@d-ai/openclaw

OpenClaw plugin that exposes D.AI image generation, face swapping, Grok image/video generation, video, and music generation through the native image_generate, video_generate, and music_generate tools, and registers the public D.AI user MCP while the plugin is enabled.

Install

OpenClaw 2026.9.3+ asks for capability consent on third-party plugins. The interactive second [y/N] can lose stdin and roll the install back (Plugin not found: dai). Pass the flag:

openclaw plugins install clawhub:@d-ai/openclaw --accept-capabilities
openclaw plugins enable dai --accept-capabilities

Local checkout:

openclaw plugins install --link ./sdks/openclaw --accept-capabilities

After install, open Control UI → Config → Plugins → D.AI and paste the API key (ddsk_...). That writes plugins.entries.dai.config.apiKey. You can also set DAI_API_KEY, or write the same config by hand:

{
  plugins: {
    entries: {
      dai: {
        enabled: true,
        config: {
          apiKey: "ddsk_...",
          // baseUrl: "https://dai.itbug.shop"
        }
      }
    }
  }
}

Default API host is https://dai.itbug.shop. Enabling the plugin also registers MCP server dai at https://mcp.itbug.shop/mcp with the same key. Optional mcpUrl is for private MCP hosts. Restart the Gateway after changing the key. Chat/TUI turns see the MCP tools; cron/heartbeat/subagent runs do not.

Models

Ask the agent to generate media, or pass a model ref. The native providers cover every public image/video model and their media-producing operations; the original generation functions remain compatible.

| Surface | Default | Examples | |---|---|---| | Image | dai/gpt-image-2.5-flare | OpenAI (gpt-image-2, gpt-4o-image), Gemini (gemini-3.1-flash-image), Novel (nai-diffusion-4-5-full), Flux (flux-2-max), Grok (grok-image), Midjourney (midjourney) | | Face swap | — | dai/swapface | | Video | dai/veo-3.1-quality | Gemini (omni-flash), Flux (flux-3-video), Grok (grok-imagine-video-1.5), Midjourney (fast), HappyHorse (happyhorse-1.1-r2v), Wan (wan3.0-video) | | Music | dai/chirp-v5-5 | dai/chirp-v5 |

Grok Image (dai/grok-image) supports text-to-image and image edits with one or more reference images. It accepts aspectRatio values 1:1, 16:9, 9:16, 3:2, and 2:3; output resolution is fixed by D.AI. Grok Imagine Video 1.5 (dai/grok-imagine-video-1.5) requires exactly one public or local reference image and supports 1–15 seconds and 480p, 720p, or 1080p; it does not support text-to-video or video-to-video. Both Grok surfaces are asynchronous and the native tools wait for completion.

dai/midjourney supports imagine, action, blend, and inpaint through operation; describe and seed return JSON and therefore remain available through the D.AI user MCP. dai/fast supports Midjourney video submit and operation: "extend". HappyHorse accepts its backend input/parameters, and Wan accepts the same fields for frame, reference, link, or file inputs.

dai/gpt-image-2.5 is GPT Image 2.5（经济·非官方）: explicitly select it for the lower-cost unofficial async channel. It generates one image only (count omitted or 1); size, quality, aspectRatio, outputFormat, and background are unsupported and rejected. The default remains official dai/gpt-image-2.5-flare; official flare/sunburst requests never switch to economy automatically. flux1-schnell is image-to-image only. Face swaps use dai/swapface; provide named targetImage and swapImage fields when supported by the host, or exactly two image references in target/source order. Spend is billed to the D.AI wallet.

Skill

Enabling the plugin loads dai-media, which tells the agent when to pick a dai/... model. Workspace skills with the same name override it.