runblob · API
runblob connects several image models through one pay-per-use API. This page gathers the models that create images from a prompt and edit them from references: pick one by the inputs you need, then open its page with live rates and a playground.
Text-to-image builds an image from a written scene description — useful when you have no source yet. Reference editing starts from existing images: you can keep the subject, composition or palette and change only the details you need. All three families accept text and reference image URLs; the list below states which inputs each one takes and how many references are available.
Images are usually billed per generation, so the total depends on the model, version and quality you choose, and retries also count toward your budget. No prices are quoted here: live USD rates load from the catalog on each model page and in the playground, and the estimated cost is shown before you start.
Open the model you need and build the request in its playground: it generates a cURL with the exact endpoint, request fields and Bearer authorization. Move the request into server code so the secret key never reaches the browser, and pass reference images as service-accessible URLs. Authentication, generation status and webhooks are described in full in the API documentation.
Billing is pay-per-use: you top up your runblob balance and spend it on API generations. Top-ups are charged in USD, so budgeting matches the rates shown on each model page.
Inputs: Prompt and reference images
Google identifies Nano Banana Pro as Gemini 3 Pro Image and Nano Banana 2 as Gemini 3.1 Flash Image; in runblob Standard accepts up to 4 references, Pro and 2 up to 10.
OpenAI
Inputs: Prompt and up to 8 references
OpenAI introduced ChatGPT Images 2.5 in September 2026; in its API these are gpt-image-2.5-flare and gpt-image-2.5-sunburst, and both modes are available in runblob.
Kuaishou
Inputs: Prompt and references
A dedicated Kuaishou family for still images; the runblob catalog offers O1 and O3 variants, with their settings shown in the playground.
Text-to-image builds an image from a written description; reference editing starts from an existing image and preserves what you ask to keep. Nano Banana, ChatGPT Images 2.5 and Kling Image accept both text and reference image URLs; the number of supported references depends on the model and version.
Prices are not quoted in the text. Open the model page you need: live USD rates load from the catalog, and the estimated cost for the chosen version and quality is shown before you start generating.
Yes. Add the image URL as a reference and describe what to change and what to keep. The URL must be reachable by the service; local file paths are not accepted.
Yes. runblob provides one integration and a shared balance: a single key and one Bearer authorization format are used for all the models listed here; only the version and request fields differ.