Text-to-image, reference-image edits and upscaling on one endpoint family. Standard-tier models are one flat price per image, premium models another — published up front, returned on every call, never a token-math surprise.
POST /v1/image with a prompt (≤2000 chars), optional species_name, n 1–4
and size. Each image comes back as base64 and/or a hosted URL.
Add images[] — URLs, data URLs or raw base64 — to edit or combine photos. GPT Image 1/1.5 take up to 16;
the Venice edit models take up to 6. The first image is the base; the rest are layers the prompt can refer to.
POST /v1/image/upscale with scale 2 or 4, or an exact target_size — Fire picks the
smallest real factor that covers it and center-crops.
Standard models are one flat price, premium models (GPT Image, Imagen 4, Flux 2 Max) another. The current numbers are on
the pricing page and on every response's meta.price.
Every image model publishes what it can render — fixed sizes, aspect ratios or dimension ranges — as size_policy
on /v1/models. The Layer 2 workflow coerces to the nearest supported size and tells you when it did.
POST /v1/workflows/image runs pre-gen compliance, prompt expansion, per-model rendering, parallel generation,
storage and post-gen review. Blocked images are deleted server-side and never returned.
A plain prompt generates from scratch. Add images[] and the
same endpoint becomes an editor. If you omit species_name while sending images, Fire picks the default
reference-capable model, not the overall default.
Edit-only models (the Venice Qwen/Seedream edit family) have no text-to-image path at all; asking a non-edit model like SDXL to edit returns a clear 404 rather than silently ignoring your images.
Pick one by species_name, or omit it and Fire routes to the default. The list is generated from the same query /v1/models serves.
On /v1/image: 256×256, 512×512, 1024×1024, 1024×1536 or 1536×1024 (default 1024×1024). Each model's real
envelope is published as size_policy on /v1/models; the Layer 2 workflow coerces to the nearest
supported size and reports size_coercion when it does.
A short, published list of models (GPT Image 1 and 1.5, Imagen 4, Flux 2 Max) billed at the premium flat rate; everything
else is the standard flat rate. Both numbers are on the pricing page and in GET /v1/capabilities.
Provider-hosted models (via Eden AI) apply their own content moderation. The Layer 2 workflow adds Fire's own pre- and post-generation compliance per site policy, and a blocked image is deleted server-side and never returned — the classifier call that blocked it is still a real, billed call.
Not unless your account opts in. The call's metadata — model, size, count, price, call_id — is always logged; the prompt text is withheld by default. See the privacy & data policy.
Free to sign up — you only pay when you top up. No card required to get your first key.