Full anime mode! 🎨 Crisp lineart, vivid colors, that classic illustrated-art feel.
### ⚙️ Recommended settings
| Setting | Value |
|---------|-------|
| **CFG** | **6.0** (range 5.0–6.5) |
| **Sampler** | Euler a |
| **Steps** | 30-40 |
| **Resolution** | 832×1216 / 1216×832 |
| **VAE** | baked in ✓ |
### 💡 Tip
Booru quality tags work fine here (`masterpiece, best quality, very aesthetic, absurdres` etc.) — this checkpoint's anime lineage expects that vocabulary, unlike the realism end of the spectrum.Description
FAQ
Comments (5)
First, thanks a lot for this very nice model. 👍Finding a checkpoint that is effective at following your prompt using strict danbooru tags is always nice.
Tho, regarding your comment on this model, "Booru quality tags work fine here masterpiece, best quality, very aesthetic, absurdres etc.)", the only Danbooru tag in it is "absurdres".
For "masterpiece, best quality" quality tags, those aren't Danbooru ones but Illustrious ones. In fact they are custom tags used during the very first Illustrious training. So their effect on generated content is 100% dependen,t of how dilluted or not the Illustrious model is inside of the merge. Anyway they don't exist inside of the Danbooru tags database.
And last but not least, "very aesthetic" is the quality tag that has the least background regarding from where it comes from. Is it one of the many placebo tags that many people use thinking that it has any effect when changes only come from the overall change in tokens weight coming from using it? Most of the time, probably. The other possibility would be that "very aesthetic" comes from a specific checkpoint that has been trained using this custom tag on a dataset or through a merged LoRA. Expect that most merge creators ignore which custom tags are part of what they use for the merge most of the time. So, merge after merge, it ends up completely dilluted.
PS: sorry for the text wall but it needs to be explained. ^^
Interesting... i have no idea. I need to do more research
Great model, very creative !!! simple prompts give great results !!!
Thanks :)
This is super good, we need update!!