AI artwork

With an image API key, --ai replaces the default initial-on-a-tile logo with a generated one and adds a generated background behind every canvas:

npx @rtorcato/brand-kit --ai                                  # first key found
npx @rtorcato/brand-kit --ai --ai-provider gemini --ai-prompt "neon line art"

Providers

Provider	Env vars	Default model
higgsfield	HF_API_KEY_ID + HF_API_KEY_SECRET (or HF_KEY=id:secret)	higgsfield-ai/soul/v2/standard
openai	OPENAI_API_KEY	gpt-image-1
gemini	GEMINI_API_KEY	gemini-2.5-flash-image
leonardo	LEONARDO_API_KEY	Phoenix 1.0 (de7d3faf-762f-48e0-b3b7-9d0ac3a3fcf3)
recraft	RECRAFT_API_TOKEN	recraftv3 (vector logo)
cloudflare	CLOUDFLARE_ACCOUNT_ID + CLOUDFLARE_API_TOKEN	@cf/black-forest-labs/flux-1-schnell

Without --ai-provider, brand-kit uses the first provider in that order whose key is set. --ai-model overrides the model, and --ai-prompt adds a style hint to both the logo and background prompts.

For Leonardo, a UUID --ai-model is a v1 model ID (as listed in Leonardo's platform models), and anything else is a v2 model name such as gpt-image-1.5; generations are polled on v1 either way.

Recraft draws the logo in its vector_illustration style, so it comes back as real SVG, and gets the accent as a colour input as well as in the prompt. The background uses the raster digital_illustration style.

Cloudflare Workers AI is the cheapest option: FLUX.1 schnell costs a fraction of a cent per image, and a brand-kit run fits in the free daily allowance. It is checked last, because CLOUDFLARE_API_TOKEN is often set for deploys, so pass --ai-provider cloudflare when another key is set too. The token needs the Workers AI permission; a deploy-only token gets a 403. Create one at dash.cloudflare.com → API Tokens with the "Workers AI" template, and copy the account ID from the dashboard. --ai-model reaches other Workers AI models, such as @cf/leonardo/phoenix-1.0 or @cf/leonardo/lucid-origin, which get the canvas width and height; FLUX.1 schnell draws square and the canvas crops it.

Keys come from the shell or a .env in the target repo; the shell wins. brand-kit warns if that .env is not gitignored.

What gets saved

The art is saved as sources, so it is generated once:

• brand/logo.png: the generated logo.
• brand/background.png: the canvas background.
• brand/favicon.svg: a self-contained SVG that embeds the logo. With Recraft the logo is vector, so it is favicon.svg (scripts and event handlers stripped) and no logo.png is written.

Renders after that are deterministic and need no key. Run --ai again only when you want new art.

Your own background

Any image works as a background. Drop it in as brand/background.png and run:

npx @rtorcato/brand-kit --update