When running some tests with Nanobanana in ComfyUI I asked it to make a picture of a grey cat drinking from a hosepipe on a slate roof. This is because the grey cat asks me to turn the tap on for her every morning. What alarmed me were the slate roof tiles in the result. Obviously I can't prove it, but I'm reasonably certain that these have been scraped from the multiple pictures of my own garden roof terrace in Google Photos.
Google emphatically deny that they have trained their AI on Google Photos. They say, "We don't train any generative AI models outside of Google Photos with your personal data in Google Photos." However, the wording here FEELS slippery. The phrase "with your personal data" might imply that the photos could be "anonymised" - stripped of geographical information etc. Or perhaps when they have trained the models inside Google Photos they then take the models and use them elsewhere? That would also satisfy that wording.
They certainly leapt far ahead of the pack with Nanobanana and Veo, and it's tempting to conclude that they allowed themselves to train their models on Google Photos. I'm sure that if I went through the fine print I would find that by using this service (even though I have paid for it for years with Google Workspace...), I had surrendered my rights to this data for their purposes.
Certainly there's an irony to me posting this on Google's Blogger (the original Sick Veg blog posts were all on WordPress so that's not the leak). But I would argue that what one posts on Blogger and YouTube is by definition for public consumption. It's a very tiny drama, but one which is so personal to me that it feels like an infringement on what is sacred
 |
| AI Cat. |
 |
| Real cats. |