shanliuling/dsh-image-gen

What it solves

DeepSeek Harness (DSH) lacked a cohesive image generation capability. This plugin fills that gap by providing a full suite for creating and editing images entirely within the DSH interface, from quick chat-based text-to-image to advanced batch creation, comparison, and management.

How it works

Installed as a DSH plugin, it adds an image generation configuration panel where you select a provider (Google Gemini, OpenAI, Seedream, DashScope, Grok, GLM-Image, or local ComfyUI), enter an API key (or log in with a ChatGPT/Grok/Google subscription to bypass keys), and optionally adjust endpoints, models, and aspect ratios. It offers five entry points: chat for conversational generation/editing, an infinite canvas for visual drafts and iterative refinement, a Studio workbench for controlled batch generation with multiple reference images, an inspiration library with 500+ prompt examples, and a gallery for managing saved images. ComfyUI workflows can be imported and invoked by name from chat, supporting private local generation.

Who it's for

Users of DeepSeek Harness who want to generate or edit images without leaving the platform, whether they have API keys, existing ChatGPT/Grok/Google subscriptions, or a local ComfyUI setup. It suits casual creators looking for quick results as well as power users who need parameter-level control and batch workflows.

Highlights

• Chat-native generation: Describe images in natural language, upload reference images, and iteratively edit with version switching on each card.
• AI canvas: Draw drafts or drag in reference images, then use dialogue to turn sketches into finished works with continuous revision.
• Studio batch creation: Generate multiple candidate images with fine control over provider, model, aspect ratio, and resolution, saving only the best results.
• Multi-model comparison: Run the same prompt across multiple providers side-by-side in one canvas.
• 500+ prompt inspiration library: Browse, filter, and copy prompts; all locally cached to avoid consuming generation credits.
• Subscription login support: Use existing ChatGPT, Grok, or Google accounts without API keys for both text-to-image and image editing.
• Local ComfyUI integration: Import multiple named workflows, use placeholders, and let the agent pick workflows for private GPU-based generation.
• Privacy-oriented: API keys and subscription tokens are stored securely via DSH Credentials, with workspace-scoped file saving and local metadata for gallery and favorites.