I’m developing Image Studio, a Perchance-based image generator and editor.

The goal is to let users upload an existing image and request a specific alteration in natural language while preserving everything else.

For example: “Change this apple tree into a lemon tree, replacing all the apples with lemons while preserving the tree’s shape, branches, leaves, lighting, background and camera angle.”

We’ve successfully implemented browser-based SlimSAM segmentation, precise recolouring and geometric transformations.

However, genuine generative editing isn’t possible through the current image-generation backend.

I noticed that “text-to-image-plugin” already contains a “referenceImage” parameter, but its source states that this functionality is deliberately disabled server-side.

Would it be possible to support image-to-image generation or masked inpainting in the future?

Even a basic API accepting an original image, prompt and optional mask would be extremely useful.

Ideally, this could work through Perchance’s existing free image-generation infrastructure, subject to whatever usage limits are necessary.

I’d be interested to know whether reference-image support is planned or technically feasible with the current backend.