<p>Well, simply put, I looked up a Grace Van Dien model and wasn't satisfied with the one I found. So I did the best thing I knew how to do, create a Lora model with a dataset. To be honest, I wasn't as careful with the dataset as I should have been but it turned out far better than I expected. </p><h1 id="heading-9265">DATA TRAINING INFO</h1><p>Just in case you want to use my methodology. Keep in mind, model training creates different results. A slight change in parameters can create a whole new look to a model. </p><p>For Grace Van Dien's training:</p><p>56 Images<br />No Caption<br />Stable Diffusion's Woman Regularization dataset set to 1 repeat</p><p>OPTIMIZER: AdamW<br />EPOCH:1<br />MAX TRAINING STEPS: 5000<br />XFORMERS: None<br />NETWORK DIM: 128<br />NETWORK ALPHA: 64<br />SCHEDULER: Consine<br />IMAGE SIZE: 768,768</p><p>These are the major components for this data training set. A slight change in any of these parameters would yield a different result no matter how slight the change is in the new result. Might not look that different to someone just starting out, but if you trained it and got used to the first Lora, you will notice the differences immediately. </p><p></p><p></p><h2 id="heading-9266">WILL THERE BE ANOTHER VERSION?</h2><p>Honestly, I don't know? I believe there is room for improvement all the time. It's why I keep trying to improve other Lora models I have. But for now, Grace Van Dien is far too accurate to even bother. The only reason to generate another model (following the exact parameters and image dataset) is to create a slightly higher quality Lora. I think the model could be slightly sharper in image renders. Obviously, up-scaling would fix this but I'd prefer the image to be a bit sharper after the Hi-res fix with upscaling being a frivolous step for a beautiful image. Otherwise, don't fix what's broken, right? I'm obsessed enough with two other lora's being perfected as it is. </p><p></p><h1 id="heading-9267"><strong>USE CASE </strong></h1><p>I expect anyone who uses this model to be respectful to the REAL LIFE, individual portrayed here. Keep your demented perverted fantasies where it is, and generate respectful images of Grace Van Dien. I ask and expect this even if you aren't posting to Civitai or some other site/forum. Treat others as we expect to be treated, right? In the case of AI and the people portrayed, treat them better than we expect ourselves to be treated. </p><p>As far as what is too far, that's a personal determination, but I peg too far to be something you wouldn't want to pop up on your screen at work, NFSW, right? </p><h2 id="heading-9268"></h2><h2 id="heading-9269">HOW TO USE</h2><p>Whelp, its already set in the Lora as is, but you're more than welcome to play with the settings. <br />WEIGHT: 0.7<br />TRIGGER: graceVanDien</p><p>A weight of 1 will not break the image, though you will see some degradation in quality, so 1 isn't really recommended. I oftentimes prefer a weight of 0.8 for Lora's I make, including this one, but I set 0.7 as its enough to do the job. </p><p>With her hair, you can play around or let randomness do its work. I prefer to add blonde to the generation parameters, but with randomness, you will get brownish blonde to blonde hair. The added blonde prompt text is for my personal vision of accuracy. </p><p><br /><br /></p>

Grace Van Dien by brassen250

<p>This is the first Grace Van Dien model. My personal rating of the current accuracy is at 85% to 90%. I only say that because no model in my experience is perfect. But compared to current models posted, this is going to be the most accurate! So, yeah, I think I have a little pride about this model. </p>

Files with SHA256: c72ebee873110818764232dd7aa6c28bf562167202dbff337e3853d147477b37