Your AI.
Your unlimited canvas.

The standard AIBackground studio includes five free credits each UTC day after signup. You can also purchase a one-time credit pack. If you already have an image-provider API account, choose My AI instead: personal-provider generations and revisions use zero AIBackground credits and have no AIBackground daily generation quota.

Your provider still charges your API account and controls its balance, model access, rate limits and acceptable-use rules. Normal request pacing applies in the app. Account email verification is required after the first successful generation under the standard policy, including in My AI mode. Unlimited describes the absence of a studio credit or daily-generation cap; it does not promise unlimited provider throughput, free third-party inference or every model capability.

1. Sign in and open My AI.
2. Choose a preset or Custom provider. Give the connection a name and enter the exact image-model identifier from your provider.
3. Add a public HTTPS API base URL or exact generation endpoint, then paste your own API key. Keys belong in the dedicated password field.
4. Enable revisions only when the selected model accepts an input image. Add a revision URL when the standard route cannot be inferred.
5. Save, optionally check the model-list endpoint, and select your connection in the studio.

You can keep multiple model connections, choose a default, edit settings, rotate a key or remove a connection. Removing a provider removes its stored key and leaves saved wallpapers in your collection.

OpenAI-compatible Images supports the standard JSON image-generation API and multipart image edits, including compatible gateways such as Cheaper Inference. You may use a base such as https://your-provider.com/v1 or an exact generation route. The standard edits and models routes can be inferred.

OpenRouter Images uses its dedicated image API and image-reference requests. Gemini generateContent sends text and, for revisions, an inline source image to a Gemini image-capable model. The exact model determines which capabilities are available.

Custom synchronous JSON lets you define separate generation and revision request templates and a response path such as data.0.b64_json or images.0.url. Templates support prompt, model, aspect ratio, seed and source-image placeholders. The API must return an image in its initial response; asynchronous job/polling APIs need a compatible gateway. Raw PNG, JPEG and WebP image responses are supported too.

Open an image from your own collection and choose Revise. Select My AI, pick an editing-capable connection, and explain the change. AIBackground sends your saved source image together with the new creative brief. A text-only model or a generation-only endpoint cannot perform a true image revision.

Your original stays saved. The new result becomes a separate private wallpaper with its own name and download options. A provider or model that is unavailable produces a clear error; the request does not silently switch to a shared provider or spend your AIBackground credits.

API keys are encrypted on the server. The app does not return saved keys in provider lists, public gallery pages or account exports. Prompts and revision images are sent to the provider you select, so its own privacy and retention terms apply.

Generation, revision and model-list URLs for one connection must use the same provider origin. Redirects and private network destinations are blocked. If you change the provider hostname, paste the key again; a saved key is not silently forwarded to a new host. Do not include passwords or tokens in extra JSON settings.

Check & list models sends a GET request to your configured model-list endpoint. It never submits an image-generation request. A successful model-list check confirms connectivity and can list available model identifiers; it does not guarantee that a model can generate or edit an image.

If an exact custom endpoint has no model-list URL, the app says that no network test was performed. To diagnose a generation failure, check the API format, exact model ID, provider balance and supported revision format. Provider errors are sanitized so credentials are not echoed into the interface.

Connect your AI, open the studio in My AI mode, or compare the standard credit options.