MiniMax H3 ecosystem roundup: new ComfyUI nodes, distillation LoRAs, 12GB-VRAM workflows
optimisticalish · reddit · 2026-08-26
Community roundup of the week's MiniMax H3 tooling:
- ComfyUI v0.34.0 adds H3 support: anchoring image/audio guides at any frame (MiniMaxH3AddGuide), single-image Empty Latent for H3, per-token video/audio latent noise masks, prompt embeddings (e.g. minimaxh3artisexplosion at 00:03.500), and a fix for the <d> dialogue tag
- Keyframe-Offset node: place first/last keyframes at arbitrary frames; new audio node lets a prompt like 'echo reverberating through the valley' plus a stony-valley reference image generate spatially matching sound
- A native ComfyUI MODEL patch for the official Fun-ControlNet-Union adapter
- WAN team ships for its competitor: Parallel Decoding Distillation LoRAs for H3, predicting multiple denoising steps per call for few-step generation (not yet Comfy-fied)
- Character sheet generator: one orbiting-camera pass, character static, grab 6 frames and stitch a reusable sheet; anime-to-real demos included
- Seamless looping GIF workflow (LoopGif nodes fix seam glitches); Mark DK Berry's RTX 3060 12GB workflows: optimized turbo LoRA, 2-pass latent upscaler (better mid-distance faces), pixel-space upscaler
Related event: MiniMax H3 Ecosystem Weekly: New ComfyUI Nodes and Speed LoRAs(2 posts)→
More from coding & agent
- LiveKit builds patient intake agent end-to-end on Grok voice models with ZDR — SpaceXAI · 2026-08-27
- Inside Rippling's four-layer agent eval pipeline: ~10 critical scenarios gate every deploy — LangChain · 2026-08-27
- How Rippling went AI-native in 6 months with Deep Agents and a 3-layer eval pipeline — LangChain · 2026-08-27
- Vercel engineer's 5-step quality workflow for vibe coders — brandon_galang · 2026-08-27
- Runable raises $21M to launch Grow, an agent for end-to-end GTM workflows — SimplyAnnisa · 2026-08-27
- Ox Alpha generates Fruit Ninja in 48 mins, beating rivals in quality — CodeByPoonam · 2026-08-27