Import MiniMax H3 Semantic Bridge workflow

2026-09-05 18:04:02 +00:00
parent 8a69f5208a
commit c31529af54
2 changed files with 1491 additions and 0 deletions
@@ -0,0 +1,49 @@
+# MiniMax H3 Semantic Bridge
+
+## Summary
+
+Experimental MiniMax H3 conditioning adapter plus example ComfyUI workflow. This is a semantic-control path, not a new base model, and it is aimed at better prompt adherence and consistency on the standard FL2VA route.
+
+## Model Family
+
+- `minimax`
+
+## Status
+
+- `imported-only`
+
+## Source
+
+- Model card: <https://huggingface.co/speach1sdef178/MiniMax-H3-Semantic-Bridge>
+- Workflow JSON: <https://huggingface.co/speach1sdef178/MiniMax-H3-Semantic-Bridge/resolve/main/examples/workflow_examples/minimaxH3_Semantic_bridge.json>
+- Workflow preview: <https://huggingface.co/speach1sdef178/MiniMax-H3-Semantic-Bridge/resolve/main/examples/workflow_examples/workflow_preview.png>
+- Original publisher: `speach1sdef178`
+- Date imported: `2026-09-05`
+
+## Developer notes
+
+- The repo describes this as a compact standalone adapter distilled from a teacher bridge, with the actual inference path staying inside standard MiniMax H3 conditioning.
+- The example workflow is explicitly a placement reference for ComfyUI, which makes it more useful than a thin model card drop.
+- The release says Ref2VA/reference-conditioned workflows are not supported in v1.
+
+## Our notes
+
+- This is the most interesting same-day MiniMax control release because it targets semantic structure rather than just motion.
+- It looks like a real prompt-adherence helper for FL2VA, but it is still MiniMax H3, so expect it to stay tight to heavy on an RTX 3070.
+- Good comparison point if you care about consistency, object relationships, or long prompt obedience more than raw style.
+
+## Required custom nodes
+
+- `MiniMax H3 Image to Video + Semantic Bridge`
+- `MiniMax H3 Semantic Bridge`
+- `MiniMax H3 Clear Semantic Bridge Cache`
+
+## Required models
+
+- `MiniMaxAI/MiniMax-H3`
+- `MiniMaxH3_SemanticBridge_v1.safetensors`
+
+## Notes
+
+- Do not treat this as a Ref2VA/reference-video fix.
+- Best used as a semantic-control add-on to an otherwise normal H3 FL2VA workflow.