Generate and edit images with MonkeyAI

Confirm the model and key group

Create a dedicated image key in key management and review live pricing. The table follows the current model listing; group descriptions do not establish tested API parameters.

Codex Pro（仅限Codex） explicitly forbids image generation. Codex Lite currently lists no image models. Although the Grok enterprise description advertises images and video, its model list contains no dedicated image ID; this guide does not promise access to grok-imagine-image or its quality variant. The current Gemini listing also lacks gemini-3-pro-image. MonkeyAI has no Muse/Hunyuan groups, so their availability is not promised.

Prepare a client or skill manually

Do not paste a real key into chat. If you already have a trusted image client or skill, inspect its origin, protocol support, custom base configuration and key handling. Otherwise, use the API scripts guide to check interfaces before installing unknown scripts.

The OpenAI-compatible base is https://monkeytheme.com/v1. Native Gemini routes use /v1beta/, typically with the root URL as the client's base. Verify generation and editing endpoints and parameters for the chosen model. Do not mix Responses, Images and Gemini fields in one request. A visible model listing does not prove that generation and editing both work.

Prepare the key on Windows, macOS and Linux

If your trusted tool explicitly supports IMAGE_GEN_API_KEY, set it in the parent terminal. Windows PowerShell placeholder example:

$env:IMAGE_GEN_API_KEY = "YOUR_API_KEY"

macOS / Linux:

export IMAGE_GEN_API_KEY="YOUR_API_KEY"

This variable is a tool convention, not a standard automatically recognized by every client. Map it according to the tool's instructions and launch the tool from that terminal to inherit it. GUI programs may not inherit shell variables; use their secure key input instead. Keep real keys out of prompts, source code and Git.

Describe three distinct image tasks

Text to image: “Create a blue and white product poster with a centered subject, space for a title at the top and no added text. Use gpt-image-2 and first identify the group and dimensions you will use.”

Reference image: “Create a new composition from my reference image. Preserve the subject silhouette and palette, but change the background and viewpoint. Confirm first that the selected model accepts image input.”

Image editing: “Change only the uploaded image's background to pale blue. Preserve subject position, proportions and existing text. Explain any limitation if local editing is unsupported.”

These prompts express tasks; they do not install a skill or guarantee a particular API call. Check the key group before switching models. Use reference files you are entitled to process and inspect which files the client will upload.

Handle results, troubleshoot and assess output

Inspect whether the tool returns an image URL, base64 or Gemini content blocks instead of assuming one format. URLs may expire: save the output, verify that it opens, then compare the subject, lettering and edited area. Dimensions, count, transparency and reference-image limits were not validated here; arbitrary values are not promised.

Image requests are billed according to actual usage. Check the key for 401, group for 403 and model ID for model_not_found. For format errors, inspect protocol and model input limits. A tool without image support cannot be fixed merely by renaming a text model. Share redacted errors and model IDs, keeping keys private. See API scripts for programmatic integration.