dsh-image-studio

After an AI generates an image, users often still need to draw a few arrows, erase stray elements, or crop to a specific ratio before handing the image back to the agent. That round-trip of downloading, opening an external editor, and re-uploading is tedious. dsh-image-studio folds editing directly into the conversation: the image renders as an inline card with a persistent Edit button in the bottom-left corner, opening a full annotate, erase, and crop workspace.

The three editing tools each have a clear output. Annotate supports freehand strokes, rectangles, and arrows with colour choice and brush width, baking marks into a PNG alongside the original. Erase produces an RGBA mask at the original image size with transparent pixels marking the painted region, ready for inpainting. Crop offers aspect presets, four corner handles, and real-time validation against PackyAPI pixel rules. Pressing Generate writes the output next to the original, composes a message listing paths, target size, and the wording the user typed, and submits it so the agent picks up the work on the next turn. The hand-off is plain text plus file paths, so any image-generation workflow can consume it.

Built for users who frequently generate and iterate on images within the DeepSeek Harness Web GUI: product managers annotating flow diagrams, designers quickly erasing for a fresh pass, content creators cropping for different platforms. The plugin has zero npm dependencies, the host side uses only Node built-ins, and the client side pulls React from the page module table, so after installation a simple page refresh is all it takes.