VULCA Concept (Design)

VULCA Concept is the "explore before you commit" node in the ComfyUI-VULCA pack. Feed it a brief (from VULCA Brief) and it generates a handful of concept variations - by default four, up to eight - so you can look at directions before spending a real generation call on a finished artwork. It's the cheapest way to sanity-check an idea in this pipeline, which is exactly why it exists.

If you've ever run a sampler ten times just to see where a prompt goes, this is that same impulse made deliberate. The concept step separates "what are the possibilities?" from "what do I actually render?", and the rest of the pack assumes you did that separation.

How it works

It's a wrapper over the SDK's ConceptPhase. Given a brief_path, it calls the concept generator with a provider and a count, writes the concept images to the project directory, and returns their paths as a JSON array. The provider defaults to mock, which is important: mock mode produces deterministic placeholder sketches so you can build and test the whole workflow before you spend any API budget.

The provider field is where you'd switch to a real backend (the SDK routes to gemini, openai, or even comfyui - yes, the SDK can call ComfyUI itself as a provider, which is a fun bit of nesting). Real concept generation needs the matching API key or a configured local ComfyUI, same as the rest of the pack.

Inputs and output

• brief_path (required) - wire it from VULCA Brief. This is the intent you're exploring.
• count - 1 to 8, default 4. How many concept variations to generate.
• provider - mock by default; swap to gemini, openai, or comfyui for real sketches.

The single output is concept_paths_json - a JSON string listing the generated file paths. Here's the pack's defining quirk in one sentence: these nodes hand you file paths and JSON, not IMAGE tensors. You can't wire concept_paths_json into a sampler; you'd load the files it points at if you want to see them in the graph.

Install and troubleshooting

ComfyUI Manager (search "VULCA") or:

cd ComfyUI/custom_nodes/
git clone https://github.com/vulca-org/comfyui-vulca
pip install "vulca>=0.12.0"

Then restart. No models to download - the heavy lifting lives in the vulca SDK.

If you leave the provider on mock and get "identical" placeholder images every time, that's expected, not a bug - that's the deterministic dry-run mode so you can check your wiring. When you're ready for real concepts, set provider to a real backend and make sure your API key is in the environment (GOOGLE_API_KEY is what the pack README uses for the Google VLM path). If a node fails with ModuleNotFoundError: vulca, the SDK isn't installed - that's the same lazy-import trap as the rest of this pack.