dsh-image-gen

In DeepSeek Harness, autonomous agents frequently need to generate images, remove backgrounds, upscale resolution, or run quality checks, yet every provider uses a different API, queue polling often hits 429 rate limits, and post-processing pipelines must be stitched together by hand. dsh-image-gen collapses image creation, transformation, and inspection into a single plugin so agents can complete the full visual workflow through one consistent tool set.

The plugin spans eight generation backends, including FAL, Replicate, OpenAI-compatible endpoints, ChatGPT subscriptions, Grok subscriptions, local ComfyUI and Automatic1111, ByteDance Seedream, and Gemini Imagen 3, and exposes eight core tools: generate_image, remove_background, upscale_image, vectorize_image, blend_images, generate_image_pack, compare_images, and inspect_image_quality. Reliability features include exponential backoff with jitter, fast-fail error classification for client-side errors, deterministic hash caching for instant local returns, 64-multiple dimension snapping for VAE bucket compatibility, and style presets with tailored negative prompts to minimize wasted API calls.

If you are building autonomous agent workflows, need to switch between multiple generation services through one interface, rely on local ComfyUI or SD-WebUI for offline rendering, or want automated sharpness scoring and defect detection over batch outputs, this plugin serves as a unified foundation for the visual layer of your pipeline.