Import same-day MiniMax H3 workflow bundles

2026-08-30 18:07:57 +00:00
parent 730858abb0
commit 237f6a18d4
4 changed files with 2025 additions and 0 deletions
@@ -0,0 +1,47 @@
+# MiniMax H3 Turbo LoRA ComfyUI Bundle
+
+## Summary
+
+Same-day `MiniMax H3` turbo LoRA bundle with a current recommended ComfyUI workflow. This is the most practical audio / speed helper from today because it packages the step-reduction path instead of just talking about it.
+
+## Model Family
+
+- `minimax`
+
+## Status
+
+- `imported-only`
+
+## Source
+
+- Workflow pack page: <https://huggingface.co/Halpak/MiniMax-H3-Turbo-Lora-ComfyUI>
+- Workflow file: <https://huggingface.co/Halpak/MiniMax-H3-Turbo-Lora-ComfyUI/raw/main/fl_minimax_h3_turbo_lora_example_workflow.json>
+- Original publisher: `Halpak`
+- Date imported: `2026-08-30`
+
+## Developer notes
+
+- The source README frames this as a ComfyUI-friendly set of MiniMax-H3 turbo LoRAs.
+- The bundle includes dynamic-rank LightX2V FL2V / Ref2V LoRAs plus pruned and curve-form compatibility conversions.
+- The current recommended workflow uses the `v4 step-600 EMA` pruned LoRA and calls out `6-8` steps with `audio sigma shift` around `4-6`.
+
+## Our notes
+
+- Good practical pick if you want a faster MiniMax H3 audio/video path without jumping straight to a heavyweight full model.
+- This looks like a real speed helper, not just a cosmetic LoRA dump.
+- On an `RTX 3070`, this is still a tight local test, but it is much more plausible than the full multimodal MiniMax H3 pack.
+
+## Required custom nodes
+
+- Not listed explicitly in the source README.
+
+## Required models
+
+- `Comfy-Org/MiniMax-H3`
+- `minimax_h3_turbo_v4_step600_ema_pruned_comfyui.safetensors`
+- Optional if you want the LightX2V variants:
+  - `lightx2v/Minimax-h3-Turbo`
+
+## Notes
+
+- Best same-day MiniMax H3 turbo/helper import in the repo for audio-sensitive low-step work.
@@ -0,0 +1,51 @@
+# MiniMax H3 I2V Javawock Pack
+
+## Summary
+
+Same-day `MiniMax H3` first/last-frame image-to-video workflow from the `javawock7618` pack. This is the missing local sibling from the pack and the cleanest new add if you want the I2V lane.
+
+## 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/MiniMax_int8_I2V-javano2608.17.2.json>
+- Original publisher: `javawock7618`
+- Date imported: `2026-08-30`
+
+## Developer notes
+
+- The source README describes this as the `First-Last-Frame` workflow.
+- It uses the same INT8 / Sol-Attn / SLA-Attention / Spectrum stack as the rest of the pack.
+- The I2V graph integrates native `Extend` and the MiniMax-H3 latent upscaler for faster high-resolution generation.
+
+## Our notes
+
+- Good practical add for a MiniMax H3 local workflow set.
+- Not as directly interesting as the face-refine or reference-retention lanes, but still useful if you want a first-frame path with the pack's newer stack.
+- Still a tight fit on an `RTX 3070` with `8 GB` VRAM, so treat it as a serious test target rather than a casual throwaway graph.
+
+## Required custom nodes
+
+- `ComfyUI-SolAttn_triton`
+- `ComfyUI-Spectrum-MiniMax-H3`
+- `ComfyUI-H3-SLA-Attention`
+- `ComfyUI MiniMax-H3 Latent Upscaler`
+- `ComfyUI-Switchboard`
+
+## Required models
+
+- `Comfy-Org/MiniMax-H3`
+- `lightx2v/Minimax-h3-Turbo` if acceleration is enabled
+- `drbaph/MiniMax-H3-Turbo-Lora-ComfyUI` if the optional turbo LoRA path is enabled
+- `LBH-123-AI/Minimax_h3_latent_Upscaler` if latent upscaling is enabled
+
+## Notes
+
+- Best same-day MiniMax H3 I2V import in this repo so far.