<h3 id="oct.29-2026-moss-version-is-updated:">Oct.29 2026, Moss version is updated:</h3>Moss V1 Features：MossImage SFT is an exceptional text-to-image model. However, after days of personal testing, I’ve identified several areas that could be optimized (purely from my own perspective): For example, the lighting and shadow tend to look flat with insufficient contrast, high-frequency details appear blurry, human skin tones lean too red, colors are overly saturated, and scene elements — especially non-human ones — lack sufficient randomness (this issue becomes even more noticeable when using a speed-up LoRA).To address these, I specifically trained the Moss-version Realism Aesthetic. After repeated testing on over 20 saved LoRAs, I selected this one to release as the V1 version. Compared to the base SFT model, applying this LoRA can mitigate the above issues to a certain extent.<h3 id="key-differences-after-using-this-lora">Key differences after using this LoRA</h3><ol><li>More realistic lighting and shadow, with improved contrast</li><li>Enhanced overall detail performance</li><li>Upgraded portrait aesthetic quality</li><li>Human skin tones no longer carry a strong red cast</li><li>Image color saturation becomes more natural</li><li>Higher scene realism with reduced "AI-generated" artifacts</li></ol>You are all welcome to test it and share your output results. Hope you enjoy it!<h3 id="recommended-settings">Recommended Settings</h3><ul><li>Weight: 0.8-1</li><li>Sampler: Euler + Simple</li><li>Steps: 25</li><li>CFG Scale: 3.0</li></ul>For now, using a speed-up LoRA alongside this one is not recommended.MossImage SFT模型是非常优秀的文生图模型，但是经过这些天我个人的测试下来，我发现它还有很多可以优化的点（仅仅针对我个人来说）：比如光影偏扁平，缺少对比度，高频细节模糊，人像的皮肤色调偏红，颜色过于饱和，场景元素尤其是非人物元素的随机性不够（用了加速lora后这个问题更明显）等。因此，我特地训练了Moss版本的Realism Aesthetic，从保存的二十多个Lora中，经过我反复测试后，筛选出这个作为V1版本发布。相比SFT模型本身，使用这个Lora后能一定程度上改善以上那些问题。使用Lora后有何不同：1.光影更真实，对比度提升。2.细节提升。3.人物美学提升。4.人物皮肤色调不再那么偏红。5.图像颜色的饱和度变得更自然了一些。6.场景真实度得到提高，AI感降低。欢迎大家测试返图，希望大家喜欢。推荐设置：权重：0.8-1，采样方式：Euler+Simple，25步，CFG=3.0目前暂不建议使用加速Lora<h3 id="sep.20-2026-krea2-version-is-updated:">Sep.20 2026, Krea2 version is updated:</h3>Krea2 Features：<ol><li>It can enhance the aesthetic appeal of characters, eliminate the distinct AI-generated look, and improve skin texture.</li></ol><ol><li>It makes landscapes, still lifes, indoor and outdoor scenes more photorealistic.</li><li>The lighting and shadows become more authentic, with better light transmission and clarity.</li><li>Richer and more details.</li><li>The overall aesthetic is significantly upgraded.</li><li>It features high generalization performance, does not alter the face of character LoRAs, and is fully compatible with both NSFW and SFW content.</li></ol>‌Recommended weight‌: 1.When used in combination with a photorealistic base model or multiple LoRAs, you can lower the weight to avoid overfitting.‌Recommended sampling method‌: Euler+Simple, ‌steps‌: 10<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/eb2dd50e-0a82-4819-814b-54cfc4514b88/original=true/eb2dd50e-0a82-4819-814b-54cfc4514b88.jpeg" />This picture shows its effects on different Checkpoints.(Krea2RedMix1.1-INT8-Convrot-Native and Krea2-rayArtshoot_v5_int8_convrot)Krea2版本特点：<ol><li>可以提升人物美感，去AI味道，提升皮肤质感。</li><li>让景观、静物、室内室外等更写实。</li><li>光影更加真实，通透感更好。</li><li>细节增加。</li><li>美学提升。</li><li>高泛化性，不影响人物lora的脸型，对nsfw/sfw均兼容。</li></ol>推荐权重：1，叠加写实大模型或多lora使用，可降低权重防止过拟合。推荐采样：Euler+Simple， 步数：10<h3 id="aug.29-2026-zit-version-is-updated:">Aug.29 2026, ZIT version is updated:</h3>Attention: ZIT lora weight can't be too high, 0.6 is recommend of this lora. Not recommend to using multiple loras because it can reduce the image quality, except to take a try on lower weight than 0.6. Lower weight than 0.6 may bring nothing to the image when using this lora, I think this is the deficiency of training on distilled model.