dsh-image-gen

dsh-image-gen is a native image creation plugin for DeepSeek Harness (DSH). Rather than adding a standalone image generation button to the interface, it enables the DSH Agent itself to plan and execute image generation tasks through natural language conversation.

Core Capabilities

• Batch Image Generation via Chat: Users can request multiple images in a single message. The Agent breaks down the prompt, generates up to 10 images sequentially, and collects them into one reply. If one image fails, the others continue processing, and failed attempts are retried automatically.
• AI Canvas with Sketch Understanding: An infinite canvas allows users to draw rough sketches or place reference images. The Agent reads the canvas content and current selection area directly, enabling workflows like converting stick figures into finished illustrations or modifying only a circled region of an existing image without altering the rest.
• Prompt Library Integration: The plugin includes over 500 built-in prompts. The Agent can search this library to find appropriate style templates based on user requests before generating images.
• Multi-Model Comparison: Users can instruct the Agent to run the same prompt across different models concurrently and display the results side-by-side on the canvas for comparison.
• GIF Animation: When using supported models like GPT Image 2.6, users can generate looping GIF animations. Frame count (2–16) and duration (20–2000 ms) are configurable. GIFs play directly in chat cards with playback speed controls and frame-by-frame timeline navigation.
• Character Consistency: During image-to-image edits, the plugin instructs the model to retain the original subject while changing scenes, clothing, poses, or styles.
• Version Switching: Users can modify the original prompt to regenerate an image and switch between historical versions within the same card.

Supported Providers

The plugin supports multiple providers: Gemini, OpenAI / Compatible, Seedream, DashScope, Grok Imagine, GLM-Image, and local ComfyUI. Users with existing ChatGPT, Grok, or Google subscriptions can log in directly to use text-to-image and image-to-image features without purchasing separate API keys. For ComfyUI, users import API Format Workflow JSON files and switch between workflows via the UI.

Installation & Requirements

Requires DeepSeek Harness stable version and Node.js ^22.19.0 or >= 24.0.0. Installed via the command line using pnpm dsh plugin --profile web add dsh-image-gen@latest. Configuration is handled through the DSH plugin panel where users select providers, input API keys or log in, test connections, and pull available models.