<h3 id="*showcase-images-include-embedded-workflow-and-are-post-processed-with-sharpening-and-noise-grain.">*Showcase images include embedded workflow and are post-processed with sharpening and noise grain.</h3>In other words, your results may vary depending on your workflow.<hr /><h3 id="recommended-settings-are-below.">Recommended settings are below.</h3><hr /><h1 id="version-features">Version features</h1><h3 id="rather-than-improvements-each-version-has-their-own-strengths">Rather than improvements, each version has their own strengths</h3><ul><li>v1.1 - 3DCG (Base)</li><li>v1.2 - Semi realistic (sometimes include outline)</li><li>v1.33 - 3DCG (Slightly less realistic than 1.1 in exchange for better lighting/composition/prompt adherence)</li><li>v1.34 - 3DCG (v1.33 + slightly more detail in background)</li><li>v2.4 - Realistic "Photoreal"</li><li>v3.1 - Realistic "Hyperreal"</li><li>v4.2 - Balanced (Merge of 1.34 and 3.1)</li><li>v5.0 - 3DCG "Video Game"</li></ul><h3 id="recommendation">Recommendation</h3><ul><li>From realistic to semi realistic = v3.1/2.4 &gt; v4.2 &gt; v5.0 &gt; v1.34/1.33 &gt; v1.1 &gt; 1.2</li><li>v5.0 - Softer and "video game like" 3DCG</li><li>v4.2 - Balance between realism and semi-real</li><li>v3.1/v2.4 - For realism</li><li>v1.34 - Good aesthetic with good face proportions</li><li>v1.1 - Best face proportions (imo)</li></ul><hr /><h1 id="version-notes">Version notes</h1><h3 id="v5.0">v5.0</h3><ul><li>Focus on new aesthetic - 3DCG with subtle skin texture and softer look</li><li>Prompt seems more responsive; lighting prompts seem to react better.</li><li>CFG 3-6 (Optimized model for CFG 5)</li><li>File size increased - no idea why - performance roughly the same</li><li>Some of the sample image prompt is different/updated</li></ul><h3 id="v4.2">v4.2</h3><ul><li>Balanced version - merge between v3.1 and v1.34</li><li>Aesthetic is in between 3.1 and 1.34; closer to earlier versions while retaining more skin texture/noise</li><li>CFG 3-6 (5 seems like the sweet spot)</li></ul><h3 id="v3.1">v3.1</h3><ul><li>Realistic skin texture, good contrast/lighting/colour, better background details</li><li>Good face proportions</li><li>Good prompt adherence</li><li>High "clarity" and ambient occlusion aesthetic</li><li>Aesthetic more illustrative (Closer to v1.34 than v2.4)</li></ul><h3 id="v2.4">v2.4</h3><ul><li>Focus more on realism</li><li>More realistic skin texture</li><li>Should have better prompt adherence</li><li>Aesthetic closer towards realistic photography than illustrative (Compared to v.1.0 variants)</li></ul><h3 id="v1.34">v1.34</h3><ul><li>Should have slightly better details to the background.</li></ul><h3 id="v1.33">v1.33</h3><ul><li>Focus on keeping realism while improving composition, background and lighting compared to v1.1.</li><li>Less creative/aesthetic than 1.2 in terms of background.</li></ul><h3 id="v1.2">v1.2</h3><ul><li>More focus towards semi realism *includes outline at certain times</li><li>Better lighting, composition, background, aesthetic</li><li>Structure feels more stable compared to v1.1</li></ul><hr /><h1 id="model-focus">Model focus</h1><ul><li>3DCG features with more detailed skin texture. (Heavily dependent on heun sampler, not recommended if using web generator or if you don't know what you're doing)</li><li>Face proportions closer to real life. Less anime like.</li></ul><h3 id="strengthquirks">Strength/Quirks</h3><ul><li>Facial features can respond to prompts</li></ul><h3 id="weakness">Weakness</h3><ul><li>Not suitable for male because of female face modifier.</li><li>Will do beauty mark/mole under eye (Apparent in lower cfg; use negative prompt)</li><li>Skin pores can look too smooth/uniform/unnatural texture depending on sampler/scheduler.