multimodalart/minimax-h3

Update README.md

Update README.md

Add a 1024x1024 1:1 canvas

Update app.py

Update requirements.txt

Delete __pycache__

int8 attention for the compiled block stack (#11)

Update README.md

Point the denoiser at the pruned bf16 DiT (#10)

Increase Gradio max_threads (#9)

A/B confirmed: the guard's in-process forward pass kills subsequent GPU worker forks — restore the subprocess guard

A/B test: in-process guard again — was the No CUDA worker failure really the guard?

Guard subprocess via plain Popen — mp.spawn re-imports app.py as main

Run the NCII guard from its spawned subprocess — in-process it poisons the ZeroGPU worker fork

Add ncii_guard subprocess module

Log the refused prompt in the guard message

Run the NCII guard only on requests with a keyframe

Add NCII prompt guard (hfmlsoc/ncii-light-guard-v01) before any GPU booking

Open badge links in a new tab

Price bookings against the AoTI blocks, generate from 2 s again, cross-link the demos

Start the duration slider at the 5 s the model actually generates

Restore encode_remote

Cache the conditioner client with functools.cache, tidy comments

Let gradio_client forward the caller's ZeroGPU token itself instead of pinning one by hand

Forward the caller's ZeroGPU token to the conditioner, plainly

Bill the conditioner call to the requesting user's own ZeroGPU token, never an org token

Pay for the conditioner call with the first ZeroGPU identity that can: the caller's token, this Space's HF_TOKEN, then none

Fall back to an anonymous conditioner call when the forwarded ZeroGPU token is refused

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

Repair and reorder the canvas table

Fast canvases for 4:3, 3:4 and 21:9

Use the public conditioner without a token

Fast default for generation, full quality examples

Default to full resolution, full quality examples

Idiomatic ZeroGPU startup packing for the transformer

Right size placement allowance

Dynamic GPU duration fitted from probes

Update app.py

Update app.py

Demo short description

Rename to MiniMax H3

Tighten description

Lazy cached examples, generate defaults

Model description and examples

Upload examples/last.png with huggingface_hub