Blame for README.md

---
tags:
- comfyui
- minimax-h3
- refcap
---

# H3 RefCap: fantasy cast

Ready-made RefCaps for [ComfyUI-RefCap](https://github.com/AEmotionStudio/ComfyUI-RefCap), so you can try MiniMax H3 reference-to-video with RefCaps before capturing your own: three characters, their voices and two locations.

## Use

1. Install ComfyUI-RefCap.
2. Put the `.safetensors` files in `ComfyUI/models/refcaps/` and refresh the browser.
3. Open the example workflow **H3 RefCap 16 - Fantasy cast** and queue. It renders the elf ranger and the dwarf warrior in the crystal cavern, both speaking in their own voices.

Each file loads in *H3 RefCap Select*; press **Pick views...** to browse its photos. A shot takes at most 9 pictures and 3 audio references.

## Files

| File | What it holds |
|---|---|
| `elf_ranger_front`, `dwarf_warrior_front`, `tiefling_sorceress_front` | 5 head views, -45 to 45 degrees |
| `elf_ranger_body`, `dwarf_warrior_body`, `tiefling_sorceress_body` | 6 full-body views: 0, ±45, ±90 and 180 degrees |
| `elf_ranger_turnaround`, `dwarf_warrior_turnaround`, `tiefling_sorceress_turnaround` | 24 head views, every 15 degrees |
| `elf_ranger_voice`, `dwarf_warrior_voice`, `tiefling_sorceress_voice` | One voice as an audio reference: `views` = `none`, `audios` = `1` |
| `fantasy_crystal_cavern_set`, `fantasy_wizard_study_set` | A location, 4 views: -90, 0, 90 and 180 degrees |

The characters and places are generated, not real people. The voices were designed with VoxCPM2.