Frequently Asked Questions

A: For light use, yes — you can try images and text without paying or creating an account. Heavier or sustained use moves into Pollinations’ own limits and Pollen-style billing. “Forever free” is the marketing line; treat costs as “check the official docs before you depend on it in production.”

A: Pollinations AI provides access to multiple models through the unified API at gen.pollinations.ai:

• Image Models: Flux, GPT Image Large, Seedream, Kontext
• Text Models: GPT-5, Claude, Gemini, DeepSeek V3.2, Qwen3-Coder-30B
• Video Models: Seedance, Veo (alpha)
• Audio: OpenAI Audio (text-to-speech, speech-to-text)

The models page has rough guidance on when to try which — upstream still owns the final word on availability.

A: Visit enter.pollinations.ai to get your API key. You can choose between:

• Publishable Keys (pk_): For client-side use, demos, and prototypes. Rate limit: 1 pollen/hour per IP+key.
• Secret Keys (sk_): For server-side applications only. No rate limits. Never expose in client-side code.

See API notes for curl examples and footguns around keys.

A: It is the unified host for Pollinations-backed generation — one base URL to configure in your client, with paths and parameters selecting modality and model. You still think in “what am I calling?” not “magic socket.”

A: Pollen is how upstream meters heavier use — roughly pay-as-you-go. Exact earning and pricing rules change; read Pollinations’ billing page before you budget a product around it. Casual tries may never touch Pollen.

A: Often yes for casual use in the browser. For anything customer-facing or high-volume, get a key, use server-side secrets correctly, and expect clearer limits — anonymous traffic is the first to get squeezed.

A: The core Pollinations work is open on GitHub — this website is separate, but rides the same public APIs. Read licenses and issues before you fork for a business.

A: Gemini Agent Tools enable advanced capabilities on Gemini models, including:

• google_search: Search the web for real-time information
• code_execution: Execute code in a sandboxed environment
• url_context: Extract and process content from URLs

Handy when the model needs the web, runnable code, or a page pulled in — not every deployment exposes all three.

A: Yes — when you mint a key you can allow everything or tick only the models you need. Narrower keys leak less damage if they escape into a client bundle.

A: You can push fairly large frames depending on model settings; upscalers like SPAN can stretch a good mid-res base. For print, still proof at real size — pixels lie.