larryvrh/MiniMax-H3-Turbo-Lora

MiniMax-H3 Turbo LoRA — 4-step audio-video generation (early preview)

A LoRA for MiniMax-H3 that renders joint video + synchronized stereo audio in 4 sampling steps instead of the usual ~20 — roughly a 5× speedup in sampling wall-clock.

⚠️ Early prototype — the weights and the tooling. These weights are an early, under-trained checkpoint (not production quality), and the ComfyUI nodes below are prototype code: functionality and compatibility are not guaranteed. It is already a clear step up from the base model at 4 steps — sharper detail, cleaner and better-synced audio — but treat everything here as a work-in-progress taste, not a finished product. If something breaks, please open an issue on the node repo.

Use it in ComfyUI (recommended)

Custom nodes: Larryvrh/ComfyUI-MiniMax-H3-Turbo — or search "MiniMax-H3 Turbo" in ComfyUI-Manager.

1. Install the nodes (Manager, or git clone into ComfyUI/custom_nodes).
2. Download a .safetensors from this repo into ComfyUI/models/loras/.
3. Start from the official MiniMax-H3 workflow (text-to-video or image-to-video) and make two changes:
   • insert MiniMax-H3 Turbo LoRA between the model loader and the sampler;
   • replace the sampler feeding SamplerCustomAdvanced with MiniMax-H3 Turbo Sampler (4-step), and set the scheduler to 4 steps (simple).

Everything else stays as in the official workflow, so both t2v and i2v work. A ready-made t2v workflow is included here (minimax_h3_t2v_turbo.json) and in the node repo — drag it into ComfyUI. The custom sampler is required: MiniMax-H3 runs video and audio on two different flow schedules, and a stock sampler over-steps the audio at 4 steps and it breaks.

• Steps: 4 works, but at this early checkpoint the model is under-trained, so the comfort zone for sharpness is 6–8 steps, not 4 — 4 comes out noticeably softer. Any count ≥ 4 is valid and more steps look better. Keep the scheduler on simple.
• Base model: use a non-pruned base (the bf16 or full int8_convrot DiT). The pruned variants (pruned_int8, pruned_fp8) use a different time-conditioning layer and are not compatible with this LoRA.

Weights

All bf16, ~744 MB, applied as a standard low-rank update (W_eff = W + lora_B @ lora_A, alpha = rank so no extra scaling):

file	~steps	notes
minimax_h3_turbo_4step_ckpt500.safetensors	~500	recommended default — newest, sharpest (non-EMA)
minimax_h3_turbo_4step_ema_ckpt500.safetensors	~500	time-averaged variant — smoother, but at this early checkpoint it can show heavy ghosting / motion smear; prefer the non-EMA file above
minimax_h3_turbo_4step.safetensors	~200	initial release, non-EMA
minimax_h3_turbo_4step_ema.safetensors	~200	initial release, time-averaged (superseded)

Training is ongoing; newer checkpoints will land here over time.

Standalone (no ComfyUI graph)

generate.py is a single self-contained file — loads the base DiT + a LoRA, encodes the prompt, runs the 4-step dual-schedule sampler, decodes and muxes an mp4. It still needs a ComfyUI checkout for the H3 model / VAE / text-encoder definitions:

# ComfyUI (pinned to the commit these weights were validated against)
git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI && git checkout 14b05228cef127ce529bc0c08660770d4af3e9a8
pip install -r requirements.txt && cd ..
pip install -r requirements.txt   # this repo: torch, safetensors, imageio-ffmpeg

# base weights from Comfy-Org/MiniMax-H3 into a models/ tree, then:
python generate.py \
  --comfyui ./ComfyUI \
  --base   models/diffusion_models/minimax_h3_fl2va_bf16.safetensors \
  --lora   minimax_h3_turbo_4step_ckpt500.safetensors \
  --te     models/text_encoders/qwen3vl_32b_minimax_h3_int8_convrot.safetensors \
  --video-vae models/vae/minimax_h3_video_vae_fp16.safetensors \
  --audio-vae models/vae/minimax_h3_audio_vae_fp32.safetensors \
  --prompt "A corgi in a chef hat flipping a pancake, sizzling sounds and a cheerful bark." \
  --width 1344 --height 768 --frames 124 --out corgi.mp4

Notes

• Resolution / duration: width/height are multiples of 32; the short edge is typically 768. Frame count is at 24 fps and snaps to the model's 17·k+5 grid (124 ≈ 5 s). Validated range 124–362 frames (5–15 s).
• VRAM: the base model is large (~33 B); an 80 GB GPU is comfortable (--offload-adaln in the standalone script trades ~13 GB of VRAM for CPU RAM).
• Audio: 32 kHz stereo, aligned to the video; the two streams ride different flow schedules and are integrated each on its own clock.