</li><li><s>Backgrounds (res_multistep + normal with high steps 50-100 - has chance to resolve scale issues but might introduce artifacts) </s>(better in version 1.2 and above)</li></ul><hr /><h1 id="recommended-settings">Recommended settings</h1><h3 id="(can-see-workflow-via-uploaded-images)">(can see workflow via uploaded images)</h3><h2></h2><h3 id="positive-prompts">positive prompts</h3>can change the face structure/matureness<pre><code>milf, adult</code></pre>makeup/beauty enhancements<pre><code>(dark maroon eyeshadow under eye, maroon blush makeup, eyeliner:1), (doe eyes makeup:0.5), shiny skin, (perfect hands:0.9)</code></pre>realism tags<pre><code>subsurface scattering, photorealistic, realistic, hyperrealistic</code></pre><h3 id="initial-sampler">initial sampler</h3>sa_solver_pece + sgm_uniform<ul><li>sweet spot: steps 30-40, cfg 5-7 @ 1024x1280 pixels</li><li>more realistic at the cost of contrast/dark/warmer image: steps 30-40, cfg 4-5</li><li>provides the most stable structure, details and subtle noise at reasonable speed. handles cfg burn not bad as well. use cfg higher than 7 for more contrast.</li></ul>dpm_adaptive + normal/exponential<ul><li>steps doesn't matter since adaptive, cfg 4-10</li><li>slower but very reliable without much errors. however compared to sa_solver_pece, it is smoother/no noise and lack small details.</li></ul>dpmpp_sde + sgm_uniform/exponential/sgm_uniform<ul><li>steps 30, cfg 4-7</li><li>feels slightly faster than the two above with reasonable structure.</li></ul>res_multistep/res_multistep_a + sgm_uniform/normal/simple<ul><li>steps 30-100, cfg 4-6</li><li>fast, reliable but slightly boring/simplistic results (use res_multistep_a for more detail/interesting results)</li><li>res_multistep is fast and converges at high steps, fixes the structural problems 90% of the time.</li></ul><h3 id="upscalehiresfix">upscale/hiresfix</h3>upscale model<ul><li>any - don't think it makes a big difference which ever used</li><li>realesrgan_x2 (only because it's fast)</li></ul>heun + normal/ddim_uniform(second choice)<ul><li>steps 15-20, cfg 3-4, denoise 0.35-0.55</li><li>more random noise pattern with cleaner details compared to 2m_sde_heun karras</li></ul>dpmpp_2m_sde_heun + karras<ul><li>steps 20-30, cfg 3-5, denoise 0.35-0.55</li><li>slightly faster than above, noise pattern more uniform compared to heun</li></ul>euler_a + sgm_uniform<ul><li>steps 20-30, cfg 3-5, denoise 0.35-0.55</li><li>use with a initial sampler that produces more noise/texture then subdue/control that noise with euler_a+sgm_uniform</li><li>produces subtle uniform noise and to avoid over realistic results</li></ul><h3 id="facedetailer">facedetailer</h3>heun + normal/ddim_uniform<ul><li>steps 10, cfg 2.5-6, denoise 0.25-0.4</li></ul>dpmpp_2m_sde_heun + karras<ul><li>steps 10-15, cfg 2.5-6, denoise 0.25-0.4</li><li>mole under eye might appear if low cfg</li></ul><hr /><h1 id="remarks">Remarks</h1><ul><li>sa_solver_pece + sgm_uniform is my new favourite combination</li><li>usual euler/2m/3m + sgm uniform/normal/karass should work though</li><li>i prefer heun as it produces natural noise/texture at lower step values and loses it at higher (recommended schedulers: normal/ddim_uniform/ays+)</li><li>for prompt adherence: higher resolution can have lower cfg, lower resolution need higher cfg</li><li>personal frequently used resolution before upscale: <s>896x1120</s> 1024x1280 or 1152×1440 (more pixels = more space for details like hands)</li><li>post-processing nodes with sharpening, bloom, grain, tone curves for more texture in final image. (see workflow in images)</li></ul>

FUBUWA MIX 3D REAL

v5.0<ul><li>Focus on new aesthetic - 3DCG with subtle skin texture and softer look</li><li>Prompt seems more responsive; lighting prompts seem to react better.</li><li>CFG 3-6 (Optimized model for CFG 5)</li><li>File size increased - no idea why - performance roughly the same</li></ul>

Files with SHA256: 1fdf7f479c1df0e928dded207beb5d1727e17689e9b4acbc1b0ae924e9707e14