Public sources read 2026-11-04
MiniMax H3 VBVR: LoRA download & ComfyUI guide
MiniMax H3 VBVR is LiconStudio’s image-to-video adapter. Find its file and loading instructions, with a separate Hailuo clip generator here.
This generator uses Hailuo 02 Standard: 6 seconds for 60 credits. Nano AI does not load the VBVR LoRA or run MiniMax H3 weights.
Duration
Ratio
Generated Videos
No videos yet
Fill out the form and click Generate Video to create your first AI video.
Tips: Use detailed descriptions for better results
- Made by
- LiconStudio
- Run on Nano AI
- Hailuo 02 Standard · from 60 credits per video
- Page updated
- November 4, 2026
Source: huggingface.co
Published example
A static frame at 4 seconds from LiconStudio’s published Example 02 VBVR.mp4, reviewed 2026-11-04. It shows the kitchen validation scene; it is not a Hailuo or Nano AI output. One frame cannot verify motion consistency.
MiniMax H3 VBVR · Scope and limits
LiconStudio’s model card, read 2026-11-04, describes Licon VBVR as an image-to-video LoRA trained on VBVR-Pro-SFT-Video data reprocessed for H3. The author targets object consistency, fewer duplicated objects and less clipping, plus more coherent action order. These are author claims, not Nano AI measurements.
The card supplies a download and a compatible H3 image-to-video loading route. It does not specify a universal LoRA strength, step count, VRAM requirement or improvement percentage. This page adds no such settings or scores.
Scope and limits
- Model here
- Hailuo 02 Standard
- Credits here
- 60 / 6s
- Route
- Text-to-video clip
Download and load the adapter
Public sources read 2026-11-04
- 1
Download the file
Download H3-VBVR-I2V.safetensors from LiconStudio’s model card. This is an adapter, not the H3 base checkpoint.
- 2
Put it in the LoRA folder
For ComfyUI, the card names ComfyUI/models/loras/ as the destination.
- 3
Select it in your H3 workflow
Use a compatible MiniMax-H3 image-to-video workflow and select the file in that workflow’s LoRA loader. The card does not supply a full workflow or sampler recipe.
File and location
The card supplies a download and a compatible H3 image-to-video loading route. It does not specify a universal LoRA strength, step count, VRAM requirement or improvement percentage. This page adds no such settings or scores.
| File | Location | Purpose |
|---|---|---|
| H3-VBVR-I2V.safetensors | ComfyUI/models/loras/ | Local H3 image-to-video + VBVR LoRA |
Before you start
What does this page run?
This generator uses Hailuo 02 Standard: 6 seconds for 60 credits. Nano AI does not load the VBVR LoRA or run MiniMax H3 weights.
Is this free?
The video costs 60 Nano AI credits. Existing or granted credits can pay only within your available balance; free generation is not guaranteed.
Can I use an API here?
The model card provides a downloadable adapter, not an API served by Nano AI. For local use, follow the card with a compatible H3 image-to-video workflow.
Compare the routes
Public sources read 2026-11-04
| MiniMax H3 VBVR | Nano AI | |
|---|---|---|
| Route | Local H3 image-to-video + VBVR LoRA | Independent browser generator |
| Model | MiniMax-H3 + VBVR LoRA | Hailuo 02 Standard |
| Cost | Local compute; not priced by Nano AI | 60 / 6s |
Primary sources
Public sources read 2026-11-04
Use the generator
- 1
Choose a scene
Choose one subject and one simple motion. These original prompts are for the actual model shown here; they do not reproduce the published sample.
- 2
Check the model and cost
This generator uses Hailuo 02 Standard: 6 seconds for 60 credits. Nano AI does not load the VBVR LoRA or run MiniMax H3 weights.
- 3
Generate and review
Sign in, review the displayed credit cost, then generate and inspect the result before downloading.
Original scene prompts
Choose one subject and one simple motion. These original prompts are for the actual model shown here; they do not reproduce the published sample.
Paper boat
A red paper boat floats through a shallow stone basin, one continuous shot, slow camera push, soft morning light, no people, no text, no logos.
Marble
A single glass marble rolls along a wooden toy track, one continuous tracking shot, natural daylight, no people, no text, no logos.
Steam
A ceramic cup stands on a wooden table as steam rises, static camera, warm side light, no people, no text, no logos.
Lantern
A paper lantern sways above an empty courtyard, slow camera arc, blue hour, no people, no text, no logos.
Before troubleshooting
- No adapter in the menu
- Check the exact folder and file name against the model card; this website does not install or configure ComfyUI.
- No measured score
- The card supplies a download and a compatible H3 image-to-video loading route. It does not specify a universal LoRA strength, step count, VRAM requirement or improvement percentage. This page adds no such settings or scores.
FAQ
MiniMax H3 VBVR · Questions and answers
Public sources read 2026-11-04
What does this page run?
This generator uses Hailuo 02 Standard: 6 seconds for 60 credits. Nano AI does not load the VBVR LoRA or run MiniMax H3 weights.
Is this free?
The video costs 60 Nano AI credits. Existing or granted credits can pay only within your available balance; free generation is not guaranteed.
Where is the official product?
LiconStudio’s model card, read 2026-11-04, describes Licon VBVR as an image-to-video LoRA trained on VBVR-Pro-SFT-Video data reprocessed for H3. The author targets object consistency, fewer duplicated objects and less clipping, plus more coherent action order. These are author claims, not Nano AI measurements.
What does the sample prove?
A static frame at 4 seconds from LiconStudio’s published Example 02 VBVR.mp4, reviewed 2026-11-04. It shows the kitchen validation scene; it is not a Hailuo or Nano AI output. One frame cannot verify motion consistency.
Can I use an API here?
The model card provides a downloadable adapter, not an API served by Nano AI. For local use, follow the card with a compatible H3 image-to-video workflow.
More from Nano AI
Related guides
Nano AI is independent and not affiliated with LiconStudio or MiniMax. The published validation sample belongs to LiconStudio.