<a target="_blank" rel="ugc" href="https://www.instagram.com/demirose/?hl=en">Demi Rose Mawby</a> is a British model and social media influencer with more than 20M followers on Instagram. <s>I have trained her TI on about 30 high quality images and for 2000 steps. The embedding seems to work fine with the models I've tested it on (StablyDiffused, DreamShaper, Clarity, base SD 1.5). Hope you like it! :)</s>Update: This TI has been retrained from scratch with my <a target="_blank" rel="ugc" href="https://imgur.com/a/d348H3k">current settings</a> and a reduced dataset of 15 images. Even if I still liked the previous version (uploaded in February, such a long time ago!) very much, I was upscaling the pics back then and this resulted in oversaturation. Which the current models hide quite well, it has to be said, but even so it was time for an update. Thanks @ElizaPottinger for the timely reminder. :)Curious about my work process? I have summarized it <a target="_blank" rel="ugc" href="https://civarchive.com/models/56570/renata-valliulina?commentId=118383&amp;modal=commentThread">here</a>.Do you have a specific idea for a TI in mind?<a target="_blank" rel="ugc" href="https://www.buymeacoffee.com/jernaugurgeh"> Visit my website </a>and let me know.<h2 id="heading-386">Building a good prompt with my TIs</h2>You're obviously free to experiment, but bear in mind that my TIs are trained with a more or less fixed phrasing, that normally starts with:"photo of EMBEDDING_NAME, a woman"So I recommend always starting your prompt like that and then building the rest of the prompt from there. For instance, "photo of beautiful (dem1rmawby:0.99), a woman, perfect hair, posing seductively, (modern photo, __attire_mix__), (closeup portrait), jewelry, 24mm, (analog, cinematic, film grain:1.3), (outside of Berghain Panorama Bar, nighttime, gritty), ((detailed eyes)), (epicPhoto), (color picture:1.1)".Please also note that I'm using the <a target="_blank" rel="ugc" href="https://civarchive.com/models/58390?modelVersionId=62833">"add detail" LoRA</a> for my example pics. I recommend setting it around 0.5 for best results. CLIP Skip should always be set at 1 when generating pics with this TI.

Demi Rose (JG)

140-step TI trained on a dataset of 15 images with<a target="_blank" rel="ugc" href="https://imgur.com/a/d348H3k"> these settings</a>.

Files with SHA256: a6b220e4d07848edea3e85729f12c3ab4790ceb2baa325260ceaf5e9179fe23f