alanzhao0128/dsh-image-plugins

Image understanding and generation for a text-only main model, over any OpenAI-compatible or DashScope endpoint you configure.

Gives a text-only main model eyes and an image generator without changing the model. The understand_image tool reads a workspace image, sends it to your vision endpoint, and returns the description as a tool result — so the description enters the session log and the main model reasons about it without ever receiving an image block. generate_image posts a prompt to your image endpoint and saves the result. Since 0.3.5, reference-image editing works through either provider instead of only DashScope, so image-to-image is available on any compatible endpoint.

Vision & Multimodal ★ 2
View on GitHub ↗

Install

dsh plugin --profile web add github:alanzhao0128/dsh-image-plugins#v0.3.5

Pinned to the v0.3.5 tag, the release the README describes (that version removed the hardcoded DashScope provider for reference images). After installing, you override the image-plugins row in your profile's cordis.patch.yml with your own endpoint and key for a vision model and for an image-generation model. No keys ship in the package.

Compatibility

Every capability is optional and enabled only when its config block is present, so an unconfigured install is inert. Endpoints must be OpenAI-compatible (POST {baseUrl}/chat/completions accepting image_url data URLs, and POST {baseUrl}/images/generations returning data[0].b64_json or a URL); the optional dashscope adapter speaks Alibaba Model Studio natively. The README states that the autoUnderstand pre-step rewrite is unit-tested but was never verified end to end in a live session, and is off by default.

Details

Alternatives

zemanzhang809/dsh-stt-plugin · TaurusWood/dsh-plugin-appshot · Cikian/dsh-cture-image

More plugins in Vision & Multimodal

Browse more in Vision & Multimodal

Guides for Vision & Multimodal plugins