Image Editing Skill

Edit and generate images using AI models via OpenRouter.

Capabilities

• Image editing: Modify existing images (change colors, rearrange objects, add/remove elements)
• Image generation: Create new images from text descriptions
• Style transfer: Apply artistic styles to photos

Supported Models (via OpenRouter)

API Format

Request (to OpenRouter)

{
  "model": "google/gemini-2.5-flash-image",
  "messages": [{
    "role": "user",
    "content": [
      {"type": "text", "text": "Move the couch to the left wall and add a plant in the corner"},
      {"type": "image_url", "image_url": {"url": "data:image/png;base64,..."}}
    ]
  }]
}

Response (from OpenRouter)

{
  "choices": [{
    "message": {
      "role": "assistant",
      "content": "Here's your edited image:",
      "images": [{
        "type": "image_url",
        "image_url": {"url": "data:image/png;base64,..."}
      }]
    }
  }]
}

Implementation Notes

Telegram Integration

1. Receiving images: Use Media.downloadPhoto to get image bytes
2. Encoding: Convert to base64 data URI for OpenRouter
3. Sending results: Use Telegram’s sendPhoto API with the base64-decoded bytes

Telegram sendPhoto API

POST /bot{token}/sendPhoto
Content-Type: multipart/form-data

chat_id: <int>
photo: <file bytes>
caption: <optional text>
message_thread_id: <optional, for topics>

Example Prompts for Furniture Rearrangement

Good prompts for interior editing:

• “Move the couch to face the window”
• “Rearrange the furniture so the TV is on the opposite wall”
• “Add a coffee table in front of the sofa”
• “Remove the bookshelf and add a floor lamp”
• “Change the wall color to light blue”

Limitations

• Image editing is generative, not pixel-perfect — results may vary
• Complex rearrangements work better with clear, well-lit photos
• Model may interpret ambiguous instructions creatively
• Large images are resized; originals aren’t preserved

Tool Definition

editImageTool :: Engine.Tool
editImageTool = Engine.Tool
  { Engine.toolName = "edit_image"
  , Engine.toolDescription = "Edit or modify an image. Provide the edit instructions and the image will be modified accordingly."
  , Engine.toolParameters = ...
  , Engine.toolExecute = \params -> do
      -- 1. Get image from context (most recent photo in conversation)
      -- 2. Call OpenRouter with image + prompt
      -- 3. Return edited image
  }

Environment Variables

• OPENROUTER_API_KEY: Required for OpenRouter API access (already configured for Ava)

See Also

• Omni/Agent/Telegram/Media.hs — Image download/upload utilities
• Omni/Agent/Provider.hs — Provider abstraction for API calls