MiniMax H3 Turbo Ref2V v0.1: 4-Step Reference-to-Video LoRA

ComfyUI Wikinews

Lightx2v and ModelTC release MiniMax H3 Turbo Ref2V v0.1, a 4-step reference-to-video LoRA with ComfyUI-ready bf16 weights for multi-reference video generation.

On August 13, the Lightx2v team and ModelTC released Ref2V Turbo v0.1, the reference-to-video counterpart of their MiniMax H3 Turbo LoRA line. Following the FL2V Turbo v1.0 release on August 11, the new checkpoint brings the same 4-step distillation to Ref2VA (reference-to-video + audio), letting a video be driven by one or more reference images instead of a single first frame.

Character reference from the Lightx2v Ref2V test assets

Character reference from the Lightx2v Ref2V test assets

What's in v0.1

ModelTasksTraining resolutionTraining shifts (video / audio)Distillation steps (NFE)Recommended inference steps
Ref2VA Turbo 4-step v0.1Ref2VA544p (mixed AR)12 / 344

The v0.1 release consists of two files on Hugging Face:

  • minimax_h3_ref2v_turbo_4step_v0.1_bf16.safetensors for Diffusers
  • minimax_h3_ref2v_turbo_4step_v0.1_comfyui_bf16.safetensors ready to drop into ComfyUI, no conversion needed

For the reference-image resizing policy, the project recommends the match mode used during distillation training, which matches the reference pixel area to the target canvas while preserving its aspect ratio and never upscaling a smaller reference. All three supported policies keep the reference aspect ratio and use the H3 resolution grid (dimensions rounded to multiples of 32) without cropping the reference content.

ComfyUI usage

MiniMax H3 has native ComfyUI support, so no custom node is required for the LoRA:

  1. Download minimax_h3_ref2v_turbo_4step_v0.1_comfyui_bf16.safetensors into ComfyUI/models/loras/.
  2. Load the base MiniMax H3 checkpoint in a Ref2V workflow and apply the LoRA between the model loader and the sampler.
  3. Connect one or more reference images through the reference-input branch and run 4 sampling steps.

The official MiniMax H3 R2V workflow template is a good starting point. The project also ships a dedicated example graph, video_minimax_h3_ref2v_lightx2v_turbo.json, which defaults to the Ref2VA Turbo 4-step v0.1 checkpoint at 960x544 (16:9, 0.5 MP).

Availability

Comments

Sign in with GitHub to join the discussion.

Loading comments…
MiniMax H3 Turbo Ref2V v0.1: 4-Step Reference-to-Video LoRA | ComfyUI Wiki