Instructions to use FastVideo/FastH3-Omni-Comfy with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusion Single File
How to use FastVideo/FastH3-Omni-Comfy with Diffusion Single File:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
FastH3 Omni (INT8 ConvRot) on AMD Strix Halo, Windows 11 + ROCm 10.1: setup, timings, gotchas
FastH3 Omni runs fine on AMD with these ComfyUI weights. Here is a full data point in case it helps other AMD users (and maybe a future ROCm row in the cookbook π).
Machine
- GMKtec EVO-X2: Ryzen AI Max+ 395, Radeon 8060S iGPU (gfx1151), 128 GB LPDDR5X unified memory; PyTorch sees about 110 GB as VRAM
- Native Windows 11 (no WSL), Python 3.12.11
- torch 2.13.0+rocm10.1.0, ComfyUI 0.42.4, comfy-kitchen 0.2.39
- Launch flags:
--disable-pinned-memory --use-split-cross-attention --bf16-vae(Dynamic VRAM on)
Models
- Diffusion:
fasth3_omniref_8step_pruned_int8_convrot.safetensors(this repo) - Text encoder:
qwen3vl_32b_minimax_h3_int8_convrot.safetensors(NVFP4 AWQ is Blackwell-only) - VAE:
minimax_h3_video_vae_fp16+minimax_h3_audio_vae_fp32 - Graph: native "MiniMax H3 Reference to Video" node,
MiniMaxH3SigmaShift12 / 3,ModelAttentionBackend= comfy kitchen attention,BlockSparseAttention(vsa, keep 10 %, start 0.2, min_tokens 12288),res_multistep/ simple, 8 steps, 24 fps
Timings (end to end, request β MP4 with audio)
Same task as Hao AI Lab's local comparison (1 image, 832Γ480, 124 frames, 8 steps)
- Warm, 3 seeds: 332.5 s mean (333.1 / 332.2 / 332.3 s), sampling 34.5 s/step (276 s)
- Cold (first run after model swap): 372 s
- For reference, from Hao AI Lab's post: 2Γ DGX Spark 130 s Β· 1Γ Spark 192 s Β· RTX 5090 62 s Β· RTX 3090 FP8 16.9 min Β· M5 Max 19.5 min
Other single runs (indicative, not averaged)
| Refs | Resolution Γ frames | Time |
|---|---|---|
| 1 image | 1344Γ768 Γ 124 (5 s) | 922 s |
| 2 images | 832Γ480 Γ 124 | 465 s |
| 1 voice (no image) | 832Γ480 Γ 124 | 321 s |
| 1 image + 1 reference video | 832Γ480 Γ 124 | 1162 s |
| 5 images, euler/simple | 1024Γ576 Γ 362 (15 s) | 2359 s (249 s/step) |
The iGPU's strength here is memory: the TE (β26 GB), the DiT (β21 GB) and the VAEs all stay resident, with no offload, and 10 s at 768p ran without OOM.
Gotchas on gfx1151 / ROCm 10.1 (Windows)
- SDPA Flash / Efficient kernels have no valid image for gfx1151 in torch 2.13+rocm10.1 ("device kernel image is invalid"). This only hits direct SDPA calls, i.e. the audio VAE encoding voice/video references. Fix: force the MATH backend (
torch.backends.cuda.enable_flash_sdp(False),enable_mem_efficient_sdp(False), plus ComfyUI'sSDPA_BACKEND_PRIORITY = [MATH]). With that, the voice-reference run went from crash to 321 s. - No Triton wheel for ROCm on Windows, so anything Triton-based is unavailable.
PYTORCH_*_ALLOC_CONF=expandable_segments:Truecauses OOM on Windows/HIP: leave it unset.- comfy kitchen attention is much faster than PyTorch attention here (about β60 % per step).
- Heavy CPU load on the same chip throttles the iGPU (down to about 1.4 GHz): keep the CPU idle while benchmarking.
- ComfyUI caches identical graphs: change the seed between benchmark runs.
Happy to run specific prompts or reference mixes on this box if useful. Next step on my side: trying FastVideo itself natively on ROCm.
Thanks a lot for sharing this. Did you try running this workflow with additional LoRa's like character replacement LoRa?