Post ASOJSrO2qt77CikCqe by bitprophet@social.coop
 (DIR) More posts by bitprophet@social.coop
 (DIR) Post #ASNuXoDxXZWPzOI2E4 by simon@fedi.simonwillison.net
       2023-02-05T23:31:49Z
       
       0 likes, 2 repeats
       
       There's a tool doing the rounds called "Stable Attribution" which claims to be able to identify the images in the Stable Diffusion training set that were used to generate an imageIt's extremely misleading. All it does is show you images in the Stable Diffusion training set that are visually similar. You can prove this to yourself by running it against a photo you have taken yourself.I took this photo. It was not generated by AI at all.
       
 (DIR) Post #ASNwHesuhEzAIYs9JI by minimaxir@sigmoid.social
       2023-02-05T23:50:46Z
       
       0 likes, 0 repeats
       
       @simon This seems like https://haveibeentrained.com with counterproductive pretentiousness.
       
 (DIR) Post #ASNweVAaywUYpS2GSe by hierarchon@inherently.digital
       2023-02-05T23:54:42Z
       
       0 likes, 0 repeats
       
       @simon conversely, here are three ai-generated images I made myself for which it completely fails to find anything similar!
       
 (DIR) Post #ASNwrtwUBpyHQKpELA by Albot@sigmoid.social
       2023-02-05T23:56:11Z
       
       0 likes, 0 repeats
       
       @simon urggh when will people learn how these models work
       
 (DIR) Post #ASO0xdkjh8FelcJzAu by ummjackson@mastodon.social
       2023-02-06T00:43:30Z
       
       0 likes, 0 repeats
       
       @simon Grifters are going to grift, unfortunately :(
       
 (DIR) Post #ASO1YRWS86kbY9BnjU by Stevenpotato@mastodon.social
       2023-02-06T00:48:23Z
       
       0 likes, 0 repeats
       
       @simon Good enough to impress a jury.
       
 (DIR) Post #ASOJSrO2qt77CikCqe by bitprophet@social.coop
       2023-02-06T04:08:23Z
       
       0 likes, 0 repeats
       
       @simon Sounds like that’s not a useful test: https://twitter.com/atroyn/status/1622366388953423872
       
 (DIR) Post #ASOT1J1lUSnsMap4ng by simon@fedi.simonwillison.net
       2023-02-06T05:58:07Z
       
       0 likes, 0 repeats
       
       @bitprophet I think they are being very disingenuousI know how image Diffusion models work. They don't take a dozen images and merge them together, which is what this tool impliesThe copy that says "these human-made source images were used by AI to generate this image" is plain wrong, and uploading your own non-AI image is a great way to illustrate that
       
 (DIR) Post #ASOpzODzUhjqqgb3Uu by segv11@mastodon.online
       2023-02-06T10:14:46Z
       
       0 likes, 0 repeats
       
       @simon It goes  beyond misleading. It's an outright lie pandering to emotion. They don't even reflect the square crop for the training data.Despite their definitive claim, my face is not generated by #stablediffusion. I also fed it four complex SD-generated images in a row, and only the last one found any "matches," and I don't see any similarity.I note they are owned by a company called Chroma, which seems to be developing a competing business plan. The whole website is astroturf.
       
 (DIR) Post #ASOtkvR48OMjK74RZA by hayley@social.applied-langua.ge
       2023-02-06T10:59:23.320369Z
       
       0 likes, 1 repeats
       
       @simon how
       
 (DIR) Post #ASPWhzSibT0h9hR2Qq by fernand@sigmoid.social
       2023-02-06T18:13:43Z
       
       0 likes, 0 repeats
       
       @simon @bitprophet I’m not sure what you mean, the reasoning behind this is correct. They use the open source LAION datasets which were used to train SD v1/2 and find the closest images to the uploaded image (has to be generated by SD v1/2) in terms of CLIP embeddings.