Licon VBVR for MiniMax-H3

Licon VBVR for MiniMax-H3

An image-to-video LoRA for MiniMax-H3 that reduces duplicated objects, clipping and odd character motion in generated clips. It suits ComfyUI users who animate still images.

Made by LiconStudio on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Licon VBVR is a LoRA that teaches the MiniMax-H3 video model to animate still images with fewer glitches. Its author rebuilt the VBVR training data around H3 and reprocessed it for that model's training pipeline. The aim is steadier objects and more sensible character movement, with fewer unwanted copies of objects and less clipping or overlap during motion and interaction.

Who it suits: It suits ComfyUI users who animate still images with MiniMax-H3 and see objects multiply or limbs pass through things. It is an add-on to the base model, so you need an H3 image-to-video workflow already.

Title: Who might use it
Industry | Who uses it | How they use it | Example
AI video hobby | ComfyUI users | Add the LoRA to an H3 image-to-video workflow | Animate a still image of a character handling an object
Creative production | Video artists | Compare output with and without the LoRA | Run one prompt twice and inspect motion and overlaps
Workflow building | Workflow authors | Offer the LoRA as an optional step in a shared graph | Wire it into the LoRA loader of an H3 graph

If you do not run MiniMax-H3 image-to-video, you can skip it.

Where to use it: It works inside ComfyUI, through the LoRA loader of an H3 image-to-video workflow.

• ComfyUI, in the models folder for LoRA files
• Any other compatible MiniMax-H3 image-to-video workflow
• Windows, macOS or Linux with Python, to download the files

Title: Jobs it does
Use case | Who | What they get back
Fewer duplicated objects | Video creators | Clips with fewer unintended copies of objects
Fewer overlaps | Animators | Cleaner contact between characters, objects and surfaces
More coherent action | Workflow authors | Movement sequences that follow a clearer logic

How to set it up: You need ComfyUI, the MiniMax-H3 base model and an image-to-video workflow for it. The LoRA is one safetensors file that goes in the LoRA folder.

• ComfyUI with a MiniMax-H3 image-to-video workflow
• Python with pip, to fetch the files from Hugging Face

1. Download the repository, which holds the file named H3-VBVR-I2V.safetensors.
2. Place that file in the LoRA models folder of ComfyUI.
3. Open your H3 workflow and select the file in its LoRA loader.

pip install -U huggingface_hub
hf download LiconStudio/Minimax-H3-VBVR

ComfyUI/models/loras/

How to use it: Run your H3 image-to-video workflow as usual, with the LoRA selected in the loader. There is no extra command or setting on the card, so you judge the result by watching the clip.

Title: The pieces
Tool | What it does
H3-VBVR-I2V.safetensors | The LoRA file you download
LoRA loader | The ComfyUI node where you pick the file
ComfyUI/models/loras/ | The folder that holds LoRA files

Example: You feed a still image of a character holding an object into your H3 workflow with the LoRA on. The card says the clip should show fewer duplicated objects and less clipping than the plain model does. The author's page has side by side comparison videos.

For newcomers and experts: If you are new, a LoRA is a small add-on file that nudges a big model. Download the file, drop it in the LoRA folder and pick it in the loader of an H3 workflow. Then run the same image with and without it.

If you know the field, the training data comes from the VBVR-Pro-SFT-Video set, rebuilt for H3. The author wants steady movement and object contact to carry over into everyday video generation. The trade-off is that it only fits H3's image-to-video pipeline.

A still image enters the H3 workflow
The LoRA loader applies the VBVR file
MiniMax-H3 generates the video
The clip shows steadier objects

Good to know: The card names MiniMax as the base model's maker and the VBVR team as the source of the dataset.

• License: apache-2.0
• Cost to run: none named, though you need the MiniMax-H3 base model and a workflow for it
• Data it reaches: the files in your ComfyUI folders
• Not for: use on its own, or with workflows that are not MiniMax-H3 image-to-video

On Hugging Face, 13 November 2026

• Library: minimax-h3
• Downloads in the last 30 days: 13,559
• Likes: 170
• License: apache-2.0
• Based on: MiniMaxAI/MiniMax-H3
• Tags: minimax-h3

Links

• Model: https://huggingface.co/LiconStudio/Minimax-H3-VBVR
• Files: https://huggingface.co/LiconStudio/Minimax-H3-VBVR/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download LiconStudio/Minimax-H3-VBVR

Then follow the usage in its model card. Nothing in this kit runs by itself.