# MiniMax H3 Ref2Image Javawock Pack

## Summary

Same-day `MiniMax H3` reference-image workflow from the `javawock7618` pack. This is the most useful extra sibling in the pack if you want reusable subject and object control instead of another video variant.

## Model Family

- `minimax`

## Status

- `imported-only`

## Source

- Workflow pack page: <https://huggingface.co/javawock7618/comfy-MiniMax-H3-workflows>
- Workflow file: <https://huggingface.co/javawock7618/comfy-MiniMax-H3-workflows/raw/main/extra/MiniMax_int8-Ref2Image-javano2608.2.json>
- Original publisher: `javawock7618`
- Date imported: `2026-09-01`

## Developer notes

- The graph is built around reference-role prompting, subject definitions, and a MiniMax H3 image workflow with latent upscaling.
- It uses a large prompt-rewrite block, so the practical value is in the workflow structure rather than a tiny graph.
- This looks closer to a reusable reference-control helper than a flashy one-off image demo.

## Our notes

- Good fit for the current reference-image watch area.
- More interesting than a generic style LoRA because it is trying to preserve subject roles and reference structure.
- Still likely heavy on an `RTX 3070` with `8 GB` VRAM, so keep expectations realistic.

## Required custom nodes

- `comfyui-ollama`
- `ComfyUI-KJNodes`
- `LBH-123-AI/Comfyui_Minimax_h3_latent_Upscaler`

## Required models

- `minimax_h3_fl2va_pruned_int8_convrot.safetensors`
- `minimax_h3_video_vae_int8_convrot.safetensors`
- `minimax_h3_audio_vae_fp32.safetensors`
- `qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors`
- `H3/minimax_h3_ref_lora_rank_256_bf16.safetensors`
- `H3/minimax_h3_ref2v_lightx2v_turbo_4step_v0.1_resized_avg_rank_20_bf16.safetensors`
- `minimax_h3_latent_upscaler_3d_fp16.safetensors`

## Notes

- Best same-day MiniMax H3 import for reference-image control.