Which images contain a prompt?
Anything saved directly by a local generator: Automatic1111, Forge, ComfyUI, InvokeAI, SwarmUI, Fooocus, SD.Next, EasyDiffusion, NovelAI downloads. Images downloaded from Civitai usually keep it; images passed through Discord (mobile), Instagram, X, WhatsApp or screenshots do not — see why an image has no metadata. Midjourney and DALL·E web exports contain little or no generation data.
What the reader shows
- Prompt / negative prompt — separated and copyable.
- Seed, steps, sampler, scheduler, CFG — the values you need to reproduce it.
- Model / checkpoint, VAE, LoRAs — with hashes when the tool wrote them.
- Raw chunks — every text chunk, pretty-printed if it is JSON, so nothing is hidden.
Per-tool notes
A1111/Forge: single parameters string — details. ComfyUI: prompts live inside CLIPTextEncode nodes of the prompt graph — details. InvokeAI: invokeai_metadata JSON — details. SwarmUI: parameters JSON with sui_image_params. Fooocus: parameters / fooocus_scheme JSON. NovelAI: Comment JSON with prompt and uc.
Privacy
Reading happens in your browser with the File API; the prompt is visible before any network request. See how it works.
Frequently asked questions
How can I see the prompt used for an AI image?
Upload the original file to pngmeta.app. If the generator embedded metadata (most local Stable Diffusion tools do), the prompt, negative prompt and settings appear immediately.
Can I read the prompt from a Midjourney image?
Usually not. Midjourney web downloads carry only basic metadata (sometimes the job URL in XMP/EXIF), not the full prompt. Discord downloads may include a description tag; PNGMeta shows whatever is present.
Does the prompt reader work on phone?
Yes — it is a normal web page. Tap the upload button and pick a file. Make sure the file is the original, not a "photo" re-share.