fixed repeated Continue generations in the same ComfyUI session, including Clip 3+.
• Older ComfyUI H3 layouts remain supported through the legacy compatibility path.
• Updated workflow Registry metadata to v1.2.2.

Validation

• 72/72 regression tests passing.
• Live-tested on ComfyUI 0.33.0.
• Multiple consecutive Continue generations completed successfully in the same ComfyUI session.
• Video continuation, native audio continuation, saved latent chaining and new Last Frame landing verified.

If you updated to ComfyUI 0.33, please update the node pack to v1.2.2.

v1.2.1 – Manager Integration Hotfix

v1.2.1 – Manager Integration Hotfix

Small metadata and packaging hotfix for the initial v1.2 release.

Changes

• Added explicit Comfy Registry metadata to all included workflow nodes
• Added node_list.json for improved ComfyUI Manager node-pack detection
• Improved compatibility with Check Missing Custom Nodes
• Updated bundled workflows to v1.2.1 metadata
• No changes to generation, continuation, freeze detection, Safe Tail Bridge or stitching behavior

v1.2.0 – Initial Release

Herrgotts H3 Infinite Continuation Suite v1.2.0

First public release of the H3 Infinite Continuation Suite for ComfyUI.

The suite combines repeated FL2VA First/Last-Frame keyframe anchoring with direct H3 video/audio latent continuation, aiming to preserve the visual quality and control of keyframe-based generation while maintaining motion and native-audio continuity across longer chains.

Highlights

• Direct H3 video + audio latent continuation
• Repeated FL2VA keyframe quality resets
• Automatic frozen-tail detection
• Phase-aligned latent handover
• Safe Tail Bridge for cleaner visual seams
• Seamless video stitching with short context-aligned crossfades
• 15 ms audio de-click crossfades
• Saved-chain stitching for separately generated clips
• Memory-conscious sequential decode and final encode
• Example workflows for Start, Continue, 3-Clip Showcase and Saved Chain Stitching

Example

h3-infinite-7-clip-example.mp4 is a 7-segment / ~61-second example generated with the included continuation workflow and stitched automatically with Safe Tail Bridge enabled. I've removed a few seconds of nonsense speach at the end using kdenlive.
No additional upscaling has been applied.
Settings:
736x1280 px
15 Steps
res_multistep

The example contains no manual editing at the clip boundaries. Very small single-frame brightness variations may still occasionally be visible.

Notes

This is an experimental community project and not an official MiniMax continuation mode. Testing across different hardware configurations and longer chains is still welcome.

See the repository README for installation, usage, prompting guidance, limitations and technical details.