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 T2V
+
+## Summary
+
+Same-day `MiniMax H3` text-to-video workflow from `ewin-reg` with the newer FP8 / turbo packaging.
+
+## 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_t2v.json>
+- Original publisher: `ewin-reg`
+- Date imported: `2026-09-06`
+
+## Developer notes
+
+- This graph uses the native MiniMax-H3 ComfyUI node path and keeps the audio branch in the loop.
+- The source card pairs the pruned int8-convrot diffusion model with the Qwen3-VL text encoder and separate video/audio VAEs.
+- The workflow is built around the `17k+5` frame grid at 24 fps.
+
+## Our notes
+
+- Best as a baseline if you want the faster MiniMax H3 route without getting lost in the bigger community packs.
+- Still heavy for an `RTX 3070` with `8 GB` VRAM, but much more practical than the raw full-fat path.
+- Good comparison point against the older MiniMax H3 templates already in the repo.
+
+## Required custom nodes
+
+- Native ComfyUI MiniMax-H3 support
+
+## Required models
+
+- `Comfy-Org/MiniMax-H3` or the equivalent files referenced in the workflow
+- `minimax_h3_fl2va_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_fl2v_turbo_8step_v1.0_comfyui_bf16.safetensors` for the turbo path
+
+## Notes
+
+- Good same-day MiniMax H3 import if you want the leaner FP8/turbo setup.