Why AI images can contain your prompt, and how to see it
Stable Diffusion, ComfyUI and other generators save the prompt, seed and settings inside the PNG. How to read them, and remove them before sharing.
· 5 min read
If you generate images with Stable Diffusion, ComfyUI or similar tools, the PNG files they save often contain your prompt, the negative prompt, the seed and every setting needed to recreate the image. Anyone you send the original file to can read it. Sometimes that is what you want; often it is not.
What different tools save
| Tool | Where | What |
|---|---|---|
| Stable Diffusion web UIs (AUTOMATIC1111, Forge) | parameters | Prompt, negative prompt, steps, sampler, CFG scale, seed, size and model hash |
| ComfyUI | prompt, workflow | The full node graph as JSON, including models, LoRAs, seeds and sometimes file names |
| NovelAI | Description, Comment | The prompt and a JSON block of generation settings |
| InvokeAI | invokeai_metadata | Prompt, model and generation settings |
These live in PNG text chunks (tEXt, zTXt or iTXt). When the same tools save JPEG or WebP, they may put the settings in the EXIF UserComment field instead.
Why it can matter
- Your prompt and settings are work you may not want to give away.
- A seed and settings let anyone reproduce the image exactly.
- Workflows can include the names of private models, LoRAs and local file paths.
How to see the prompt in a PNG
Drop the file into Inspect. The PNG text group lists every keyword and its full value, with AI-related fields highlighted. Check shows the prompt it found alongside its provenance verdict.
How to remove it
Open Clean and add the images. With Remove everything selected, the text chunks are removed along with all other metadata. To keep other blocks, choose Custom and leave only Text and comments selected. PNG chunks carry their own checksums, so removing whole chunks is lossless: the image data is unchanged.