Please note: v3.0 is out!<hr />Samara Weaving, born 1992 in Adelaide, is an Australian actress. She may be best known for her roles in various horror movies such as The Babysitter (2017), Ready or Not (2019), and the opening sequence of Scream VI (2023). LoRA v3.0 was trained on 76 publicly available photos in various SDXL compatible resolutions.Recommendations for v3.0<ul><li>Use trigger samaraweaving (together with woman) and LoRA weight 1.0.</li><li>Photorealistic models such as <a target="_blank" rel="ugc" href="https://civarchive.com/models/43977/leosams-helloworld-sdxl-base-model">LEOSAM's HelloWorld</a> tend to produce the best results.</li><li>Trained with Clip Skip 1, results without Clip Skip vary.</li><li>Positive prompts looking at viewer, wide smile, pink lipstick and earrings may be helpful and/or interesting.</li><li>Negative prompts not needed while testing.</li><li>I didn't do much testing with CFG values, but 3 to 7.5 seem to be fine.</li></ul><hr />Older version information:Recommendations for v2.0<ul><li>Use trigger samaraweaving (together with woman) and LoRA weight 0.7 - 0.9.</li><li>Photorealistic models such as <a target="_blank" rel="ugc" href="https://civarchive.com/models/43977/leosams-helloworld-sdxl-base-model">LEOSAM's HelloWorld</a> tend to produce the best results.</li><li>Trained with Clip Skip 1, results without Clip Skip vary.</li><li>Positive prompts looking at viewer, looking over shoulder, messy hair, wide smile, pink lipstick and at the Oscars may be helpful and/or interesting.</li><li>Negative prompts man and tattoo/tattoos may be needed.</li><li>I didn't do much testing with CFG values, but 7.5 to 8.5 seems to be fine.</li></ul>Recommendations for v1.0<ul><li>Use trigger samaraweaving (together with woman) and LoRA weight 0.7 - 1.2, depending on checkpoint.</li><li>Photorealistic models such as <a target="_blank" rel="ugc" href="https://civarchive.com/models/43977/leosams-helloworld-sdxl-base-model">LEOSAM's HelloWorld</a> tend to produce the best results. Unfortunately, it may take a few tries. Still needs refinement.</li><li>Portraits seem to work best.</li><li>Trained with Clip Skip 1, result without Clip Skip vary.</li><li>Positive prompts looking at viewer, looking over shoulder, messy hair, wide smile, pink lipstick and at the Oscars may be helpful and/or interesting.</li><li>Negative prompts man and tattoo/tattoos may be needed.</li><li>I didn't do much testing with CFG values, but 7 to 8 seems to be fine.</li></ul>

Samara Weaving SDXL

Expanded dataset, better captions, resulting in a much higher quality version of the LoRA.

Files with SHA256: 382724e0a5615798241d2c5bff25068759a2e9d9af712ed8a3148033b8e839eb