Preface

DeepSeek Harness (DSH)’s conversation interface excels at text reasoning and tool orchestration, but Harness itself does not directly provide image generation capabilities. To have the model generate images from prompts or perform stylized edits on existing images within the same session, a common approach is to switch to a standalone image generation service and then manually paste the results back into the conversation—a fragmented process that also hinders intelligent agents from invoking steps sequentially.

Below is an introduction to the community plugin dsh-image2-draw (maintainer JuneLearn). It calls the gpt-image-2 model through a third-party OpenAI Images-compatible relay interface, enabling text-to-image and image-to-image generation within DSH Web, and displays the results directly in the conversation. The plugin is currently categorized under “Model Inference” in the SkillHub directory, has approximately 10 stars on GitHub, and is licensed under MIT (package.json version 0.1.0).

What It Is

dsh-image2-draw is an Image2 generation plugin for the DSH Web configuration profile. After installation, an Image2 Draw configuration card will appear under Settings > Plugins > Configurable plugins; you only need to fill in the baseURL and API Key provided by the relay service, with the default model being gpt-image-2.

The plugin is not tied to a specific relay provider: the README indicates that any relay implementing the OpenAI Images protocol can be chosen. Users should evaluate pricing, stability, and privacy terms before connecting. The baseURL can be a short path (e.g., https://example.com/v1), and the plugin will automatically append /images/generations; the image editing endpoint defaults to /images/edits but can be explicitly overridden via editURL.

Core Features

The following capabilities are verified based on the GitHub README and package.json.

Configuration & Credentials

  • Only requires baseURL and API Key; models, editing endpoints, timeouts, etc., can be adjusted in the settings page.
  • The API Key is only written to DSH credentials, not into regular settings documents, and is not returned via the plugin state interface.

Text-to-Image: image2-generate

  • Can generate 1–8 images consecutively in a single call, with upstream requests sent sequentially.
  • Supports adaptive portrait, landscape, and square sizes, as well as validated custom dimensions.

Image Editing: image2-edit

  • Accepts 1–8 reference images in PNG, JPEG, or WebP format; each image must not exceed 4MB, and the total must not exceed 32MB.
  • Format is detected by file magic numbers, not file extensions; relative paths are resolved from the current session’s working directory.

In-Session Display & Saving

  • Generation results are displayed in the conversation via dedicated tool cards, supporting thumbnails, enlargement, and Save As.
  • When the client card parses image attachments, it does not inject image content into the context of DeepSeek models that do not support visual input.
  • Images are also saved to the outputs/image2/ directory under the current session’s working directory; files with duplicate names are automatically numbered and do not overwrite existing files.

Request & Security Boundaries

  • Default timeout is 180 seconds, with a configurable range of 1–3600 seconds.
  • Validates settings writes, response sizes, timeouts, and input images.
  • HTTP 524 errors and timeout failures are not automatically retried to avoid duplicate billing if upstream has already generated the image.

Specific capabilities (e.g., gpt-image-2, editing, custom sizes, and quality tiers) depend on the chosen relay’s implementation; if an HTTP 400 or 404 is returned, you should refer to the service provider’s model name and Images endpoint documentation for troubleshooting.

Installation & Enabling

Environment Requirements

Before installation, ensure the following are available on your system:

  • Node.js: DSH currently supports 22.19.x or 24 and above; Node.js 24 LTS is recommended. It includes npm and npx.
  • Git: Used to pull the plugin from GitHub.
  • pnpm: dsh plugin uses pnpm to install or remove plugins in the profile directory; you can first run corepack enable.
  • Network access to registry.npmjs.org and github.com.

The plugin runs with the permissions of the current dsh process; it is recommended to read the repository source code and MIT license before installation to ensure that the storage of the relay API Key and outgoing requests comply with your security policies.

Method One: npx (For Regular Users)

No need to clone the Harness source code or globally install dsh:

npx --yes -p @deepseek-ai/dsh dsh plugin --profile web add github:JuneLearn/dsh-image2-draw

After installation, start the Web:

npx --yes -p @deepseek-ai/dsh dsh web

Method Two: Harness Source Code + pnpm (For Developers)

If you have cloned deepseek-harness and are developing in the source root directory:

cd D:\deepseek-harness
pnpm install
pnpm dsh plugin --profile web add github:JuneLearn/dsh-image2-draw
pnpm dsh web

The Web interface listens on http://127.0.0.1:3080 by default; if the port is occupied, another port will be used. To upgrade, simply re-run the corresponding add command; there is no need to manually maintain profile patches. Uninstall example (npx):

npx --yes -p @deepseek-ai/dsh dsh plugin --profile web remove dsh-image2-draw

After restarting dsh web, the Image2 Draw card and related tools will be removed.

Typical Usage

  1. Open Settings > Plugins > Configurable plugins > Image2 Draw.
  2. Enter the API Key provided by the relay service.
  3. Enter the relay endpoint, e.g., https://example.com/v1.
  4. Adjust the model, editing endpoint, or timeout as needed; the default values are usable in most scenarios.
  5. Click Save and wait for the Saved status to appear.
  6. Start a new session and have the model call image2-generate for text-to-image generation; or provide a reference image path and call image2-edit.

Text-to-image example prompt:

Call image2-generate to create a portrait cinematic poster of a future city at high quality.

Image editing example prompt:

Call image2-edit with D:\images\room.png and restyle the room with light Japanese wood while preserving the layout.

Use Cases & Considerations

Who It’s For

  • Developers who are already orchestrating intelligent agents in DSH Web and want to complete “description → image generation → continued reasoning” within the same conversation.
  • Teams that have or plan to integrate with an OpenAI Images-compatible relay and have upstream support for gpt-image-2 or equivalent model names.

Things to Know Before Use

  • Image generation quality, billing, and data retention are determined by the third-party relay; the plugin itself does not host the model.
  • Reference images and generated images have size and quantity limits; be mindful of disk usage in the session directory for large-scale image generation.
  • The community directory SkillHub is an independent site with no official affiliation with DeepSeek / High-Flyer; plugin lists and GitHub repositories are published by the maintainers themselves.

The author’s DSH Reasoning Settings (dsh-reasoning-settings) is used to configure reasoning levels for third-party models in Harness and can be used alongside this plugin, though they have no hard dependency on each other.

Links