V2 and V2 Dash

This model was created to refine Canopy: The Digital Alchemist to better suit my personal workflow. My goal was to ensure zero errors with my existing WebUI A1111 ADetailer settings, while pursuing a specific style: distinct outlines, exaggerated body proportions, and unique facial deformation (deformé).

For the past few days, I’ve worked hard to improve style consistency and eliminate AD issues without sacrificing prompt adherence, all while maintaining the original vibe I loved in the previous version.

These past few days have been a realization that to truly achieve the results I want, I need a deeper understanding of the NoobAI model structure and the technical ability to actually train the models myself. (I also learned the hard way that experimenting with different settings is incredibly time-consuming and that my current computing resources are quite limited.)

As a result, I’ve reached a compromise with v2 and v2-Dash. To be honest, they’ve drifted away from the v1 style I initially aimed for, and I can’t guarantee an increase in prompt adherence. However, I have made significant progress in improving style consistency and noticeably reducing AD issues (though it's not perfect).

Interestingly, I suspect these issues persist partly because the prompt adherence is actually higher than before, causing it to behave differently from the base model. If you are like me and don't apply specific prompts to ADetailer, the issues should be much less noticeable. (I personally use wildcards for emotions and facial features to maximize randomness). I’ve also improved the unintended 'over-detailing' issue.

While it's not without its lingering regrets, I think this is as far as I can take this particular model.

Have fun~!

V1

This model was made for my own use and is compatible with WebUI A1111.

The overall quality is slightly lower than typical NoobAI models, but it still outperforms typical Illustrious XL models.

Exaggerated Stylized illustration(comics) style.

Please note that style consistency is a bit lacking and occasional artifacts may occur.

As for updates to fix these weaknesses, it's currently undecided.

I noticed that the style consistency is significantly lacking compared to my expectations

I uploaded the wrong file.

Apologies for wasting your time due to my oversight.

I've finished re-uploading it.