A request of @Inmel. Hope you like it, mate!<a rel="ugc" href="http://www.veronika-zemanova.info/Index/default.html">Veronika Zemanová</a> is a retired Czech porn actress. For her TI I have used a dataset of 30 images, taken from various sites. The method I'm using as of lately tends to produce pretty accurate results already before reaching the 1000 steps, and it seems to be the case again with this embedding, which, as usual, might be a tiny bit overcooked already. However, and as the example pictures prove, you can easily build a pretty good prompt without even touching the embedding weight. I've basically replicated the training process I used for Maria Guardiola and <a rel="ugc" href="https://civarchive.com/models/14680/camila-morrone-actress">Camila Morrone </a>(have a look at the "About this version" section of the latter if you want more info on my process).

Files with SHA256: 6dff971bd0c16aace2a880dbc988f35065b2204b8926f9f076a379fcfe8bc03e