AI SDK Image GeneratorCloudflare

AI Image Generator

One prompt, several image models side by side.

ai-image-generator

Vercel's AI SDK Image Generator — one prompt, and the image each provider's model draws from it, side by side, with a model picker per provider, a performance/quality switch and each image's time — running on Cloudflare Workers + Static Assets through @opennextjs/cloudflare, and packaged as an Orun baseline: one command takes a new site from nothing to live on stage and prod, in its own repository, under its own name.

The template's code is the code. What the port adds is the Workers build, the monorepo wrapper, a test package and the deploy contract; what it changes is what Workers and a fresh, unattended deployment force: Workers AI is a fifth provider, through the Worker's own AI binding, so a site draws from its first deploy with no key of anyone else's; the template's four providers (OpenAI, Replicate, Fireworks, Vertex AI) appear once their keys are stored — Orun secrets the owner adds after the build, handed to the Worker as Worker secrets and never seen by the build; the page lists, and the image route allows, only those providers and only the models the page offers; Vercel Analytics is gone and the "Deploy with Vercel" button is the site's own source link. What the baseline adds is everything needed to make a copy of it somebody else's — built the way cirrus-baseline is. See specs/epics/ai-image-generator for the design, the milestones and the as-built status, including every departure.

Instantiating a site

git clone --depth 1 --branch <baseline-tag> https://github.com/sourceplane/ai-image-generator
cd ai-image-generator
orun new --blueprint repo-blueprint.yaml --out ~/sites/acme-images --run-hooks --resume \
  --set reponame=acme-images --set productname="Acme Images" \
  --set githuborg=acme-inc --set subdomain=<workers-dev-subdomain>

Or, once the platform's registry lists it, from the Orun console (Baselines → ai-image-generator → Build). Two phases — 01-product (the repository created, then the whole site in one pull request, deployed to stage and prod and smoke-tested) and 03-docs (the live deployment recorded) — each landing one pull request in the new repository. BOOTSTRAP.md is the operator guide; docs/phases/ has a page per phase.

The new repository receives the site only — the app, its tests, its CI, its tooling, a README and an AGENTS.md of its own and its operating contract — renamed throughout to the site's name, repository and Cloudflare Worker. None of this repository's machinery ships. The other providers are the owner's to add afterwards (the site's README, "Add a provider"): no platform brokers them, so the card asks for no provider key, and the site goes live drawing with Workers AI alone.

What makes it a baselineWhere
The console card (inputs, secrets, programme)blueprint.yaml
The build document (modules, phases, hooks)repo-blueprint.yaml
This tree's identity, and the bootstrap's task contracts.baseline/
The rename engine (product-name seam: apps/web/app/site.ts)tooling/rebrand/rebrand.mjs
The bootstrap's helpers (deferrals, redeploy marker, work-tree strip)tooling/bootstrap/
The deployment-record rendererhooks/render-deployment-docs.sh
The site's README, agent page and context pack.baseline/product/README.md, .baseline/product/AGENTS.md, ai/context/
The baseline's tests: contract, leak, offline bootstrap, end to endtesting/, .github/workflows/baseline.yml
Cutting a baseline-vN tag.github/workflows/tag.yml

What is here (the site)

SurfaceWhere
The site's name, source link and originapps/web/app/site.ts
The page, the provider cards, the promptapps/web/app/page.tsx, apps/web/components
The providers and their modelsapps/web/lib/provider-config.ts
Which providers a deployment has, and each one's image modelapps/web/lib/providers.ts
Workers AI's image models over the AI bindingapps/web/lib/workers-ai-image.ts
/api/generate-images, /api/health (the deploy's smoke)apps/web/app/api
The Worker-secret push (the providers' keys)tooling/wire/worker-secrets.mjs
Workers build and deploy contractapps/web/next.config.ts, open-next.config.ts, wrangler.jsonc, component.yaml
Teststests/web (vitest)

Getting started

pnpm install
pnpm typecheck
pnpm lint
pnpm test
pnpm build            # next build + opennextjs-cloudflare build → apps/web/.open-next

The build needs no key and no Cloudflare credential. Workers AI has no local simulator, so pnpm --filter @app/web dev and preview reach your Cloudflare account through the AI binding: pnpm --filter @app/web exec wrangler login once. Another provider's key goes in apps/web/.env.local (apps/web/.env.example).

How it maps to Vercel

VercelHere
Next.js hosting, the image route a Vercel Function (maxDuration: 60, vercel.json)One Cloudflare Worker (.open-next/worker.js) + Static Assets; the route runs in the Worker, which waits on a provider without a wall-clock limit (the template's own 55 s timeout kept)
Four providers, each needing its key in the project's env vars, all four shown whatever is setWorkers AI added as a fifth, through the Worker's own AI binding, needing no key; the other four Orun secrets per environment (OPENAI_API_KEY, REPLICATE_API_TOKEN, FIREWORKS_API_KEY, Vertex AI's four), handed to the Worker as Worker secrets by every deploy, and shown only when set
Any model id the client sends, passed to the providerOnly the models the page offers; any other is refused before a call
Vercel AnalyticsRemoved
"Deploy with Vercel" button (clone the template)The site's own source link; orun new --blueprint repo-blueprint.yaml is the clone — a new repository, branded, deployed
Deploy on pushOrun deploys <repo>-web-{stage,prod} on merge to main (apps/web/component.yaml), smoke-tested

CI

WorkflowWhat it runsShips to a site
ci.ymlthe Orun pipeline, and the only workflow a site carries: orun plan --changed, then one job per lane. A pull request runs the verify lanes (the site built and dry-deployed per environment; the tests' quick-check lane installs and checks the package, it does not run them), except the bootstrap's own landings (orun/BASE-*), where plan alone is the check; a merge deploys stage → prod. It plans nothing while intent.yaml names the placeholder workspace ws_CHANGE_ME, as this repository does. A manual run (Orunbase's Deploy button, or Run workflow on main) plans the environment and components it names under the push-to-main trigger, so they deploy as a merge deploys themyes
baseline.ymlthe baseline's own tiers: contract tests, the blueprint read by the pinned orun, the leak gate, the offline bootstrap, the site built, dry-deployed and served over a Workers AI stand-in — with no key (every Workers AI model drawing, every keyed provider refused) and with made-up keys (no key in any file of the build, the keyed providers listed)no
tag.ymlcuts a baseline-vN tag (manual)no