Focused mainly on realistic portraits and flexibility, with higher native resolution support. (V8: 896x1152, up to 1024x1280, 1280x1280)Example images are somewhat cherry-picked, no hi-res upscale used, no face-fix.CLIP Skip 2 might help with realism in some cases while sacrificing fantasy aspects, experiment with it.<h3 id="heading-78">V9/V8/V7 comparison (896x1152):</h3><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3e58c395-6fa6-4a60-b2a4-8bce59013aba/width=525/3e58c395-6fa6-4a60-b2a4-8bce59013aba.jpeg" /><h3 id="heading-79">V8/V7 comparisons:</h3>1024x1280<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/ef254b53-448f-455f-ab56-f2cdf6c83582/width=525/ef254b53-448f-455f-ab56-f2cdf6c83582.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/0a92eff7-cbd9-410c-ba40-36198074555a/width=525/0a92eff7-cbd9-410c-ba40-36198074555a.jpeg" />768x1024<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/f894426e-bbb1-47db-b756-559f05244e3c/width=525/f894426e-bbb1-47db-b756-559f05244e3c.jpeg" /><h3 id="heading-80">V1 Recipe (might not be exactly how I did it):</h3>Exp1-Tier1: (using add difference for all)<ul><li>GreyModel v12.0.0 + Dreamlike Photo v2 (25%)</li></ul><ul><li>+ Woop Photo (25%)</li><li>+ Protogen Infinity (25%)</li><li>+ ArtEros (15%)</li><li>+ HSPU (15%)</li><li>+ URPM (15%)</li><li>+ F222 (15%)</li><li>+ AlmostDirty (15%)</li><li>+ BondageV10 (5%)</li><li>+ TWAM (5%)</li><li>+ SXD1.0 (5%)</li><li>+ 3DKX1.1 (1%)</li><li>+ Hassan1.5 (1%)</li><li>+ OpenJourneyV2 (1%)</li><li>+ StrangeDedication2 (1%)</li></ul>Exp1-Tier2-Part1: (still add diff)<ul><li>GreyModel12.0.0 + AnalogPhoto (25%)</li><li>+ Dreamlike Photoreal v1 (25%)</li><li>+ Amphibia Diffusion (15%)</li><li>+ Artstation (15%)</li><li>+ CALARTS (15%)</li><li>+ ChromaV5 (15%)</li><li>+ Corporate Memphis (15%)</li><li>+ NitroDiffusion (10%)</li><li>+ Dreamlike Diffusion (10%)</li><li>+ FracTime (10%)</li><li>+ InkPunk (10%)</li><li>+ Midjourney-v4-diff (10%)</li><li>+ RPGv2 (10%)</li><li>+ SeekArt MEGA (10%)</li><li>+ Colorjizz768 (5%)</li><li>+ Double Exposure (5%)</li><li>+ MegaPals (5%)</li></ul>Exp1-Tier2-Part2: (add diff)<ul><li>GreyModel12.0.0 + MosaicArt (5%)</li><li>+ RealisticCartoon (5%)</li><li>+ Strahd (5%)</li><li>+ Technicolor (5%)</li><li>+ DMT Entity (1%)</li><li>+ Elysium (1%)</li><li>+ BalloonArt (1%)</li><li>+ DnDClasses (1%)</li><li>+ DarkVictorian (1%)</li><li>+ ComicDiffusion V2 (1%)</li><li>+ DiscoDiffusion (1%)</li><li>+ Dvmjv4 (1%)</li><li>+ Halloween (1%)</li><li>+ Lichtspiel (1%)</li><li>+ Lnkdn (1%)</li><li>+ Midjrny-v4 (1%)</li><li>+ Mtg-style (1%)</li><li>+ NekoMix (1%)</li><li>+ Nekrofaerie (1%)</li><li>+ SDHeroCalm (1%)</li><li>+ SpiderVerse (1%)</li><li>+ Ssmoky (1%)</li><li>+ SynthwaveV2 (1%)</li><li>+ TDST (1%)</li><li>+ VanGogh (1%)</li><li>+ Vectorartz (1%)</li><li>+ Voxelish (1%)</li><li>+ Wa-vy (1%)</li><li>+ Woolitize768 (1%)</li></ul>Exp1-Tier3 (add diff):<ul><li>GreyModel12.0.0 + Babetopia (20%)</li><li>+ Instababe (20%)</li><li>+ PovSkinDream (15%)</li><li>+ PowerfulMix20 (15%)</li><li>+ GTMult v1? (15%)</li><li>+ LatexRaw (10%)</li><li>+ ClassicNegative (10%)</li><li>+ Furrystaber (5%)</li></ul>Exp1-Tier4 (add diff):<ul><li>GreyModel12.0.0 + RealismMix (20%)</li><li>+ Hassan1.4 (15%)</li><li>+ AttractiveMix (10%)</li><li>+ VoluptuousMix (10%)</li><li>+ SamDoesBimboes (1%)</li></ul>The Tiers separately are UNUSABLE because of all the AddDiffs, but merging them together brings them back to usable territory.Probably merged these with Weighted sum as( (Tier1 + Tier2Part1)*0.5 + (Tier2-Part2 + Tier3)*0.5 ) + Tier4 * ?Then merged it with a custom DB model (trained on a face, but it also learned a bit of analog photo style) to tone it down a bit= Exp1FinalExp2 (using weighted sum):<ul><li>ArtEros + RealVision1.2 (50%)</li><li>Grey12.0.2 + ProjectPhoto BETA (50%)</li><li>SexyRealism + TheGirlNextDoor (50%)</li><li>Deliberate1.1 + AnythingFurry (25%)</li><li>Visigen + Futagen (25%)</li></ul><ul><li>ArtRealV + GreyProjectPB (50%)</li><li>DelibAnyFur + VisiFuta (50%)</li><li>DelibAnyVisiFuta + SexyRGirlND (35%)</li></ul><ul><li>ArtRealGreyPro + DelibAnyVisiFutaSexyRGirlND (35%)</li><li>+ Emotional Puppeteer (1%)</li></ul>= Exp2FinalExp1Final + Exp2Final weighted sum 50%?= Exp3Moving it back a bit towards photorealism:Exp3+ (not sure about the weights from here on, but used weighted sum)<ul><li>(Analog + SimpMaker3k1 + DoubleExp + Attractive + Licthspiel)</li><li>+ (SexyRealism + TheGirlNextDoor)</li><li>+ (RealVision1.2 + ProjectPhotoBeta)</li><li>+ (Dreamlike PhotoV2 + Deliberate1.1)</li></ul>= WyvernMix V1

WyvernMix (1.5 & XL)

probably a bit better faces, still not perfectseems to be more biased towards femalesadded merges: WonderMix, CyberRealistic1.2, Faetastic, RealisticAnimeV2, PhotoMaxUltra, Pinocchio, Pixar, WoopAlpha2, Dungeons'n'Waifus2, another custom dreambooth (768), custom lora (768)might need extra weighting for the custom trigger words, as they were merged with low weightrealistic person vs fake person were two different datasets with real and AI generated faces, not sure how much it picked up from that

Files with SHA256: bb191491f9631908ae4a30718fd6b33e9fee8e8cdace7d41b63851fac88b4a33