Image Editing
Open Settings → Plugins to install and configure image-edit. In chat, the agent uses EditImage to change an existing PNG or JPEG from the current project or a public URL, guided by a text instruction. Each call saves a new PNG into the current project under generated-images/ and returns its path; an authorized sender can use ShareImage({path}) to show it in chat. To create a picture from nothing but a description, see Image Generation.
The plugin adds no slash command or page to the web interface. It has its own card under Settings → Plugins.
Configure an image provider
An administrator adds an OpenAI-compatible provider or connects a ChatGPT account in Settings → Brain. Then open the plugin's settings from Settings → Plugins and select its provider. An OpenAI-compatible provider reuses its configured API key and base URL; the ChatGPT account uses its existing connection, and its sign-in token never enters plugin code.
The Model setting selects an image model enabled for that provider in Settings → Models; leave it empty to use the provider's default. With the connected ChatGPT account the default is gpt-image-2.5-sunburst; with an API-key provider it is gpt-image-1. The tool is unavailable until a usable provider is configured; without one, the agent is not offered the image tool.
Edit an existing image
EditImage takes a required instruction, plus either a path to a PNG or JPEG in the current project or a public HTTP(S) url. Supply exactly one source; when neither is present the call is refused.
For example, set path to assets/room.jpg and instruction to "remove the clutter from the desk and make the wall light blue." You can instead give a public image URL. If the URL response carries a content type, it must be PNG or JPEG; a response without one is treated as PNG. Fetching a source URL has a two-minute limit and follows at most five redirects.
The optional size accepts 1024x1024, 1536x1024, 1024x1536, or auto. Leaving it out or using another value lets the model choose. Editing makes a new render, not a pixel-exact patch, and the original file is never overwritten. The result is a path to the saved PNG: creating it does not send it automatically. Use output_path to save elsewhere in the project; an existing file, including the PNG source, is only replaced with overwrite: true.
Install and access
Install image-edit from Settings → Plugins → Available. Version 0.2.11 requires Elowen 0.28.53 or newer. Configure a provider as described above. The plugin is not user-grantable, so authenticated accounts can use its tool once a provider is configured. It remains subject to account tool permissions.
Enabling the plugin requires no consent confirmation. Installing, enabling, disabling or saving plugin settings requests an Elowen restart, announced as Restart. If active work delays it, the change is saved and the restart waits for that work to settle.
Limits
Image editing can take up to two minutes before timing out. A missing edit instruction is refused before the provider is called. A source image must be a PNG or JPEG. The plugin can fetch a public image URL, but cannot use a private-network address. Image requests run through Elowen, which keeps provider credentials out of the plugin.