Everything you need is located in this repo. (Minus Comfys MiniMax VAEs): <a target="_blank" rel="ugc" href="https://huggingface.co/realrebelai/FastH3-V2_GGUFs">https://huggingface.co/realrebelai/FastH3-V2_GGUFs</a> Comfys Minimax VAEs: <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/tree/main/vae">https://huggingface.co/Comfy-Org/MiniMax-H3/tree/main/vae</a> GGUF quantizations of FastVideo FastH3 8-Step V2 for use with ComfyUI / ComfyUI-GGUF.These GGUFs were converted from the full official BF16 FastH3 V2 transformer weights released by FastVideo.<blockquote>These are NOT quantizations of the pruned ComfyUI checkpoint.</blockquote>FastH3 8-Step V2 is a distilled MiniMax-H3 checkpoint from FastVideo designed to generate synchronized video and audio using only 8 transformer forwards.The official V2 release uses:<ul><li>35B parameter MiniMax-H3 architecture</li><li>8-step / 8-forward generation</li><li>Data-Free DMD2 distillation</li><li>VSA-H3 sparse attention</li><li>80% attention sparsity</li><li>Synchronized video + audio generation</li></ul><h2 id="available-ggufs">Available GGUFs</h2><ul><li>QuantDescriptionQ3_K_M Lowest VRAM/RAM option. Maximum compression.</li><li>Q4_K_M Recommended balance between size and quality.</li><li>Q5_K_M Higher-quality quant with increased memory requirements.</li></ul>Additional quantizations may be added later.<h2 id="full-weight-conversion">Full-Weight Conversion</h2>The official FastH3 V2 transformer is distributed as a large sharded BF16 checkpoint.For these releases, the complete transformer was reconstructed and converted to the tensor layout expected by ComfyUI before GGUF quantization.The conversion preserves the FastH3 architecture, including:<ul><li>All 50 main transformer blocks</li><li>Token refiner blocks</li><li>Fused QKV projections</li><li>VSA <code>to_gate_compress</code> weights</li><li>SwiGLU MLP weights</li><li>AdaLN projections</li><li>Video projection layers</li><li>Audio projection layers</li><li>Conditioning projections</li><li>Time embeddings</li><li>RoPE data</li></ul>High-precision projection/conditioning tensors are preserved as F16/F32 where appropriate rather than indiscriminately quantizing every tensor.<h2 id="comfyui-installation">ComfyUI Installation</h2>ComfyUI-GGUF is required.Place the downloaded GGUF in:<pre><code>ComfyUI/
└── models/
 └── diffusion_models/
 └── FastH3-8-Step-V2-Q4_K_M.gguf
</code></pre>Restart ComfyUI or refresh your model list after installing the model.Load the model using your GGUF diffusion-model loader.<h2 id="recommended-generation-settings">Recommended Generation Settings</h2>FastH3 V2 is an 8-step model.Start with:<pre><code>Steps: 8
Sampler: Euler
Scheduler: Simple
Denoise: 1.0
</code></pre>The official FastVideo V2 checkpoint uses a trained video scheduler shift of 10.Exact workflow settings may depend on the current ComfyUI MiniMax-H3 implementation and any attention/VSA patches being used.<h2 id="required-supporting-models">Required Supporting Models</h2>The FastH3 V2 GGUF contains the transformer only.You will still need the supporting MiniMax-H3 models required by the ComfyUI FastH3 workflow:<ul><li>Qwen3-VL text encoder</li><li>MiniMax-H3 Video VAE</li><li>MiniMax-H3 Audio VAE for audio generation</li></ul><h3 id="gguf-text-encoder-wan-architecture-users">⚠️ GGUF Text Encoder / WAN Architecture Users</h3>If you are running the workflow through the WAN architecture implementation and want to use a GGUF Qwen3-VL text encoder, use the MiniMax-H3 GGUF encoders from my MiniMax-H3 repository:<a target="_blank" rel="ugc" href="https://huggingface.co/realrebelai/MiniMax-H3_GGUFs/tree/main">https://huggingface.co/realrebelai/MiniMax-H3_GGUFs/tree/main</a>Available GGUF encoders include:<ul><li><code>qwen3vl-32B-MiniMax-H3-Q2_K.gguf</code> — lower memory usage</li><li><code>qwen3vl-32B-MiniMax-H3-Q4_K_M.gguf</code> — recommended for better quality if you have enough RAM/VRAM</li></ul>Place the encoder in:<pre><code>ComfyUI/
└── models/
 └── text_encoders/
 └── qwen3vl-32B-MiniMax-H3-Q4_K_M.gguf
</code></pre>These are the MiniMax-H3-specific Qwen3-VL GGUF encoders intended for this architecture. If you are using the WAN architecture path with GGUF encoder loading, use one of these rather than a generic Qwen3-VL GGUF.<h3 id="vae-files">VAE Files</h3>You will also need the MiniMax-H3 Video VAE and Audio VAE.Official ComfyUI-compatible supporting files can be found here:<a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/tree/main/vae">https://huggingface.co/Comfy-Org/MiniMax-H3/tree/main/vae</a>Both VAEs are required if you want synchronized video + audio generation.<h2 id="q4_k_m">Q4_K_M</h2>Q4_K_M is the recommended starting point for most users.It significantly reduces the storage and memory requirements of the original ~35B BF16 transformer while retaining substantially more of the original model than lower-bit quantization.Users with more available system/GPU memory may prefer Q5_K_M, while Q3_K_M is intended for more memory-constrained systems.<h2 id="important:-vsa-attention">Important: VSA / Attention</h2>FastH3 V2 was trained using VSA-H3 sparse attention at 80% sparsity.Attention implementations, VSA patches, alternative attention backends, and post-processing/upscaling nodes can affect output quality and temporal consistency.When troubleshooting generation quality, test the raw FastH3 output before adding video restoration, interpolation, or upscaling stages.<h2 id="scope">Scope</h2>The official FastH3 V2 checkpoint supports text-to-audio-video generation.According to FastVideo, FL2VA and Ref2VA were not distilled for this checkpoint.Quality on difficult motion, very fine details, and some audio scenarios may remain below the original MiniMax-H3 base model.<h2 id="credits">Credits</h2>All model architecture, training, distillation, and original weights belong to their respective authors.FastH3 / FastVideo: <a target="_blank" rel="ugc" href="https://github.com/hao-ai-lab/FastVideo">https://github.com/hao-ai-lab/FastVideo</a>Original FastH3 V2 model: <a target="_blank" rel="ugc" href="https://huggingface.co/FastVideo/FastVideo-FastH3-8-Step-V2">https://huggingface.co/FastVideo/FastVideo-FastH3-8-Step-V2</a>Original base model: <a target="_blank" rel="ugc" href="https://huggingface.co/MiniMaxAI/MiniMax-H3">https://huggingface.co/MiniMaxAI/MiniMax-H3</a>GGUF conversions: Rebel AI / realrebelai<h2 id="license">License</h2>These quantizations inherit the licensing requirements and usage restrictions of the original FastH3 V2 / MiniMax-H3 model.Please review the original model license before using or redistributing these files:<a target="_blank" rel="ugc" href="https://huggingface.co/FastVideo/FastVideo-FastH3-8-Step-V2">https://huggingface.co/FastVideo/FastVideo-FastH3-8-Step-V2</a><hr />If these GGUFs helped you run FastH3 V2 on hardware that couldn't comfortably handle the original BF16 transformer, consider dropping a ❤️ on the repository.

Files with SHA256: 3d70ea49101a74da6989ed10ebd30bc875b410c4a92242b265cd07befcb30277