Preface¶

DeepSeek Harness (DSH) integrates Agents, tool invocation, and conversational flows, making coding, research, and task execution seamless. However, many developers often find that when they want an Agent to “draw a picture,” they still need to switch to Midjourney, DALL·E, or various cloud platform consoles—copy prompts, wait for generation, and paste the image back into the chat, disrupting the workflow and risking lost context.

The community has filled this gap. dsh-image-gen is an open-source DSH image generation plugin maintained by shanliuling, categorized under “Model Inference” in the SkillHub plugin directory. Its GitHub repository description reads: “Generate images directly in DeepSeek Harness chats.” Its goal is straightforward: to enable ChatGPT-like functionality in the DSH chat window, where users describe the scene in natural language, and the Agent automatically calls the generate_image tool, embedding the image directly into the current session.

This article is based on cross-verification of the plugin’s GitHub README, package.json, and the SkillHub Directory API. It introduces its capabilities, installation methods, and typical usage. Note that SkillHub (skillhub.cn/plugins) is a community-maintained DSH plugin directory with no official affiliation to DeepSeek or High-Flyer. DSH itself adheres to an “everything is a plugin” philosophy, with most capabilities contributed by the community.

What This Is¶

dsh-image-gen is an open-source plugin (MIT license) for DeepSeek Harness Web, distributed via the npm package name dsh-image-gen. It registers a generate_image tool for the DSH Agent, forwards text prompts to external image APIs, embeds returned images into the conversational flow, and integrates with DSH’s Attachment/Conversation system.

In simple terms, it solves the problem of “DSH can write code and invoke tools but lacks a native, configurable multimodal image generation pipeline.” Users bring their own API keys (BYOK) from various cloud platforms, select the Provider and model in the settings page, and use it immediately. Keys are managed by DSH’s credentials service with write-protection isolation, eliminating the need to embed them in project source code.

Core Features and Highlights¶

According to the official README, the plugin primarily offers the following capabilities:

1. Direct Image Generation in Conversations: Describe the desired image in natural language in the chat box, and the Agent automatically calls the tool, eliminating manual website switching or prompt copying.
2. History Image Gallery: A “Gallery” tab in the session header aggregates all historically generated images, supporting keyword search, vendor filtering, single-image deletion (with accidental-click confirmation), copying, and downloading.
3. Interactive Image Operations: Supports fullscreen preview, clipboard copying, local download, and opening in a new tab.
4. Multi-Provider Support: Covers Google Gemini, OpenAI Images, OpenAI Compatible API, ByteDance Seedream/Volcano Ark, and Alibaba Cloud DashScope (Tongyi Wanxiang/Qwen-Image). Providers, models, and endpoints are customizable in the Web settings.
5. Session-Persistent Images: Generated results integrate with DSH’s attachment system, ensuring historical images remain accessible when reopening sessions.
6. Auto-Save to Workspace: By default, images are saved to the current session workspace (can be disabled or customized with subdirectories), with the tool returning the file’s absolute path.
7. Native Settings Interface: Configuration is completed in DSH Web’s Settings → Plugins → Image generation, avoiding manual config file edits.

Default models and endpoints for each Provider (from the README, subject to the settings page):

In the SkillHub directory, this plugin is marked as verified (installation validation passed), with approximately 150+ GitHub stars (directory sync data shows 153 stars and 7 forks; real-time GitHub data may be higher).

Installation and Activation¶

Execute the following commands in the DeepSeek Harness project root directory. The official README recommends installing the latest version via npm with the --profile web flag (this plugin is for the Web client):

# Recommended: Install or upgrade to the latest version
pnpm dsh plugin --profile web add dsh-image-gen@latest

# If dsh is installed globally, you can omit pnpm:
dsh plugin --profile web add dsh-image-gen@latest

For direct installation from GitHub or local debugging, the README provides two additional methods:

# Install the latest code from the GitHub repository
pnpm dsh plugin --profile web add git+https://github.com/shanliuling/dsh-image-gen.git

# Local clone and development installation
git clone https://github.com/shanliuling/dsh-image-gen.git
pnpm dsh plugin --profile web add ./dsh-image-gen

After installation, open the DSH Web page (default http://localhost:3080), go to Settings → Plugins → Image generation, select the Provider, and enter the API Key. Optionally enable “Save to Workspace” and customize the subdirectory, then click save to activate.

Typical Usage Examples¶

After configuring the Key, simply describe the desired image in the conversation. Example prompts from the README:

帮我画一张雨夜霓虹街头的赛博朋克猫咪。

Or:

生成一张极简主义的现代建筑客厅插画。

The Agent will call generate_image, and the image will be displayed inline in the current conversation flow. To review past works, click the “Gallery” tab at the top of the session to browse, search, and manage all generated records collectively.

If you prefer the Agent to handle installation, you can directly say:

帮我安装生图插件，执行命令：pnpm dsh plugin --profile web add dsh-image-gen@latest

Applicable Scenarios and Notes¶

Who is this suitable for:

• Developers already using DSH Web for Agent development or daily conversations, seeking to add “text-to-image” capabilities and reduce context switching;
• Users with existing API keys for Gemini, OpenAI, Volcano Ark, or Tongyi Wanxiang, preferring BYOK integration;
• Scenarios requiring iterative image generation within the same session with preserved historical records.

Important notes before use:

1. The plugin runs with the current dsh process permissions. Before installation, review the GitHub source code and MIT license to ensure the behavior aligns with your security policy.
2. Image generation costs are billed by your selected Provider’s API. The plugin itself does not charge fees; safeguard your API Key and monitor usage and quotas on each platform.
3. --profile web must be configured. The plugin’s dsh.client.platform is web, which may differ from installation methods for TUI or other profiles. Refer to the README.
4. Models and endpoints may change with vendor updates. The default values in the table are from the README; actual available models depend on each platform’s documentation and the plugin settings page.
5. In domestic network environments, accessing overseas APIs like Google/OpenAI may require additional network configuration; domestic services like Volcano Ark and Tongyi Wanxiang typically offer easier direct connectivity.

Summary¶

dsh-image-gen brings the ChatGPT-style “image generation in conversation” experience to DeepSeek Harness: with one installation command, entering the API Key in settings, and describing the scene in the chat box, the Agent can invoke generate_image and retain the result in the session, complete with gallery management and workspace saving. For developers already working within the DSH ecosystem on Agent workflows, this is a model inference plugin worth trying.

• Directory page: https://www.skillhub.cn/plugins/shanliuling/dsh-image-gen
• GitHub: https://github.com/shanliuling/dsh-image-gen