Add same-day MiniMax H3 FP8 workflows

2026-09-06 18:05:20 +00:00
parent c31529af54
commit 493f436778
6 changed files with 6411 additions and 0 deletions
@@ -0,0 +1,49 @@
+# MiniMax H3 Turbo FP8 R2V
+
+## Summary
+
+Same-day `MiniMax H3` reference-to-video workflow from `ewin-reg`, with explicit support for images, videos, and audio references.
+
+## Model Family
+
+- `minimax`
+
+## Status
+
+- `imported-only`
+
+## Source
+
+- Model repo: <https://huggingface.co/ewin-reg/MiniMax-H3-Turbo-FP8-ComfyUI>
+- Workflow file: <https://huggingface.co/ewin-reg/MiniMax-H3-Turbo-FP8-ComfyUI/raw/main/workflows/video_minimax_h3_r2v.json>
+- Original publisher: `ewin-reg`
+- Date imported: `2026-09-06`
+
+## Developer notes
+
+- This graph runs `MiniMaxH3ReferenceToVideo`.
+- It accepts up to 9 reference images, 3 reference videos, and 3 standalone audio references.
+- The prompt tags the inputs with `<Picture 1>`, `<Video 1>`, and `<Audio 1>` style references.
+- The sampler setup is tuned for reference-heavy prompts.
+
+## Our notes
+
+- Most interesting of the three for identity, voice, and scene continuity work.
+- Best choice if the goal is a character pack, a reference-sheet workflow, or mixed media control.
+- Still very heavy for an `RTX 3070` with `8 GB` VRAM, so treat it as a serious test target.
+
+## Required custom nodes
+
+- Native ComfyUI MiniMax-H3 support
+
+## Required models
+
+- `minimax_h3_ref2va_pruned_int8_convrot.safetensors`
+- `qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors`
+- `minimax_h3_video_vae_fp16.safetensors`
+- `minimax_h3_audio_vae_fp32.safetensors`
+- `minimax_h3_ref2v_turbo_4step_v0.1_comfyui_bf16.safetensors` for the turbo path
+
+## Notes
+
+- Best same-day MiniMax H3 import if you care about reference locking and multi-input control.