Every image model.
One port.
Give your agent — or your app — image generation with every major model behind one endpoint and one prepaid balance. Top up, it works.
Works with Claude Desktop · Claude Code · Cursor · Windsurf · any OpenAI SDK
images generated & counting 12+
models, one endpoint 3.0s
median, fastest $0.041
per image, from
One key. One balance.
Zero provider accounts.
Nano Banana, GPT Image, Gemini — behind one OpenAI-compatible endpoint. New models land the week they launch; your code doesn't change.
The MCP server is the front door: connect once in Claude Desktop or Cursor, and your agent generates, edits, and iterates — with a balance warning before it ever stalls.
"args": ["-y", "@pixfaro/mcp"],
"env": { "PIXFARO_KEY": "pf_live_…" } }
No subscription, no per-provider billing. Prepaid balance, pay per image, credits last 365 days, optional auto-reload so long agent runs never stop mid-pipeline.
Your agent writes the post.
Pixfaro draws the picture.
The founding use case: an agent drafts your post or article, calls Pixfaro for the picture, and publishes — one pipeline, zero tabs. One sentence: "add a visual to this post."
Thumbnails, album covers, book covers, coloring pages — one account, every model.
Avatars, stickers and logos inside your own product — the same API, your users never see the providers.
One more, every second.
Every frame, real output.
Logos, products, characters, scenes — each one came out of the Pixfaro API, prompt in, image out. The queue keeps growing because the agents using it don't stop.
Every image, signed.
One parameter.
Models can't spell your handle or draw your logo reliably — so we don't ask them to. Upload your logo once, then add overlay to any request and Pixfaro composites your logo or text into a corner, pixel-exact, with automatic contrast. Nobody else in the space does this.
Real brand logos and handles above — each a different overlay, composited by Pixfaro into the corner, never generated by the model.
"overlay": { "logo_id": "logo_8f2a…", "position": "bottom-right" }}'
# or "text": "@yourbrand" — upload logos with POST /v1/logos
We benchmark every model weekly.
You pick by numbers.
| Model | Best for | Median speed | Price / image | |
|---|---|---|---|---|
| gemini-flash-lite | drafts and thumbnails where speed beats fidelity | 3.0s | $0.041 | Try |
| nano-banana-2 | general-purpose images: blog art, social posts, mockups | 10.7s | $0.080 | Try |
| gemini-pro-image | detailed scenes and typography-heavy images | 20.8s | $0.164 | Try |
| gpt-5-image | complex prompts needing strong instruction following (async only) async coming soon | 53.0s | $0.238 |
Prices are provider cost + 20%, updated when providers move. Full table → /pricing
OpenAI-compatible.
Your SDK already works.
client = OpenAI(
base_url="https://api.pixfaro.com/v1",
api_key="pf_live_…")
img = client.images.generate(
model="nano-banana-2",
prompt="a lighthouse at night")
CLI — npx pixfaro gen "prompt" -o cover.png
Remote MCP — one-click connector for claude.ai, OAuth, no keys in configs
Webhooks — async jobs post back when slow models finish
Top up $10.
Generate ~240 images today.
$1 free credit on signup. No card required to try.
Start freeFair questions,
straight answers.
Something else? support@pixfaro.com — a human answers.
Can Claude generate images with this?
Yes — that's the point. Claude can't generate images natively; connect the Pixfaro MCP and every major image model becomes a tool Claude can call, in Desktop, Claude Code, or via claude.ai connectors.
How is this different from OpenRouter?
OpenRouter routes text models; Pixfaro is the same idea for images — one endpoint, one prepaid balance, every major image model, with per-image cost and latency in the response.
What happens when my balance runs out mid-run?
Generations stop cleanly with a clear error — never a silent failure. Turn on auto-reload and a long agent run tops itself up (with a monthly cap) so it never stalls.
Do you store my images and prompts?
Images are stored so you can fetch them by URL; prompts are retained for a bounded window then scrubbed. It's your data — export or delete anytime.
Which model should I pick?
Start with the fastest for drafts and agent loops, step up to a quality model for hero shots. The pricing table lists speed and price per model so you choose by numbers.