注意：ZIT的lora生图时，权重不要过高，经我测试推荐设置为0.6。 我不推荐该lora叠加其他更多lora一同使用，因为可能导致图像劣化，除非你自行测试降低权重，然而低权重可能会效果不明显，我感觉这些问题就是在蒸馏模型训练上训练lora的不足之处。<h3 id="recommended-settings:trigger-word:-virealismweight:">Recommended Settings: Trigger Word: virealism Weight:</h3><ul><li>Z-Image Base LoRA used on Z-Image Turbo: 1.5</li><li>Z-Image Base LoRA used on Z-Image Base: 0.7</li><li>Z-Image Turbo LoRA: 0.6</li><li>Krea2 LoRA: 1</li></ul>ZIT comparison:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/ab65e77e-47e5-41af-a61e-d58b1d6b0b71/original=true/ab65e77e-47e5-41af-a61e-d58b1d6b0b71.jpeg" />Here is the comparison between euler+ddim_uniform with euler+simple:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/aef0206d-9fe0-4cb1-9cf5-13b6e1db5218/original=true/aef0206d-9fe0-4cb1-9cf5-13b6e1db5218.jpeg" /><h3 id="aug.28-2026-zib-version-is-updated:">Aug.28 2026, ZIB version is updated:</h3>I've long wanted to train a LoRA to reduce the artificial, AI-generated look of my images. To achieve this, I prepared a training dataset of real-world photography, and trained on Z-Image Base, Z-Image Turbo and Krea2. I will release these three LoRA models then. I found that while the performance of the LoRAs varies slightly across different base models, the following improvements are achieved overall:<ul><li>The overall lighting and shadow realism of the image is enhanced</li><li>Human skin appears fairer and gains more translucency</li><li>The contrast between light and dark areas is more pronounced</li><li>Textures and details look more authentic</li><li>The overall aesthetic presentation of human subjects is improved</li><li>High generalization performance, no negative impact on the facial features of character LoRAs</li></ul><h3 id="recommended-settings:trigger-word:-virealismweight:">Recommended Settings: Trigger Word: virealism Weight:</h3><ul><li>Z-Image Base LoRA used on Z-Image Turbo: 1.5</li><li>Z-Image Base LoRA used on Z-Image Base: 0.7</li><li>Z-Image Turbo LoRA: 0.6</li><li>Krea2 LoRA: 1</li></ul>Sampler: Euler Scheduler: simple / ddim_uniformComparison/来对比一下看看：<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/050bfb84-a48c-4e8b-92ee-f8ccab0c2ba3/original=true/050bfb84-a48c-4e8b-92ee-f8ccab0c2ba3.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3f0a21ba-5178-4c01-afa1-3897914254a6/original=true/3f0a21ba-5178-4c01-afa1-3897914254a6.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/00191a51-396f-4e2b-8cac-09687a504a8a/original=true/00191a51-396f-4e2b-8cac-09687a504a8a.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/15b85577-6066-458c-9ca6-e2e6d8ccb321/original=true/15b85577-6066-458c-9ca6-e2e6d8ccb321.jpeg" />我一直想训练个自己的lora来降低生图的AI感，于是我准备了真实摄影的训练集，在zimage base、turbo和krea2进行了训练，我会先后发布它们。我发现虽然在不同大模型上，lora效果表现略有差异，但总体得到了以下改善：<ol><li>画面整体光影真实感得到提升</li><li>人物的皮肤更加白皙、通透感增加</li><li>明暗对比呈现增强</li><li>纹理与细节更真实</li><li>人物美学提升</li><li>高泛化性不影响人物lora面部</li></ol><h3>推荐设置：</h3>触发词：virealism权重：zimage base lora 在turbo上使用：1.5， zimage base：0.7zimage turbo lora：0.6krea2 lora：1采样器：Euler 调度器：simple / ddim_uniform

TJ Realism Aesthetic

V1 Features:<ol><li>More realistic lighting and shadow, with improved contrast</li><li>Enhanced overall detail performance</li><li>Upgraded portrait aesthetic quality</li><li>Human skin tones no longer carry a strong red cast</li><li>Image color saturation becomes more natural</li><li>Higher scene realism with reduced "AI-generated" artifacts</li></ol>Recommended Settings<ul><li>Weight: 0.8-1</li><li>Sampler: Euler + Simple</li><li>Steps: 25</li><li>CFG Scale: 3.0</li></ul>For now, using a speed-up LoRA alongside this one is not recommended.

Files with SHA256: 122a128ca3c12bd35a98fbd93f4b88893af03d5e249ff8967ece1614b39054d4