Hao AI Lab integrates MiniMax H3 Flex into FastVideo

Nov 4

1. 03:42

   UCSD Hao AI LabFirst-partySelected

   Hao AI Lab Integrates MiniMax H3 Flex into FastVideo, Single-GPU…

   Hao AI Lab announced that Pika Labs' open-source MiniMax H3 Flex has been integrated into FastVideo. The team claims its inference speed is up to 7 times faster than the official open-source inference code, generating a 5-second 480p video with audio in 5.8 seconds on a single NVIDIA GB200. Under the same prompt and random seed, the official code takes 59.7 seconds for a 5-second clip with audio, while FastVideo BF16 takes 26.6 seconds, NVFP4 QAD takes 16.8 seconds, and FP4 attention takes 15.2 seconds; the official code only supports a single GPU, whereas FastVideo adds multi-GPU support, making a 15-second clip 7 times faster on 4 GPUs. NVFP4 QAD uses BF16 Flex as its distillation target, coming 14-17% closer to BF16 than ordinary quantization; FP4 attention reduces a 768p 15-second clip from 101.8 seconds to 85.3 seconds. A 5-second 480p first-and-last-frame clip takes 59.5 seconds on a single DGX Spark, and it supports the Open-H3-Regenerate-2K model from Hao AI Lab and NVIDIA. The weights, inference code, and complete training recipe have all been released.

   Expand summary
2. 03:42

   UCSD Hao AI LabFirst-party

   Hao AI Lab integrates MiniMax H3 Flex into FastVideo, single-GPU…

   Hao AI Lab integrated Pika's open-source MiniMax H3 Flex into FastVideo. On a single GB200, generating a 5-second clip with audio takes 59.7 seconds with the official code, 26.6 seconds with FastVideo BF16, 16.8 seconds with NVFP4 QAD, and 15.2 seconds with FP4 attention. The official code only supports a single GPU, while FastVideo adds multi-GPU support, generating a 15-second clip 7 times faster on 4 GPUs. NVFP4 QAD, by distilling into BF16 Flex, comes 14-17% closer to BF16 than ordinary quantization; FP4 attention reduces a 768p 15-second clip from 101.8 seconds to 85.3 seconds. The project also supports running on DGX Spark, taking 59.5 seconds for a 5-second 480p first-and-last-frame clip, and supports the 2K model of Open-H3-Regenerate-2K, with open weights, inference code, and a complete training recipe.

   Expand summary
3. 03:42

   UCSD Hao AI LabFirst-party

   Hao AI Lab uses NVFP4 QAD quantization to bring video generation close to BF16 precision

   Hao AI Lab adopted NVFP4 QAD in FastVideo to perform distillation quantization on BF16 Flex; the result is closer to BF16 than ordinary quantization, narrowing the gap by 14-17%. Long video clips are limited by attention computation, and FP4 attention reduces the generation time of a 768p 15-second clip from 101.8 seconds to 85.3 seconds.

   Expand summary
4. 03:42

   UCSD Hao AI LabFirst-party

   Hao AI Lab integrates MiniMax H3 Flex into FastVideo, delivering up to 7x inference speedup

   Hao AI Lab integrated Pika Labs' open-source MiniMax H3 Flex into FastVideo, achieving inference speeds up to 7 times that of the official open-source inference code. On a single GB200, generating a 5-second 480p video with audio using the same prompt and random seed took 59.7 seconds with the official code, 26.6 seconds with FastVideo BF16, 16.8 seconds with NVFP4 QAD, and 15.2 seconds with FP4 attention. The official code only supports a single GPU; after FastVideo added multi-GPU support, generating a 15-second video on 4 GPUs was 7 times faster; generating a 5-second 480p first-and-last-frame video on a single DGX Spark took 59.5 seconds. The project also releases the weights, inference code, and complete training recipe, and supports the 2K model of Open-H3-Regenerate-2K.

   Expand summary