Prompt Metadata in Downloaded Images
The NanoGPT web app can write an image’s prompt and generation settings into the file when you download it. Tools such as SD Prompt Reader and Civitai can then read them back, and you keep a record of how each image was made. This setting is off by default. Anyone you share a file with can read the prompt in it.Turn It On
- Open Settings in the web app.
- Go to Privacy & analytics.
- Turn on Embed prompt in downloaded images.
What Gets Written
NanoGPT writes the text in the Automatic1111parameters format:
- The first line is the image prompt.
- A
Negative prompt:line appears only when you set a negative prompt. - The settings line always starts with
Steps:, because readers look for it there.Steps: 0means the model has no step count NanoGPT knows. - Sampler, CFG scale and seed are included when the request set them. The seed is written only when the request made a single image.
- Size is read from the image itself. Model is the public model ID.
Supported File Types
The image’s pixels are not re-encoded, so quality stays exactly the same. Other formats, and images signed with C2PA content credentials, download unchanged. The text would make a signed image look tampered with.
Where It Applies
With the setting on, the prompt is written when you save a generated image with NanoGPT’s own download controls:- the download button in the image viewer, in chat and in Image Studio
- downloads from your gallery
- gallery ZIP exports (images over 25 MB in a ZIP export are left unchanged)
- images made by the image generation tool in chat
Not Covered Yet
- Saving with your browser’s own Save image as (right-click or long-press). Use the download button instead.
- Images returned by the API. They come back exactly as the model made them.
- Videos.