This mix was originally inspired by the Perfect World technique - a simple 50/50 mix of a super-responsive anime-like base - the incredible <a target="_blank" rel="ugc" href="https://civarchive.com/models/7371/rev-animated">Rev v1 - </a>with the realism of Basilmix, and was since extended rather haphazardly with the aim of realistic skin tones, diversity and trying to achieve a more natural facial structure than the somewhat stylised one available with Rev.The exact recipe is lost to the sands of time, but includes greater or lesser quantities of:<ol><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/7371/rev-animated">Rev v1</a> - <a target="_blank" rel="ugc" href="https://huggingface.co/nuigurumi/basil_mix">Basilmix</a> 50/50</li><li>some <a target="_blank" rel="ugc" href="https://civarchive.com/models/6424/chilloutmix">Chilloutmix</a></li><li>a dash of <a target="_blank" rel="ugc" href="https://civarchive.com/models/4481/pov-skin-texture-dreamlike-r34">PoV Skin Texture - Dreamlike r34</a></li></ol>And a few others, not sure what anymore. But I like it and thought perhaps you might too. The sample images aren't touched up in any way other than hires-fix, and aren't really cherry-picked for quality, just a sample of pics I like to represent the range.<ul><li>I pretty much always add the default <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/sd-vae-ft-mse-original/blob/main/vae-ft-mse-840000-ema-pruned.ckpt">stability 84000 VAE</a> but it doesn't really mind if you want to try something different.</li><li>It takes LORAs and embeddings very well, and is capable of some significant diversity of characters.</li><li>Hands and feet aren't perfect, but usually turn out alright. Faces can be a little dodgy at distance, but isn't that always the way?</li><li>Reasonably forgiving of high canvas sizes, doesn't care too much about dimensions - seems equally happy square or 2/3 or 3/2, 640x640 is a nice spot for square aspect ratio. 512x768 is the natural sweetspot but it even remains coherent at 640x960.</li><li>2GB safetensor is the only option for now, it's pruned to bf16 and ema-only. If there's overwhelming interest I'll try to recreate the recipe via some form of trail and error and then we'll be able to generate submixes and alternate formats.</li><li>VERY HORNY model, be warned</li></ul>Kinda 2.9D out of the box. For max photorealism, it helps to add pre-prompts such as: (8k, RAW photo, highest quality, best quality, masterpiece, ultra-detailed, hires, absurdres, sharp focus:1.2), (photorealistic, hyper-realistic:1.37),Go easy on the negative prompts, I tend to stick with something likeEasyNegative, (worst quality, low quality:1.4), (malformed hands, poorly drawn hands, mutated fingers:1.4)

Files with SHA256: 8c250e7c06c07cc7381c7656ef53ae46add827695c93fdc8dd4decc527f4194e