# MiniMax H3 INT8 Reference-to-Video

## Summary

`MiniMax H3` reference-to-video workflow with `INT8`, Motion Context, and a dedicated reference LoRA path.

## Model Family

- `minimax`

## Status

- `imported-only`

## Source

- Workflow repo: <https://huggingface.co/javawock7618/comfy-MiniMax-H3-workflows>
- Workflow file: <https://huggingface.co/javawock7618/comfy-MiniMax-H3-workflows/resolve/main/MiniMax_int8_R2V-javano2608.13.json?download=true>
- Original publisher: `javawock7618`
- Date imported: `2026-08-12`

## Developer notes

- This is the `R2V` branch from the source pack.
- The source README describes it as an `INT8` workflow built around:
  - `Sage-Attention`
  - `Spectrum`
  - `Lightx2v LoRA`
  - optional `Turbo-LoRA`
  - `Motion Context`
  - latent upscaling
- The source README says the `R2V` branch expects `minimax_h3_ref_lora_rank_256_bf16.safetensors`.
- The pack also ships `FLF` and `TTS` workflows, but this import is the dedicated reference-video graph.

## Our notes

- This is the strongest same-day MiniMax H3 workflow hit because it is actually about reference conditioning and consistency, not just another turbo LoRA.
- It is still a heavy ask for an `RTX 3070 / 8 GB`, but it is much more likely to teach you something useful than a novelty-only MiniMax drop.
- Best MiniMax workflow to test if you want a real comparison point against `LTX 2.3` consistency work.

## Required custom nodes

- `ComfyUI-Spectrum-MiniMax-H3`
- `ComfyUI-H3-Motion-Context`
- `ComfyUI-MiniMaxH3_LatentUpscaler`
- `ComfyUI-SolAttn_triton`
- `ComfyUI-GGUF` if you swap to GGUF-based variants
- `comfyui-easy-use`
- `comfyui-kjnodes`
- `comfyui-impact-pack`
- `comfyui-ollama`
- `comfyui-videohelpersuite`
- `WhatDreamsCost-ComfyUI`
- `comfyui_nvidia_rtx_nodes`
- `was-node-suite-comfyui`

## Required model files

- `Comfy-Org/MiniMax-H3`
- `minimax_h3_ref_lora_rank_256_bf16.safetensors` from `Kijai/MiniMax-H3-experimental`
- optional `Kijai/MiniMax-H3_comfy`
- optional `drbaph/MiniMax-H3-Turbo-Lora-ComfyUI`

## Notes

- This is the best current MiniMax reference-video import in the repo.
- If your goal is faster low-step novelty renders, the turbo paths are easier to demo, but this one is the better consistency-oriented test.