dsh-image-base64
Run the following command in DeepSeek Harness:
dsh plugin install pureexe/dsh-image-base64
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install pureexe/dsh-image-base64 in DeepSeek Harness to install this plugin; the source is available at https://github.com/pureexe/dsh-image-base64
About this plugin
When routing vision-model requests through an OpenAI-compatible gateway that enforces a strict MIME whitelist—LM Studio being the most common offender—screenshots or PNGs with an alpha channel get re-encoded to WebP upstream, and the gateway rejects the payload with a 400 before it ever reaches the model.
This plugin wraps the attachment store's readImageRequest() call (the exact point where dsh-llm-pi-ai fetches bytes and MIME type for the wire) and transcodes the result to PNG via sharp only when the media type falls outside an accepted list (image/png, image/jpeg, image/gif by default). Already-accepted formats pass through untouched, so the transcode cost is paid exclusively by images that would otherwise be rejected.
It is designed for anyone whose vision pipeline goes through a gateway with a rigid MIME whitelist, particularly LM Studio users who paste screenshots or attach transparent PNGs. If your gateway does accept WebP, you can narrow the acceptedMediaTypes list via a patch layer to skip the redundant PNG re-encode.
Use Cases
- Screenshots rejected with 400 when calling vision models through an LM Studio gateway
- Transparent PNG pastes blocked because upstream encodes them as WebP
- Automatically adapting DSH attachment layer to strict MIME whitelists on OpenAI-compatible gateways
Best For
- DSH users routing vision models through LM Studio
- Developers sending multimodal requests via OpenAI-compatible gateways
- Ops teams hitting MIME whitelist rejections and needing a quick fix
Related Plugins
A unified suite combining hot runtime injection, task-aware thinking-mode routing, and a graded session protocol with red-team gates to sustain model diligence across long-horizon inference.
ModLens is a vision plugin for DeepSeek Harness that gives text-only models sight by reading images pasted directly into chat, with zero-config setup and multiple vision engines.
On-demand vision for text-only DeepSeek Harness agents: built-in free keyless vision chain and 14 vision tools, routing image turns as tool calls to vision models with pixel fidelity, no Python needed, one-command install.
Give text-only models in DeepSeek Harness eyes, enabling image Q&A, long-screenshot OCR, UI restoration, and GUI visual tasks.