HarryLi-7/dsh-vision

README excerpt

dsh-vision DSH (DeepSeek Harness) vision plugin: gives a text-only model two tools — describe image (image recognition) and generate image (image generation). Built for personal use: engines ride what you already have (your logged-in Codex / ChatGPT account) with a free Zhipu fallback, every engine parameter is editable in the harness Settings page, and all temporary data is delete-after-use (用完即焚). Tools Tool What it does Primary engine Fallback describe image Reads a local image, returns a detailed text description; only the text enters the conversation (image bytes are never stored in the session) Codex CLI ( gpt-5.6-luna ) Zhipu glm-4.6v-flash (free) generate image Generates an image from a prompt; result is saved as a conversation attachment and rendered inline in the chat Codex (GPT Image via codex imagegen skill) Zhipu cogview-3-flash (free) Why two engines - Codex — your own ChatGPT login, top-tier models ( gpt-5.6-luna and friends), no extra key, works with gemini-flash-latest -style model names swapped freely. Requires the Codex CLI to be installed and logged in. - Zhipu — free tiers ( glm-4.6v-flash vision, cogview-3-flash generation), China-direct, works without a proxy…

Category