requirements.txt

# `diffusers` is installed from the canonical MiniMax-H3 pull request,
# https://github.com/huggingface/diffusers/pull/14371 ("Minimax h3 follow up (review & refactor)"), pinned to a
# **commit** rather than to its `minimax-h3-refactor` branch: the PR is a WIP and its head moves, and this Space's
# blocks subclass its block classes. Re-pin — and re-check `h3_split_blocks.py` against the block names of the new
# head — whenever the PR updates.
#
#   665f578278365ea4a3318cb8c9b66ce6c01204b9 = refs/pull/14371/head at the time of this deploy
--extra-index-url https://download.pytorch.org/whl/cu130
diffusers @ git+https://github.com/huggingface/diffusers.git@665f578278365ea4a3318cb8c9b66ce6c01204b9
torch==2.11.0
torchvision==0.26.0
# Ref2VA preserves each audio/video reference's native sample rate, then resamples it to the audio VAE rate.
torchaudio==2.11.0
# The Qwen3-VL processor decides the vision patch count, so a different minor changes the conditioning.
transformers==5.8.0
accelerate==1.14.0
# diffusers pins <2.
huggingface-hub==1.24.0
gradio[mcp]==6.20.0
spaces==0.51.3
# Blackwell-native NVFP4 GEMMs and the fused Q/K RMSNorm + split-half RoPE kernel used by h3_nvfp4.py.
# CUDA 13 is mandatory: older builds emulate this path and are slower than BF16.
comfy-kitchen==0.2.26
# NVIDIA Sol-Engine's released sparse-attention package, pinned to the H3 release. The Space selects its portable
# Triton backend because the separately released CuTe DSL/TVM-FFI packages currently have an incompatible ABI.
sol-attn @ git+https://github.com/NVlabs/Sana.git@46031940ba8af5d18054217e571149579424c0b1#subdirectory=techniques/sparse_backends
# No `kernels` pin on purpose: the Hub attention backends want `kernels>=0.12.3`, and that version breaks
# transformers 5.8.0 at import.
# PyAV muxes the generated soundtrack onto the frames (`encode_video`).
av
pillow
numpy
requests
safetensors>=0.8.0