VELVET VICE MiniMax H3 — Live Preview Bridge
The same live preview, positioned where the graph actually needs it
- model
- video_vae
- model
If you compare VelvetViceMiniMaxH3LivePreview and VelvetViceMiniMaxH3LivePreviewBridge side by side, the inputs and outputs are identical, the description is identical, and you'd be forgiven for thinking one of them is a duplicate that slipped into the pack. It isn't. The bridge exists because of where in the graph the preview needs to live - and how ComfyUI's subgraph execution treats nodes differently depending on their category and position.
In this pack, the live preview you'd add by hand in the main graph is the LivePreview node. But the reference workflow's System Hub wraps its model loading in a subgraph, and the preview needs to hook the sampler that lives inside that structure. The bridge is the same node re-registered under the Internal category - the author's own framing in the code is that the bridge just calls the base preview's logic. Same wrapper, same quality tiers (LOW latent2rgb / MEDIUM TAEHV / HIGH full VAE), same guaranteed "preview failures never abort or modify final sampling."
Why the separation is deliberate
ComfyUI's newer subgraph feature isolates graphs, and nodes that cross the boundary have to behave predictably. The typed outputs in this pack (the VELVET_VICE_H3_RENDER_CONFIG and VELVET_VICE_H3_MODEL_CONFIG bundles) are a running theme in the code comments about "avoiding COMBO values crossing a ComfyUI subgraph boundary." The bridge is the same idea applied to the preview: the pack ships a public-facing preview node for hand-built graphs and an internal-facing twin for the reference workflow's own wiring, so the reference graph doesn't have to reference a "Preview" category node in an internal spot where it doesn't belong.
Inputs and outputs
model(MODEL) - passed through unchanged; the preview wrapper rides along.enabled(default true),preview_fps(24),preview_quality(0–3),max_preview_edge(up to 1024).video_vae(VAE, optional) - for the HIGH tier.
Output: model (MODEL).
Install and practical note
Same pack:
cd ComfyUI/custom_nodes
git clone https://github.com/Velvet-Vice/velvet-vice-minimax-h3
then restart. The MEDIUM preview tier still wants models/vae_approx/taeh3_decoder.safetensors.
Practical advice: if you're hand-building a workflow, reach for the plain LivePreview node - it's in the Preview category for a reason and you'll find it in the right menu. If you're opening the reference workflow and see this bridge sitting inside the System Hub's subgraph, don't delete it or swap in the other one; that would break the isolation the pack's authors set up. Think of the bridge as the same tool mounted on a different bracket.
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| enabled | BOOLEAN | true | — |
| preview_fps | FLOAT | 241–30 | — |
| preview_quality | INT | 30–3 | — |
| max_preview_edge | INT | 1024384–1024 | — |
| video_vaeopt | VAE | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |