Downloads · 30 days
0
minimax-h3-fl2va-lightning-4step
cokeadrink/minimax-h3-fl2va-lightning-4step
minimax-h3-fl2va-lightning-4step is a image-to-video model from cokeadrink. Use it for the image-to-video task on the model card, and read the license before you ship it in a product. It is set up for comfyui. The card lists the license as other.
The base diffusion checkpoint is unchanged pruned INT8 ConvRot FL2VA, without merged Turbo weights. The repository name does not force four-step sampling.
Access
Public
Updated Oct 28, 2026
Repo size
67.3 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors46.4 GB · 100%
At a glance
- Task
- Image-to-Video
- Library
- comfyui
- License
- other
- Access
- Public
- Created
- Sep 1, 2026
- Updated
- Oct 28, 2026
- SHA
- fdbb9c5b
Model card
From the Hugging Face model README
MiniMax H3 INT8: base or optional 8-step Turbo
The base diffusion checkpoint is unchanged pruned INT8 ConvRot FL2VA, without merged Turbo weights. The repository name does not force four-step sampling.
Base sampling and optional 8-step Turbo use the following configurations:
| Configuration | Steps | Sampler / scheduler | Video / audio shifts | LoRA |
|---|---|---|---|---|
| Base | 20 | res_multistep / simple | 12 / 3 | none |
| Turbo | 8 (fixed) | Euler / simple | 6 / 3 | existing 768p-trained 8-step, strength 1.0 |
Both configurations use the same audio VAE for native audio. Audio quality, speed, and use of the 768p-trained LoRA at 480p require validation.
Model files
diffusion_models/minimax_h3_fl2va_pruned_int8_convrot.safetensorstext_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensorsvae/minimax_h3_audio_vae_fp32.safetensorsvae/minimax_h3_video_vae_fp16.safetensors
Base weights total 42.47 GB (decimal). Turbo additionally requires:
loras/minimax_h3_fl2v_turbo_8step_v1.0_768p_comfyui_bf16.safetensors
The optional 8-step LoRA is 1.956 GB, bringing total required model storage to 44.43 GB when used. Base sampling does not require the optional LoRA file.
Provenance and license
The four base files match Comfy-Org/MiniMax-H3 at
e5eb578a89295337b8ff433a035929ce0279e0b6. The optional 8-step 768p LoRA matches
lightx2v/Minimax-h3-Turbo at 30f485ed874aa44c8183348ad361c5eabad318da:
SHA256 08cfe946033af7d27719b964b6e0a0e50c32138daabbd6ce4137e23df6bf9980.
See the author's model specifications at
https://github.com/ModelTC/Minimax-H3-Turbo#1-model-specs for this variant.
No weights are modified by this metadata update. See LICENSE and NOTICE
for the MiniMax H3 Community License Agreement and attribution.
- Task
- Image-to-Video
- Library
- comfyui
- License
- other
- Created
- Sep 1, 2026
- Updated
- Oct 28, 2026