/edit_image

name: Edit Image
allowed-tools: Bash, Read
description: Edit existing images via Replicate using direct curl API calls

Edit Image

This command edits existing images based on provided edit prompts using Replicate's API directly via curl.

Prerequisites

• REPLICATE_API_TOKEN must be exported in the environment
• Requires base64 command for inline image encoding

Variables

DROPPED_FILE_PATH: [[FILE_PATH]] DROPPED_FILE_PATH_ARCHIVE: agentic_drop_zone/edit_images_zone/drop_zone_file_archive/ IMAGE_OUTPUT_DIR: agentic_drop_zone/edit_images_zone/image_output/<date_time>/

• This is the directory where all images will be saved
• The date_time is the current date and time in the format YYYY-MM-DD_HH-MM-SS

MODEL: google/nano-banana API_ENDPOINT: https://api.replicate.com/v1/models/google/nano-banana/predictions

Workflow

• Note: REPLICATE_API_TOKEN will be available, don't search for it and NEVER read the .env file directly.
• IMPORTANT: If for some reason you run into issues with authentication, abort immediately - do not continue with any image processing. Mention that the REPLICATE_API_TOKEN might be missing.
• First, read `DROPPED_FILE_PATH`.
• Create output directory: `IMAGE_OUTPUT_DIR/<date_time>/`
• IMPORTANT: For every image edit detailed in the `DROPPED_FILE_PATH` do the following:

<image-loop>

• Extract the source image path and edit prompt from the dropped file
• Use curl with inline base64 encoding to send image directly to Replicate API:

    curl -s -X POST \
      -H "Authorization: Bearer $REPLICATE_API_TOKEN" \
      -H "Content-Type: application/json" \
      -H "Prefer: wait" \
      -d '{
        "input": {
          "prompt": "YOUR_EDIT_PROMPT_HERE",
          "image_input": ["data:image/jpeg;base64,'$(base64 -i /path/to/source/image.jpg)'"],
          "output_format": "jpg"
        }
      }' \
      https://api.replicate.com/v1/models/google/nano-banana/predictions

• IMPORTANT: Replace `YOUR_EDIT_PROMPT_HERE` with the actual edit instruction
• IMPORTANT: Replace `/path/to/source/image.jpg` with the actual source image path which should be detailed in the `DROPPED_FILE_PATH`
• IMPORTANT: The base64 encoding happens inline using `$(base64 -i /path/to/image.jpg)`
• Parse JSON response to extract the `output` URL field for the generated image
• Response format: `{"id":"...", "output":"https://replicate.delivery/...jpg", "status":"succeeded"}`
• Extract URL with `jq -r '.output'`
• Save the executed edit prompts to `IMAGE_OUTPUT_DIR/<date_time>/edit_prompt_<concise_name_based_on_prompt>.txt`
• Include both the source image path and the exact edit prompt that was executed
• Download the edited image from the output URL: `curl -o IMAGE_OUTPUT_DIR/<date_time>/<MODEL_NAME_underscore_separated>_edited_<concise_name_based_on_prompt>.jpg "OUTPUT_URL"`
• Display:
• Source image path
• Edit prompt used
• Generation time (if available in response)
• File size of downloaded image
• Full path to saved edited image
• Replicate output URL for reference

</image-loop>

• After all images are edited, copy all original source images to the output directory for reference:
• For each source image that was processed, copy it with "original_" prefix
• Example: `cp /path/to/cat.jpg IMAGE_OUTPUT_DIR/<date_time>/original_cat.jpg`
• This allows easy before/after comparison in the same directory
• After copying originals, open the output directory: `open IMAGE_OUTPUT_DIR/<date_time>/`
• When you finish editing images, move the `DROPPED_FILE_PATH` into a `DROPPED_FILE_PATH_ARCHIVE` directory

Example Usage

# The REPLICATE_API_TOKEN is sourced from the .env file


# Execute image edit with inline base64 encoding
RESPONSE=$(curl -s -X POST \
  -H "Authorization: Bearer $REPLICATE_API_TOKEN" \
  -H "Content-Type: application/json" \
  -H "Prefer: wait" \
  -d '{
    "input": {
      "prompt": "Make the cat blue",
      "image_input": ["data:image/jpeg;base64,'$(base64 -i ./source_image.jpg)'"],
      "output_format": "jpg"
    }
  }' \
  https://api.replicate.com/v1/models/google/nano-banana/predictions)

# Extract output URL and download
OUTPUT_URL=$(echo "$RESPONSE" | jq -r '.output')
curl -o edited_image.jpg "$OUTPUT_URL"