shanliuling/dsh-image-gen

What it does

dsh-image-gen is a native image-creation plugin for DeepSeek Harness (DSH). Rather than adding a single image button to the chat box, it teaches the agent to plan and generate images itself: it can turn one request into a batch of up to 10 images, read sketches and selections on an infinite canvas, and look up matching style templates in a built-in prompt library of 500+ examples.

It supports text-to-image, image-to-image with multiple references, character consistency across scenes, GIF generation on supported providers, multi-model side-by-side comparison, and a parameter-driven Studio workspace. A gallery, favorites, and workspace-scoped storage handle the generated results.

Quick start

Requirements: a stable DSH release and Node.js ^22.19.0 or >= 24.0.0. Run the install command in your DSH project root:

pnpm dsh plugin --profile web add dsh-image-gen@latest

If dsh is installed globally, use dsh plugin --profile web add dsh-image-gen@latest; you can also install from Git or a local clone. The package is published on npm, and the installation section lists the alternatives.

After restarting DSH, open the plugin panel via the sidebar Plugins icon, find dsh-image-gen under Installed, and open the image-generation settings form. Choose a provider and enter its API key, or expand a subscription provider (ChatGPT, Grok, Google) and click Login to authorize in the browser without an API key. Use Test connection to verify, Fetch models to list available image models, and for ComfyUI, enter a DSH-reachable server address and import an API-format workflow JSON.

Then simply describe the image in the chat, for example:

Draw a cyberpunk cat on a neon rainy street, cinematic lighting, 16:9.

For a whole set, state the count and a shared style in one message. Open the Gallery entry at the top of the session to reach the gallery, Studio, inspiration library, and favorites. The provider configuration reference covers default models, endpoints, aspect ratios, and size tables.

Constraints

• The plugin runs inside DSH; it is not a standalone application.
• Studio and multi-model comparison only support cloud providers, not ComfyUI.
• GIF output is limited to OpenAI and OpenAI-compatible channels that support GIF; the ChatGPT subscription image interface does not currently support GIF.
• GLM-Image does not support image-to-image upstream, and Grok image editing is limited.
• API keys and subscription tokens are stored via DSH Credentials and not echoed in settings.
• Prompts and reference images are sent to the selected provider, and upstream image downloads reuse the host's network/proxy configuration; the README states there is no complete SSRF target restriction.
• Conversation batches cap at 10 images per call; images are generated one by one so a single failure does not stop the rest.
• Licensed under Apache-2.0.

HotRepo's take

dsh-image-gen recorded 65 new stars in a single TypeScript daily ranking snapshot on 2026-11-12, when it ranked 15th in that language list with 1,352 stars and 107 forks. There is no recorded all-language daily rank and no star-curve sampling, so this is one day's observation rather than a trend. What stands out technically is the depth of provider coverage — Gemini, OpenAI and compatible relays, Seedream, DashScope, Grok, GLM-Image, local ComfyUI, plus subscription login without API keys — combined with agent-driven batch generation and canvas-aware editing, which goes beyond a thin image button wrapper.