reloadium.com

editImage

Edit an image using a text instruction, and show the result in the chat. Do NOT use this to turn several product photos into a carousel or video — call generateCarousel once with each slide's photo as imageUrl plus an imagePrompt (parallel, and composed to crop to every format); separate edits here are slow and not composed for cropping. Use for follow-ups like 'make it a night scene', 'add falling snow', 'remove the tree', 'change the style to a realistic photo' — anything that modifies an existing image. By default edits the most recently generated image in THIS conversation (from generateImage or a previous editImage call) — there's no way to edit a photo the user attached directly, and this can't reach an image from a different conversation. Pass `imageUrl` instead to edit a real photo from the web (e.g. one found via getBestImage) rather than something generated in this chat — fetched server-side, so it works regardless of the source site's CORS policy. If the instruction refers to a specific part of the image you haven't already identified precisely ('the item on the left', 'the third car'), call annotateImage first and pass its result as `annotations` so the edit is grounded to the right region instead of guessed. The model behind this makes one holistic edit per call, not a sequence of small ones — if the user wants several changes at once, describe them all together in a single instruction rather than calling this tool repeatedly for each one. Each call is a real, billed generation — don't call it speculatively.

Page: /

Input schema

{
  "type": "object",
  "properties": {
    "instruction": {
      "type": "string",
      "description": "What to change about the image. Combine multiple changes into one instruction rather than calling this tool repeatedly."
    },
    "imageUrl": {
      "type": "string",
      "description": "Optional: a public image URL to edit instead of the most recently generated image (e.g. one found via getBestImage)."
    },
    "annotations": {
      "type": "array",
      "description": "Optional: the exact 'annotations' array from a prior annotateImage call on this same image, to precisely constrain the edit to specific labeled region(s) — only pass the box(es) actually relevant to this instruction, not every box that was detected.",
      "items": {
        "type": "object",
        "properties": {
          "label": {
            "type": "string"
          },
          "box": {
            "type": "array",
            "items": {
              "type": "number"
            },
            "description": "[ymin, xmin, ymax, xmax], 0-1000 scale"
          }
        }
      }
    }
  },
  "required": [
    "instruction"
  ]
}