Read the user guide below!
AniEase is a merge of some SDXL anime checkpoints intended to be able to produce high-quality anime images.
This model uses Animagine's way of prompting:
1girl/1boy, character name, from what series, everything else in any order, masterpiece, best qualityIt is also recommended to include the style name of the images you want to generate in the prompt (eg. anime screencap, sketch, pixel art, famous artist name, etc.).
The standard negative prompt recommended for this model:
nsfw, lowres, (bad), text, error, fewer, extra, missing, worst quality, jpeg artifacts, low quality, watermark, unfinished, displeasing, oldest, early, chromatic aberration, signature, watermark, artistic error, username, scan, [abstract], realistic, 3d,You can add negative embeddings to the negative prompt, but I think it's not necessary at all.
The recommended sampler is Euler a with ~5-7 CFG and ~25-30 steps.
What is the difference with AingDiffusion XL?
AingDiffusion XL is considered as "continued training" of Animagine XL v3, fine-tuning the model with my own trained LoRA. While AniEase is a merge of models, containing AingDiffusion XL itself and some other models.
Legals
This model is using Fair AI Public License 1.0-SD. Information about the model merge is included in the About Version section of each respective version.
Because of some reason, I decided to make the merging recipe clearer which I will put in the "About Version" section starting from version v1.4.1.
===
For my Indonesian friends: AniEase. Haha, get it?
Description
No v1.2 because #AsalBukan02
AniEase v1.3 is here!
Pros:
More optimization for all use cases (OG character creation and characters known by the model).
Better detail on some aspects.
Cons:
Still performs worse on sketch and some styles compared to Animagine XL v3.
Using Animagine XL v3, AAM XL, anima_pencil-XL, and SwampMachine for this merge.
FAQ
Comments (6)
i posted the same seed image in both versions. compared to v1.1, v1.3 looks like AingXL. its unfortunate that the prompt "anime screencap" has lost its effect. however, by increasing the CFG from 5 to 10, it improves
Hi, some styles have indeed weakened due to the nature of merged models. However, I completely disagree that "anime screencap" has completely lost its effect.
I've noticed that you also didn't follow the prompting rules I put in the model description, which makes it worse. Using the prompting rules and the recommended negative prompt will keep the model in line with how Animagine XL v3 is trained.
I also noticed that you use PonyDiffusion negatives, which won't work at all, because AniEase didn't use PonyDiffusion for merging.
ty reps, I understand that there is no pony merge, therefore that prompt is my mistake. I followed the info again and rebuilt the prompt and it improved a bit, and rest is a matter of preference. in any case, I have no doubt that this is of high quality compared to other SDXL models
Would you consider making a model like this based on PonyDiffusionV6XL? In my opinion I think it's much better than AnimagineXL so I'm curious what you would be able to make with it.
I tried. PonyDiffusion XL's way of training is just too different to merge into other models. It completely breaks Pony's way of prompting and other model's way of prompting, making it very hard to work with.
Maybe I will try some more, but no promises.
@kayfahaarukku No worries I understand. There's another pony based model called AutismMix, if you need help maybe you could ask the creator of that model for advice, I'm sure they would be willing to help out considering their model is even more accurate when it comes to hand anatomy/lora training compared to PonyXL. :)