FastVideo/FastH3-Omni-NVFP4-QAD

FastH3 Omni NVFP4 QAD

The Blackwell local-inference variant of FastH3 Omni. It generates video with synchronized audio from a prompt and ordered image, video and audio references in eight transformer evaluations. Quantization-aware distillation optimizes the model for NVFP4 execution.

Release collection · FastVideo local inference guide · BF16 release

Local AI with FastVideo

Use the device profiles for NVIDIA DGX Spark (spark), RTX 5090 (rtx-5090) or RTX PRO 6000 Blackwell (rtx-pro-6000). Follow the guide for installation, model composition, the AdaLN table, reference inputs and decoder selection. The RTX 5090 profile includes the memory and offload settings needed for a 32-GB device.

The standard Blackwell profiles retain higher precision for the text, reference and audio attention prefix. Keep those profile defaults for the release inference path.

Contents

Component	Purpose
transformer_ref/	NVFP4 reference-conditioned transformer
transformer/	Alternate directory containing the same transformer weights
text_encoder/	NVFP4 Qwen3-VL text and vision encoder
vae/, audio_vae/	Video and audio VAEs
vae_decoder_nvfp4/decoder-nvfp4.safetensors	36-layer quality video decoder
vae_decoder_nvfp4_fast/decoder-nvfp4.safetensors	8-layer fast video decoder
scheduler/, audio_scheduler/, processor/, tokenizer/	Sampling and input processing
fastvideo_inference.json	Eight-evaluation FastH3 Omni inference configuration

Use the 36-layer decoder for quality or the 8-layer decoder for lower decode cost. The local inference guide documents how to pass the decoder and prepare the modulation table.

NVFP4 kernels require Blackwell. For Hopper or for UniServe server deployment, use the BF16 checkpoint and the UniServe guide. For ComfyUI, use the NVFP4 QAD ComfyUI variant.

Quantized output can differ from BF16 output.

Local AI resources

Run FastH3 Omni on DGX Spark or RTX 5090 through FastVideo using the FastVideo-format NVFP4 QAD weights. Start with the FastVideo cookbook and FastH3 Omni recipe builder; use the Spark recipe or RTX 5090 recipe and the local inference guide for the AdaLN table, offload and decoder setup.

FP8 weights are a separate FastVideo variant; the FP8 recipe for RTX 4090 illustrates the non-FP4 path with layerwise offload and compatible BF16 encoder components. Use each runtime's matching checkpoint format.

Acknowledgements

FastH3 Omni was built in close collaboration with Nuva Lab, which co-developed the model with FastVideo, defined and curated the omni reference dataset, and brought production-grounded creative video workflows to FastH3. For dedicated production deployment, fine-tuning on your own data and a creative agent on top, see Nuva Lab's FastH3-Omni page.

Thanks also to the NVIDIA FastGen team.

License

Use of these weights is governed by the MiniMax-H3 Community License Agreement.