<p><strong><em>📢Please read the version notes and change log below for SDXL generations 📢</em></strong></p><h1 id="heading-61">Dolly Little</h1><p>is a retired American model, actress and cosplayer, born on May 25, 1995 in Texas. She lives up to her name by being a mere 4'9" (145cm) and weighing a scant 85lbs (39kg). She has intense red hair and an infectiously innocent smile. She was nominated for several industry awards in 2016. Dolly speaks French and credits Kota Skye as her role model, while her best friend is actress Elsa Jean. She performed in the industry for two years, from 2015 to 2017.  Her favorite male performer to work with was Brick Danger, while her favorite female performer was Hope Harper. Dolly's fan base contributed to rescuing her from a very emotionally and financially abusive situation in Oregon, and she moved closer to her family and agency in Florida. She was nominated for several awards, including Customs Making Mogul of the Year and Cam Girl of the Year in 2016, Customs Specialist of the Year and Ravishing Redhead of the Year in 2017. Dolly grew up in Portland, Oregon, where she was studious in high school and played Friday Night Magic on weekends and hung out with D&amp;D groups. She did a Cosplay of Jean Grey for SuperCon and has led Lightsaber Choreographies as both a Sith and a Jedi.</p><p></p><h1 id="heading-2583">🏗Change Log🏗</h1><h2 id="heading-9">SDXL v1.2</h2><ul><li><p>note the different trigger use: <strong>ohwx woman</strong></p></li><li><p><strong>far more </strong>flexible model than previous models</p></li><li><p><strong>trained at 1024x1024px,</strong> top native resolutions are <strong>512/576/640/704/960x1024, </strong>and <strong>1024x960</strong></p></li><li><p><strong>not recommended to</strong> go below 1024x1024, but if you do using resolutions smaller than that will require you to <em>up the strength</em></p></li><li><p>long distance shots will also require you to<em> up the strength</em></p></li></ul><p></p><h3 id="heading-62">(Outdated) Generation Advice</h3><h3 id="heading-63">Prompt and sampling</h3><p>Most images have been generated with Lora settings at 1. All examples done with CyberRealistic v1.4. If you look at the description of most prompts, they are more about the theme and feel. This seems to generate great images every time for Dolly, close-ups, full body, etc.</p><p>Samples used 512x640 with hires fix @ 0.45-0.55 denoising either using Latent or 8x NMKD-Faces as upscaler. No restore faces used.</p><p>Works without, but use D011lyLittle as the trigger for more facial accuracy.</p><p></p><h3 id="heading-64">Upscaling</h3><p><a target="_blank" rel="ugc" href="https://github.com/Coyote-A/ultimate-upscale-for-automatic1111">Ultimate SD upscale</a> script used for all upscales ~.15 - .4 denoising.</p><p></p><h3 id="heading-65">Sidenote</h3><p>Dolly Little is one of my personal favourites. This may be my best LORA yet, which is funny as I failed probably 4x generating this one over the past week or so. I also had to use <strong>D011yLittle</strong> as trigger word as typing DollyLittle would trigger doll like effects for some models. I have not tested this on any other model this time as I'm lazy, let me know how you go.</p>

Dolly Little + XL

<ul><li><p>best resolutions trained on: <strong>448/512/576/640/704/960x1024 and 1024x960</strong></p></li><li><p>If her face output seems inaccurate, try again but use <strong>1.1</strong> to <strong>1.25</strong> strength for better output</p></li></ul><p>SD XL has done a good job of capturing Dolly Little versus any of my previous attempts. I've switched up the trigger word from DollyLittle (which appears to be trigger things embedded into most models) for the ever so neutral ohwx. I may be doing this for all my builds in the future as it allows you to know what the "trigger" word is always, and it avoids adding the wrong data when training the LoRA.</p>

Files with SHA256: c42e990cd6c44d96088cea48653f2679a20312d62801343d4d82aa0c4b67c206