Stable Diffusion Image Detector Free

Stable Diffusion is unusually easy to detect when you have the original file, because the most popular apps save the full recipe inside the image.

What Stable Diffusion apps save

• Automatic1111 and Forge: a PNG text field called "parameters" with the prompt, negative prompt, steps, sampler, CFG scale, seed, size and model name.
• ComfyUI: "prompt" and "workflow" fields holding the whole node graph, which can be dragged back into ComfyUI to recreate the image.
• InvokeAI and others: their own metadata fields.

These are saved in PNG files by default. Converting to JPG, editing, or uploading to most sites usually removes them.

How to read them for free

• In Chrome: AI Image Detector (made by Patrick Bushe, who runs this site) checks images for these fields when you right-click them. It looks for keys such as "parameters," "prompt" and "workflow," and for telltale text like "Steps:" with "Sampler:", and shows the text it found.
• In Automatic1111: the PNG Info tab shows the parameters of any image you drop in.
• ExifTool: lists all PNG text chunks on your computer.

When the metadata is gone

Images shared on social media, Discord or most image hosts have usually been stripped. Then you're left with:

• Visual detectors such as Hive, which often recognize Stable Diffusion images and sometimes guess the model.
• Reverse search to find the image on Civitai, AI art communities or the creator's page.
• Visual clues: common Stable Diffusion sizes (512 or 1024 pixels on a side), and artifacts in hands, text and backgrounds.

A note on fine-tuned models

Stable Diffusion is open-source and widely fine-tuned, so its images vary enormously in style. Visual detectors trained on the base models may miss outputs from custom ones.

See Midjourney metadata and metadata fields that hint at AI.