🧩 MiniMax-H3 Workflows showcase

An interactive reference for javawock7618/comfy-MiniMax-H3-workflows — optimized INT8 ComfyUI workflows for MiniMax-H3: FLF image-to-video, reference-to-video, Bridge transitions, Face Refine, TTS/Video-to-Audio, plus Music 3 & Ref2Image extras.

📦 Source workflows 🤖 Base model ⚡ Spectrum 🎭 Face Refine 🪜 Latent Upscaler 🔀 Switchboard

What is this collection?

Seven drop-in ComfyUI workflows built around the INT8 quantized MiniMax-H3 model (FL2VA and Ref2VA checkpoints on Comfy-Org/MiniMax-H3). They keep first-frame/last-frame or reference-image conditioning, add ControlNet-Union video control (OpenPose / Canny / Depth), support the Latent Upscaler (low-res generation → high-res decode), and are wired through ComfyUI-Switchboard Group Bypass groups so optional stages can be toggled without editing the graph.

Setup

  1. Update ComfyUI + ComfyUI-Manager.
  2. Install the required custom nodes (Spectrum, FaceRefine, Latent Upscaler, Switchboard).
  3. Download the INT8/quantized MiniMax-H3 model files into ComfyUI/models.
  4. Grab the R2V Ref LoRA (and ControlNet-Union / Lightx2v / Turbo LoRAs) as needed.
  5. Load the workflow JSON from this Space or the source repo and run it.

Each tab below lists the workflow's node inventory (incl. nested sub-graph nodes), the model files it references, and its embedded notes. Use Preview JSON to inspect a workflow, and Download to grab it directly.

LLM prompt-engineering system prompts

The authors ship two system prompts for writing generation-ready captions. Feed them to any capable LLM, give it your rough idea, and paste its output into the workflow prompt node.

Workflows © javawock7618 · built as a showcase Space · follow each component's own license.
Copied!