<h3 id="hi-there!-v30sprp26">Hi there!</h3>Here's a quick intro to this LoRA:it's built to recreate the early visual style of <a target="_blank" rel="ugc" href="https://civarchive.com/models/1471272?modelVersionId=1664142">AGGA Utopia</a>, while being more compatible with other SDXL checkpoints. It works best with prompts like <code>anime</code>, <code>cartoon</code>, <code>shiny</code>, and <code>shiny skin</code>.<hr /><h3 id="il-or-v2-or-rescaling-lfe05nfti">IL | V2 | RESCALING</h3>At first, I considered retraining the LoRA to fix the issue caused by excessive strength, which limited its versatility. The problem was an overload of information, making it necessary to use very low weights (like 0.3 or less) to achieve balanced results.However, a simpler solution was to rescale it. By internally reducing the intensity of the weights, it can now be used with standard values like 1.0 without losing control or quality. For example, a previous weight of 0.34 is now equivalent to using it at 0.7-0.8, maintaining the intended aesthetic with better compatibility and flexibility.P.S: I rescaled it too much, but it is quite functional. If you use it at 1.0 weight it will be the same as using it at 1.5 (very saturated). I recommend 0.7, although you can try it to your liking.PS2: Use weights less than <code>0.2</code> when using it with my checkpoint and don't use triggers, those are made for other checkpoints since the model has them integrated but they are quite diluted.<pre><code>WAI-NSFW v14.0 | Steps 24 | CFG Scale 5.5 | Euler a |「Style Executor」:0.8</code></pre><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/1ec02448-932a-4c2d-8e42-8a3f8cf0e110/width=525/1ec02448-932a-4c2d-8e42-8a3f8cf0e110.jpeg" /><hr /><h3 id="il-or-v2-9ua59ilv2">IL | V2</h3>This LoRA captures the visual essence of early AGGA Utopia versions while improving compatibility across a wide range of SDXL-based checkpoints. It’s designed to reduce the reliance on complex prompts or stacked triggers by deeply integrating key stylistic traits—especially when using common keywords like <code>anime</code>, <code>cartoon</code>, <code>shiny</code>, or <code>shiny skin</code>. While these are not strictly required, they’re still recommended for consistency, especially on checkpoints without similar visual biases.The style emphasizes bold contrasts, polished cartoon aesthetics, and vibrant, glossy textures, making it particularly suited for dynamic anime-style illustrations. The visual signature is stable and refined, even at low LoRA weights, due to extensive training (~60,000 steps).🔹 Recommended strength: <code>0.3–0.4</code> for most checkpoints. Use higher values if you want to push the aesthetic even further—but be aware that weights above <code>0.4</code> may produce extreme or distorted results depending on the base model.Perfect for users seeking a consistent, stylized base to emulate AGGA Utopia’s rendering style across other checkpoints.PS: Use weights less than <code>0.1</code> when using it with my checkpoint and don't use triggers, those are made for other checkpoints since the model has them integrated but they are quite diluted.<hr /><pre><code>📦 UtopiaXL V3.0 | Steps 17 | CFG Scale 4.5 | Euler a |「Style Executor」:0.08</code></pre><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/053def2d-926c-49c6-be7d-231345bb33d8/width=525/053def2d-926c-49c6-be7d-231345bb33d8.jpeg" />

AGGA's Utopia —「Style Executor」

This version includes parts of T1 and T2 from Utopía_v1 (the first version), meaning that "CuteLatte_PS" is included in the trigger words, along with "anime, cartoon, shiny skin, shiny." However, I'm not sure if I'll bring over other versions since they are a bit more diluted regarding the base style. I might add the UNet + T1 version as an alternate variant.As an extra file, there's the visual-only version—it's lighter and doesn't touch the TE, meaning it relies entirely on what the model already knows. This is actually the true v3: just the clean style. But since I've been tinkering with the "mini-brains," I left the full version available to test out since it's the most faithful one.What I definitely want to do next is a version for v2 or v4. Even though my new workflow is much more complete than before, I haven't tried guided recovery yet—we'll see how it goes. If it works out, expect several styles rolling out back-to-back!

Files with SHA256: bbd94f7c1a5cecbab441fa6679c8806378335204ab7d35734109f8067b776c15