Questions, answered straight.

A brand studio that runs as a full product, not a prompt box. You define a brand once — palette, typography, voice, objectives — and BrandCrafter generates every asset from that one definition: through the web studio, a scriptable CLI, or typed tools for AI agents. The prompt engineering is done once, encoded as curated, typed workflows, and reused.

36 asset types across eight families: social (Instagram posts, story formats, share cards, covers, avatars), content (thumbnails, posters), identity (logos, favicons, app icons, sticker packs), web (website heroes, banners, OG images, email headers), print (business cards, flyers, letterheads, product mockups), developer assets (CSS themes, Tailwind configs, email templates), long-form copy (bios, value props, mission statements, press releases, landing copy), and motion (ambient video loops, animated heroes). Logos export as crisp, recolorable vector SVG — not just pixels.

Both major providers, routed per asset type by what tested better for the job: Gemini draws logos, writes copy, and reads brands (vision); Veo 3.1 generates video; and OpenAI’s gpt-image-2 renders most other imagery — posters, thumbnails, social. The two image lanes fall back to each other, and if no provider can deliver, every lane degrades to a deterministic, on-brand fallback instead of erroring: you always walk away with a real asset.

Yes. Vela, Lumen, Pulse, and Strata — their identities and every asset you see — were generated end-to-end by BrandCrafter’s own production pipeline, then curated the same way any user would: pick from ranked options, approve, move on. Nothing in the demos was designed by hand.

Memory and composition. Assets build on each other: once a logo is canonical, business cards, product mockups, and covers are generated with that exact mark composed in — not re-imagined from a text prompt each time. The whole kit stays coherent because everything derives from one brand definition and its prior assets.

Every generation is schema-validated and checked before it joins the brand — contrast against WCAG AA, logo flatness for clean vectorization, structured output against typed schemas. Each asset records which model made it and what it was grounded in, so nothing is invented silently.

No. The heuristic lane is free and needs no account: palette extraction from your own images (processed on your device), deterministic logo and card rendering, and the whole brand-definition flow. Credits only come in when AI does the heavy lifting — per-asset pricing goes live with the hosted release.

With the hosted release, which is in preparation now: accounts, saved brands, the full generation studio, and per-brand galleries. This site is the read-only preview of that product.

Yes. A scriptable CLI ships with the full release — brandcrafter generate --type logo --brand vela will run the same engine as the studio, with a --dry-run flag that quotes the cost before you spend a single credit.

Yes — an MCP server is on the way. brandcrafter mcp will serve the generation system over the Model Context Protocol, so Claude Code, Claude Desktop, Cursor, or any MCP host can list asset types, read a brand’s identity, quote a generation with dry_run, and generate real assets as typed tools — no prompt engineering. It arrives with the hosted release.

Yes. A public, fully typed API — the same generation calls the studio itself makes — ships with the hosted release. Bring your own code; the brand and its assets come back typed, with API keys included in the paid plans.

Video generation as background jobs with progress and email/browser notifications when your hero loop is ready; social posting and scheduling from inside the studio; autopilot runs that keep a brand’s feed stocked; sign-in with Google/GitHub; and paid plans going live. And way more — the engine is built; the release wires it to the world.

Fully. This site is static: there is no server behind it, no account system, no tracking scripts, and it makes no API calls — nothing you do here leaves your browser.

Everything that needs the hosted product — creating or importing a brand, pricing, signing in, live AI generation — routes to one honest page instead of pretending. What does run here is real: the four brand demos are the production pipeline’s recorded output, replayed exactly as it generated them.