Please leave feedback in the model review or write in DM!The LoRA was trained on all my images generated using the Pony Diffusion V6 XL model in 2024 and then my images generated using the Illustrious model in 2025 and 2026. This is why it may lean toward the lewd side.<h1 id="krea-2-v1.0">Krea 2 v1.0</h1>All images were generated with the following configuration:Base model: Krea 2 Turbo bf16Resolution: 1280x1856sampler: eulerscheduler: betasteps: 8cfg:1But I tested other configurations, and all worked well!<h1 id="anima-v1.0">Anima v1.0</h1>I didn't test it too much, so please leave your feedback!I'll leave more details here later.<h1 id="il-v2.0">IL V2.0</h1>The biggest change - it should be more compatible with all Illustrious models, not only WAI:v11Also, I cleaned the training dataset, so I hope there will be fewer problems with the handsIMPORTANT! It appears that this version can produce artifacts on the lips when you use "Uniform" schedule type. As soon as this is the most popular type in civitai online generator, here are your options:<ul><li>Use DPM++ 2M Karras sampler (set it in advanced settings). This works reliably even on a low number of steps</li><li>Or use Euler a sampler with a large steps value (&gt;40). This may not work each time.</li><li>Or use less lora strength (0.5-0.7). This may not work each time.</li></ul>Sorry for this mess, I tried to retrain the model three times, but it looks like this is a dataset issue.<h1 id="il-v1.0">IL V1.0</h1>It was trained based on WAI-NSFW-illustrious-SDXL:V11.0 model, so it may not work so well on other models.

Carefull Style [Krea 2 | Anima | IL]

All images were generated with the following configuration:Base model: Krea 2 Turbo bf16Resolution: 1280x1856sampler: eulerscheduler: betasteps: 8cfg:1But I tested other configurations, and all worked well!

Files with SHA256: 1d7dd8770f2e0038c73d6d310935493c18b248ceeadf6174f409d00c74d603d7