Introduction¶
The design philosophy of DeepSeek Harness (DSH) is “everything is a plugin.” Developers can extend an Agent’s capabilities through plugins. In multimodal scenarios, image-generation and video-generation capabilities are often scattered across different scripts or tools, and configuring multiple providers, managing keys, and polling asynchronous tasks can be cumbersome. The dsh-media-gen plugin consolidates these two types of capabilities and provides a unified provider catalog, settings panel, gallery view, and CLI interface, allowing the Agent to directly invoke image-generation and video-generation tools.
What Is This¶
This is a DSH plugin maintained by EmberwingAviation. It forked and merged the dsh-image-generation and dsh-video-generation projects, unifying the interfaces for image and video generation. The plugin is licensed under MIT, supports multiple providers such as OpenAI, DashScope, Volcengine, and Google Veo, manages API keys through the DSH credential service, and does not write keys into configuration files.
Core Capabilities¶
The plugin provides three core tools, all available in newly created sessions:
image_generate: Handles text-to-image and reference-image editing. It supports passing 1–5 images from the conversation as references. Results are stored in the workspacegenerate/image/directory and displayed inline in the conversation.generate_video: Handles text-to-video. It uses an asynchronous task plus automatic polling mechanism. Results are stored in the workspacegenerate/video/directory, can be played in the conversation, and are included in the video gallery.animate_image: Handles image-to-video. The source image can be a workspace file, a conversation attachment, or “the latest image in the conversation.”
The plugin supports multiple providers, including OpenAI, DashScope (Tongyi Wanxiang), Volcengine (Seedance), and Google Veo. The tools include an automatic exponential backoff retry mechanism (default initial delay of ≥60 seconds, up to 3 retries) to handle upstream rate limits or load saturation.
Installation and Enablement¶
To install the plugin, use the DSH client command:
dsh plugin --profile desktop add dsh-media-gen
After installation, you must fully restart DSH to load the plugin. The tools are only available in new sessions because they are snapshotted when a session is created. Therefore, after installing the plugin, open a new conversation window.
Typical Usage¶
In a newly created session, simply state your request to the Agent to invoke the corresponding tool:
- Text-to-image:
“Draw a humpback whale leaping above the sea surface in the late night, in a cinematic movie-poster style.”
- Image-to-video:
“Turn this image into a 5-second video.”
- Text-to-video:
“Generate a video: a cyberpunk city on a rainy night, with the camera slowly pushing forward.”
Use Cases and Notes¶
- Unified management: Suitable for developers who need to use multiple providers simultaneously, such as API relays or official APIs. You can switch models and manage keys through a unified settings panel.
- Asynchronous processing: Video-generation tasks can take a long time. The plugin automatically polls task status, so developers do not need to manually track task IDs.
- Artifact storage: All generated artifacts are stored under the workspace
generate/directory, making them easier to process or reference later. - Permissions and licensing: The plugin is forked from an MIT-licensed project. Please comply with the MIT License terms when using it. It is recommended to review the source code and licensing notices before use.
Conclusion¶
dsh-media-gen encapsulates image-generation and video-generation capabilities into a unified toolset. Combined with DSH’s credential service and gallery features, it reduces the configuration cost of multimodal development. See the GitHub repository for plugin details and source code.
