Forward the caller's ZeroGPU token to the conditioner so one request bills as one request

Load the AoTI packages from the public multimodalart/minimax-h3-aoti, move the upsample toggle under the prompt

Migrate to canonical diffusers PR 14371 + public MiniMaxAI/MiniMax-H3 weights, add prompt upsampling toggle

Update README.md

Duration slider from 2s

Lower minimum duration to 2s

The keyframe half moved to multimodalart/minimax-h3

Mirror the conditioner's canonical table, grouped by ratio

Update README.md

Mirror the conditioner's 16-canvas table (fast 4:3, 3:4 and 21:9 added)

Point at the public multimodalart/qwen3vl-conditioner, called without a token

Point at the public multimodalart/qwen3vl-conditioner, called without a token

Fast canvas stays the default; the examples carry full-resolution canvases

Full 16:9 as the default canvas; match the conditioner's renamed fast label

Upload subject.png

A real person as the subject reference (Pexels, Oliver Dohrn)

A real person as the subject reference (Pexels, Oliver Dohrn)

Real speech for the audio example: LibriSpeech dev-clean 1462-170145-0022

Real speech for the audio example: LibriSpeech dev-clean 1462-170145-0022

A real subject for the reference example

Dynamic GPU duration from the packed sequence, cached examples, torchaudio note

torchaudio for off-rate reference soundtracks; title, fillable width, slot height

torchaudio for off-rate reference soundtracks; title, fillable width, slot height

Rename to MiniMax-H3 Reference; describe the tab order and image slots

Images/audio/video tab order, expandable image slots, native attention for the VAEs

Mirror the generator's title, theme and advanced-options layout

Describe the reference tabs

One tab per reference modality, in reading order

MiniMax-H3 ref2va, the denoising half of the split deployment

initial commit