barelymining/ComfyUI-MiniMax-H3-FastVideo

FastH3 VSA-DataFree LoRA + Gate for ComfyUI (experimental)

Two files that let ComfyUI run MiniMax H3 video generation in 4 steps using FastVideo's Video Sparse Attention. On a 24 GB RTX 3090 Ti, this drops a 5-second, 0.8 MP clip from ~8 minutes (base, 20 steps) to under 1 minute (draft) or ~2.5 minutes (best quality). Requires the companion custom node — you can't load these with a normal LoRA loader alone.

Most people should probably use one of the established H3 turbo LoRAs (e.g. Comfy-Org's official 4-step turbo, lightx2v's 8-step turbo) — they're simpler to set up and don't need a custom node. This project exists for people who want to experiment with FastVideo's sparse-attention approach and see if the speed/quality tradeoff works better for their use case, or for people on older cards (RTX 3090 / 4090) who can't run FastVideo's latest official releases that target Hopper/Blackwell GPUs. Compare and pick what you like.

⚠️ Experimental. Tested on one machine (RTX 3090 Ti). Requires the companion custom node — this is not a drop-in for standard LoRA loading. See the node repo before downloading.

Files

• fasth3_vsa_v5.safetensors (2.05 GB) — the LoRA (backbone + AdaLN projected into the pruned base's compressed AdaLN space, with q/k/v fused into qkv_proj to match Comfy's H3 layout)
• fasth3_vsa_gate.safetensors (3.6 GB) — 50 to_gate_compress weight tensors extracted from FastVideo's VSA-DataFree checkpoint, loaded at runtime by the companion custom node

Why two files?

FastVideo's VSA training added a to_gate_compress layer to each transformer block. That layer doesn't exist in ComfyUI's stock H3 model. A LoRA can only modify existing layers — it can't add new ones. So the gate weights ship separately and are injected at runtime by the custom node.

Required companion custom node

You need ComfyUI-MiniMax-H3-FastVideo — a ComfyUI custom node that injects the gate layers and routes attention through FastVideo's Triton VSA kernel.

Installation, requirements, workflow, performance numbers, and known limitations are all documented there.

Base model

Designed for minimax_h3_fl2va_pruned_int8_convrot.safetensors from Comfy-Org. Other pruned INT8 FL2VA variants should also work if their adaln_t_table matches. Not tested with the non-pruned or non-INT8 variants.

Quick usage

1. Install the companion node and its vsa Python dependency
2. Place both .safetensors files here in ComfyUI/models/loras/
3. Load with a standard LoraLoader (for the _v5 file) + the MiniMax H3 VSA (FastVideo + gate) node (for the gate file)
4. 4 steps, CFG 1.0, topk_ratio 0.10 for drafts or 1.0 for finals

Provenance / conversion

Converted from FastVideo's VSA-DataFree checkpoint. Conversion involved:

• Renaming Diffusers-style keys to Comfy's H3 naming
• Fusing separate q/k/v LoRA branches into Comfy's qkv_proj (block-diagonal B, concatenated A, alpha = 3 × rank)
• Projecting adaln_proj.linear.lora_A from full BF16 space ([96768, 2688]) into the pruned model's 8-dim compressed space via least-squares fit against the base's adaln_t_table (residual ≈ 0 — mathematically lossless)
• Extracting to_gate_compress.set_weight tensors into a separate file for runtime injection

Attribution

• Original weights: FastVideo/FastVideo-FastH3-4-step-Preview-v1-VSA-DataFree
• Base model: MiniMaxAI/MiniMax-H3, pruned INT8 by Comfy-Org
• VSA kernel: hao-ai-lab/FastVideo
• Distributed under the MiniMax H3 Community License