Skip to content
4 providers · 22 models · 240+ edge nodes

The AI Router for Images

ImageGPT is the gateway for AI image generation built for developers and agents. Write an <img> tag; we route it to the best current model, generate once, cache at the edge, and return the image. No client secrets, no SDK, no storage, no backends.

See how it worksType a description of your image here and hit Generate.
a red canoe on a still lake at dawn
No SDKNo storageNo backendNo secrets to leakNo provider accounts

You always get an image

Generation is flaky — providers time out, rate-limit and refuse. We absorb every one of those and keep going until an image comes back. Your app never handles a retry.

The URL is the hosting

Everyone else hands back bytes or a URL that expires, and the storage problem becomes yours. Ours is permanent and edge-cached from the first request. No bucket, no CDN, no cleanup job.

Stop tracking models

New models ship weekly across price, quality and speed. Name the outcome you need and we serve the most cost-efficient model that hits it — re-mapped as the field moves, without a code change.

Routing

Two ways to ask for an image.

Pin a model by alias, or name the outcome and let us pick. Every route, and most aliases, has more than one provider behind it.

route=<outcome>/<level> · model=<alias>
By modelSingle model, usually several providers
labGoogle DeepMind
aliasgemini-3-pro-image
labBlack Forest Labs
aliasflux-2-pro
labBlack Forest Labs
aliasflux-2-dev
labBlack Forest Labs
aliasflux-1.1-pro-ultra
labIdeogram
aliasideogram-v3
labRecraft
aliasrecraft-v3
labByteDance Seed
aliasseedream-4.5
labAlibaba Qwen
aliasqwen-image-2512
labRunDiffusion
aliasjuggernaut-flux-pro

An alias is a canonical model name, not a provider endpoint. Most aliases resolve across several providers for that same model, so pinning gains reproducibility without giving up redundancy. See all models

By outcomesay what matters, not which model
routequality/fast
best forMost use cases and real-time apps. The default when route is omitted.
routequality/balanced
best forHigher quality needs without the full latency cost.
routequality/high
best forFinal output and marketing materials.
routetext/fast
best forPrototyping layouts that contain words.
routetext/balanced
best forProduction content with readable text.
routetext/high
best forSigns, posters, logos, text overlays.
routerealistic/fast
best forQuick photoreal drafts.
routerealistic/balanced
best forProduction realistic imagery.
routerealistic/high
best forPortraits and product shots.

Three outcomes — quality, text, realistic — at three levels each. Omit route and you get quality/fast. We map each route to the best current model and re-map as models ship; your URLs never change.

Features

Control, reliability, and performance.

Outcome-based control. Quiet defaults. Predictable behaviour.

Multi-provider

Do not be confined to one model or one provider. Access best-of-breed from a single image URL.

Outcome-based control

Choose what matters — quality, speed, text adherence — without managing models.

Model routing & reliability

Requests are routed across multiple providers to return the best available result.

Reference images

Provide reference images to supported models for grounding and style foundations.

Presets

Define re-usable image configurations and prompts for predictable generations.

Predictable behaviour

Stable URLs and consistent outputs, designed for production usage.

The whole integration

Request in, image out. Nothing in between is yours to run.

The response is the image itself — not JSON, not base64, not a job ID to poll. Drop the URL into an img tag, an email, a CMS field, or an agent's output and it works.

request
<img src="https://demo.imagegpt.host/image
     ?prompt=matte+ceramic+mug+on+concrete
     &route=quality/high
     &aspect_ratio=4:3">
response
200 image/png
cache-control: public, max-age=31536000, immutable
x-cache: EDGE_HIT
What you stop operating

Every piece of image plumbing, retired.

This is the list most teams build before they ship their first AI image. With ImageGPT it stays unbuilt.

01Provider accounts and their keysNo keys at all on your side, and your host can be scoped to your domains.handled
02Base64 decoding and file writesThe response is already an image.handled
03Object storage and lifecycle rulesThe image lives at the URL you wrote.handled
04CDN and cache headersCached on 240+ edge nodes, sub-50 ms.handled
05Retries, timeouts, failoverMulti-provider routing on every route, and on most aliases.handled
06Model evaluationRoutes re-mapped as models ship; URLs never change.handled
Use cases

ImageGPT is for builders.

Editorial clarity. Practical integration. Designed for teams that ship. Every thumbnail is a live URL — click one to see the call that made it.

Agentic coding

Agents like Claude Code, ChatGPT and Cursor generate images by constructing a URL. No API calls, no auth tokens, no SDK install.

Image-based frontend apps

Redesign rooms, create marketing assets, generate headshots — all from the frontend. No backend infrastructure to touch.

Mobile apps

Ship native iOS and Android features without backend infrastructure. Construct URLs in Swift or Kotlin; sharing a creation is sending the URL.

Bulk image generation

Bulk generate simply by building URLs. When a URL is rendered in a browser, the image is generated on demand.

Personalised images

Generate a unique image for every customer from personalised inputs, and only pay for the ones actually rendered.

Infographics

Rich, data-driven visual content with ingredient breakdowns and layered composition — for health, food and educational content.

Presentations

Slides with integrated charts, text and 3D visualisation — business decks and pitch material that look polished without design overhead.

Video game assets

Prototype 3D-style assets for scenery, characters and structures — quick visual iteration in early development, or concept art for a pitch.

Technical

Zero infrastructure. One img tag.

Works anywhere images work. Designed for frontend apps and LLMs.

Agent-native

Built for AI agents. Claude Code, ChatGPT, Cursor and any LLM can generate images with a URL.

One img integration

Generate images directly from an image URL — no backend, no SDKs, no API keys.

Automatic generation

The first request generates the image. Subsequent requests return cached images instantly.

240+ edge nodes

Images cached on Cloudflare's global network. Sub-50 ms response times worldwide.

Pay per generation

Billed only when images are generated. Cached requests are free. No surprises.

4 providers, 22 models

Multi-provider routing with automatic failover. Your images always generate.

Models fromGoogle DeepMindBlack Forest LabsRecraftIdeogramAlibaba QwenByteDance SeedZ.aixAIRunDiffusionVyro

Your first image is one request away.

Free 7-day trial included. Cancel any time.