(Documentation in English)
What it does
This tool turns a text prompt into a 1024×1024 image. It runs FLUX.1 [schnell] (@cf/black-forest-labs/flux-1-schnell), a 12-billion-parameter text-to-image model, on Cloudflare Workers AI — the same network that serves this site. No account, no API key, no setup: type what you want and press Generate.
Generation uses a fixed 4-step inference pass — schnell’s design point — and always outputs a single 1024×1024 PNG. There is no size or style control in v1; the prompt is the only input.
Privacy, honestly stated: this is not a client-side tool. Your prompt travels from your browser to a CosmoDev Worker, which forwards it to the image model and returns the image. That transit is the only reason a backend exists here — we store no prompts, no images, and no raw IP addresses. The rate-limit counters keep only anonymous counts keyed by a truncated
hashhashA one-way function that maps data of any size to a fixed-length digest. Same input, same digest; any change, different digest; not reversible.
of your IP.How to use it
- Type a description into the box — what the subject is, the style, the lighting. Anything from
a red cube on a white studio backgroundtoisometric pixel-art library, warm colors, 16-bit. Up to 600 characters; multi-line prompts are fine. - Press Generate image (or pick one of the example chips to fill the box first). Generation takes a few seconds; the button shows a spinner while it runs.
- The result appears below with two actions: Download PNG saves the image, Copy share link copies a URL that carries your prompt (
?prompt=…) — opening it re-runs the same generation. - If something goes wrong, the tool says which of these happened: the hourly limit (10 images), the shared daily limit (40 images), a model error (try rewording), or a network failure.
Examples
Product-style shot — a red cube on a white studio background, soft light returns a clean studio render with soft shadows.
Pixel art — isometric pixel-art library, warm colors, 16-bit returns a retro isometric scene in a warm palette.
Minimal illustration — minimal watercolor mountain range, two colors only returns a two-tone watercolor landscape.
Good to know
- Rate limits, by design. Anonymous tools with per-image model cost need them: 10 images per hour per visitor, and 40 images per day shared by everyone — both reset at 00:00 UTC. The daily cap keeps the whole tool inside Cloudflare’s free daily AI allocation, so it stays free for everyone.
- Your prompt is the whole input. There is no negative prompt, no seed control, no image-to-image in v1 — describe what you want in positive terms.
- Nothing is kept. No prompts, no generated images, no raw IPs are stored. The counters are anonymous totals; closing the tab erases the image from your side too (download it to keep it).
- Errors are honest. A 429 answer means a limit was hit (the message says which and when it resets); a model error usually means the prompt produced nothing renderable — reword and retry.
- First CosmoDev tool that sends your input to an AI model through our Worker. Other egress tools (WHOIS/
RDAPRDAPThe modern JSON REST protocol for WHOIS-style registration data (domains, nameservers, entities), replacing the legacy plain-text WHOIS protocol.Learn more
) query registries browser-direct; this one needs our Worker because the model binding lives there. The page footer states which situation you are in on every tool.