Old Photo Restorer

Problem

Old photos often combine scratches, fading, blur, tears, and compression artifacts. Manual retouching is slow and can easily turn a historical portrait into a modern headshot or an over-processed image. This skill treats restoring the real person as the first principle. It enhances clarity, reduces damage marks, and performs restrained colorization when needed, then returns a result image URL.

How It Works

The caller must provide at least one reference photo and fill in options based on the image, instead of writing long style prompts. Useful fields include:

• colorize: use true for black-and-white or sepia photos, and false for already colored photos
• damage: specify scratch, fade, blur, tear, or mixed
• restore_level: use light, standard, or strong, with standard as the default

The skill submits the request to the fixed model_key and image_edit capability, while the service applies its fixed recipe. The task is asynchronous: poll the task_id from pending to running until succeeded, then read the image from data.result.data[].url.

Boundaries And Notes

It is not suitable for generating images without a reference, face swapping, rejuvenation, or stylized cinematic output. If the user's text conflicts with the identity in the photo, the photo takes precedence. Paid calls may return 402; after successful payment, retry with the same JSON body. If the result is weak, adjust colorize, restore_level, or note first rather than rewriting long style prompts on the client.