Updated pics with locally trained lora "VETE" on 30/03/2026Updated pics on illustrious XL2.0 05/03/2026Mimic style of "POP929""V1.X" dataset is weak for eyes generation, use Facedetailer"V1.Y" dataset expands x2.5, higher definition and better tags, can do close up portrait without facedetailerNo trigger words, but depending your base model; Tags such as 1990s_\(style\) or flat_colors can greatly help to make the styling lora to "pop out"About V1.Y check models details ----&gt;<ul><li>Illustrious XL 2.0 V1.Y is trained on classic pattern 1024P / kohya / Official illustrious XL 2.0</li><li>V2 ILL V1Y is trained using beta version of civit AI toolkit at 1536P with civitai illustrious base</li><li> VETE is the locally trained lora, using VETE training checkpoint ; I believe this one is the closest to the artist styling. Start with 0.8 weight</li></ul>Give both a try to see what does work the best for you, Illustrious XL 2.0 might be more flexible.About V1.X:V1 use custom illustrious model for trainingV2 use baseline illustrious (civit) model for trainingV3 is SD 1.5 Variant because why not ¯_(ツ)_/¯ , weight can be pushed up to 1.2 before it goes badbase model used for generation is <a target="_blank" rel="ugc" href="https://civarchive.com/models/76489?modelVersionId=111497">https://civarchive.com/models/76489?modelVersionId=111497 </a>V2 yamimeand mostly Plantmilk "Walnut" for illustrious showcase pictures

Pop styling 929 - V1.X and V1.Y

High quality picture (up to 1536pxl)Datasets include Introduced older arts, and few black and white comics and rough sketchBaseline illustrious from civit trainer was used<pre><code>"engine": "ai-toolkit", "adamw8bit" ; "resolution": 1536</code></pre>

Files with SHA256: 287948fae777d7e22ee621d6949a8cc6c30dc9503515887b7f3bda3a5fc342ac