@haoailab

Hao AI Lab at UCSD. Our mission is to democratize large machine learning models, algorithms, and their underlying systems.

haoailab.com/
Joined March 2024
  • Tweets734
  • Following379
  • Followers8.8K
272Media
As promised, FastVideo open-sources @MiniMax_AI FastH3 Omni, training code included! 🚀 - Real-time gen @ 12x faster on NVIDIA Blackwell - NVFP4/FP8 for Spark + RTX! - ComfyUI day-0 Collab with @NVIDIAAI , Nuva Lab @nuvalab and Reactor @reactorworld Open source must win! ♥️
32
68
45
600
59,519
(1/7) FastVideo FastH3 Omni goes local on DGX Spark, RTX GPUs, and Apple MLX 🚀 - FastH3 Omni, a 5 s 480p clip from a photo and a prompt takes ~3 min. On two Sparks, ~2 min - FastH3 T2AV is 52.6 s on one Spark, 30.7 s on two Sparks Blog: haoailab.com/blogs/fasth3-om… Cookbook: haoailab.com/FastVideo/cookb… HF 🤗 huggingface.co/collections/F… @NVIDIARTXSpark @NVIDIAAI @MiniMax_AI Every clip in this video was generated on NVIDIA DGX Spark. 8-step NVFP4 checkpoint.
4
16
7
65
7,035
more replies
(7/7) Get started in the FastVideo cookbook, with recipes for DGX Spark, 2× Spark, RTX 5090, 24 GB RTX and Apple Silicon. Blog: haoailab.com/blogs/fasth3-om… Cookbook: haoailab.com/FastVideo/cookb… Models: huggingface.co/collections/F…

FastH3 Omni Goes Local: DGX Spark, RTX and Apple Silicon

The same FastH3 Omni request, with matched prompts, references and seeds, run end to end on DGX Spark, RTX GPUs and an Apple M5 Max.

haoailab.com
1
4
409
We thank - @posi_posi8 for the M5 Max runs ♥️ - @nuvalab For the amazing work on data curation, training eval and inference optimizations - @reactorworld For providing the compute and early feedback for production API serving of FastH3 Omni! - NVIDIA FastGen (Julius Berner @julberner, Chao Liu @cxlclb, Arash Vahdat @ArashVahdat) - NVIDIA Enterprise Products (Pengcheng Li @landy0220, Cliff Woolley) for the collaboration on FastH3! 🚀 Special thank to @NVIDIAAI, @mbzuai, and @vllm_project for the continued support of FastVideo!
1
8
807
Hao AI Lab retweeted
Yi Sun@yisun0618Oct 20
UniServe supports FastH3 Omni on Day 0! FastH3 Omni produces high-quality output with fast generation, and it reaches realtime speed in some settings. Our team had a lot of fun building with it internally. github.com/hao-ai-lab/uniser…

GitHub - hao-ai-lab/uniserve: Preview release of a unified serving engine, built for the fastest...

Preview release of a unified serving engine, built for the fastest FastH3 serving. - hao-ai-lab/uniserve

github.com
Hao AI Lab
@haoailabOct 20
As promised, FastVideo open-sources @MiniMax_AI FastH3 Omni, training code included! 🚀 - Real-time gen @ 12x faster on NVIDIA Blackwell - NVFP4/FP8 for Spark + RTX! - ComfyUI day-0 Collab with @NVIDIAAI , Nuva Lab @nuvalab and Reactor @reactorworld Open source must win! ♥️
6
1
15
2,413
Check out FastVideo FastH3 Omni on ComfyUI!
ComfyUI
@ComfyUIOct 20
FastH3 OmniRef by @haoailab's is now supported in ComfyUI. This distilled version of Minimax H3 allows multi-modal reference generation of videos. It runs locally on your hardware and provide great balance between quality and speed. Click the links below for open weights and cloud ⬇️
2
1
8
1,537
Hao AI Lab retweeted
Will Lin
@wlsaidhiOct 20
We keep our promises :)
Hao AI Lab
@haoailabOct 20
As promised, FastVideo open-sources @MiniMax_AI FastH3 Omni, training code included! 🚀 - Real-time gen @ 12x faster on NVIDIA Blackwell - NVFP4/FP8 for Spark + RTX! - ComfyUI day-0 Collab with @NVIDIAAI , Nuva Lab @nuvalab and Reactor @reactorworld Open source must win! ♥️
7
6
2
48
2,395
Hao AI Lab retweeted
Nuva Lab
@nuvalabOct 20
What If malfoid went to high school in Japan #Malfoid 🧵 check comment for how it's made👇
2
2
1
16
5,350
6/7 Serve it: one image + one voice on 4x— GB200, then 8 GPUs across two hosts for bigger reference mixes. Pinned checkpoints and configs. Star the repo, try it in ComfyUI, and tell us what you make. Blog: haoailab.com/blogs/fastvideo… Model: huggingface.co/collections/F… Uniserve: github.com/hao-ai-lab/uniser… FastVideo Repo: github.com/hao-ai-lab/FastVi…

FastVideo FastH3 Omni: Open Source Real-Time MiniMax H3 Ref2VA Video Generation on NVIDIA Blackwell...

FastVideo is open-sourcing FastH3 Omni, a distilled MiniMax H3 model for reference-guided video and audio. Give it a prompt plus images, video clips and a voice reference to generate a finished video...

haoailab.com
1
18
2,732
7/7 We thank -@nuvalab For the amazing work on data curation, training eval and inference optimizations - @reactorworld For providing the compute and early feedback for production API serving of FastH3 Omni! - NVIDIA FastGen (Julius Berner @julberner, Chao Liu @cxlclb, Arash Vahdat @ArashVahdat ) - NVIDIA Enterprise Products (Pengcheng Li @landy0220, Cliff Woolley) for the collaboration on FastH3! 🚀 Special thank to @NVIDIAAI, @mbzuai, and @vllm_project for the continued support of FastVideo!
12
2,205
(1/6) The all-new FastVideo FastH3-QAD-v2 and the all-new H3 NVFP4 VAE decoders are here! 🚀🚀 This entire 768p video was generated on two DGX Sparks, using QAD-v2 and both new decoders. 8-step video + audio, QAD weights 💾 @NVIDIARTXSpark @NVIDIAAI @MiniMax_AI
5
17
8
103
17,809
more replies
(6/6) Tomorrow: FastH3 OmniRef. The general model and training code, plus a consumer-hardware release that uses this QAD recipe and the NVFP4 decoders. Blog:haoailab.com/blogs/fasth3-v2… Model :huggingface.co/FastVideo/Fas… Decoders: huggingface.co/FastVideo/Fas…

FastH3-QAD-v2 and NVFP4 VAE Decoders: Better Quality, Faster Decoding, and Community LoRAs

FastH3 V2 runs an eight-step video and audio model in NVFP4 on consumer GPUs. FastH3-QAD-v2 raises quality without changing the architecture, memory footprint or file layout by training the 4-bit...

haoailab.com
1
1
7
990
We thank - NVIDIA FastGen (Julius Berner @julberner, Chao Liu) - NVIDIA Enterprise Products (Pengcheng Li @landy0220, Cliff Woolley) - @nuvalab for help with data curation, evaluation and inference optimizations Special thanks to @NVIDIAAI, @mbzuai and @vllm_project for their continued support of FastVideo! ♥️ See you tomorrow!
6
727
Load more