The reason for this model is quite... Hmmm... Simple in my opinion... When trying to generate some images, I realised that some characters I would have liked to be present in the ONE SHOT version were not. For example, Anna Yamada, who is a character I like (even though I've never read the manga or watched the anime). I know, it's complicated, I actually fell in love with the opening song of the anime The Dangers in My Heart (season 2) and ended up finding the character cute... 😅😅😅

Anyway, she wasn't in the Illustrious dataset or any Checkpoint so far, just like the other characters, so I thought: You know what, I'll put them in my dataset and release a Christmas present model. Anyway. Here it is.

I took the opportunity to fix a few things, such as sharpness and colour.

Like all ONE FOR ALL models, it works with most samplers, but my recommendation is:

DPM++ 3M SDE

SGM Uniform
70 steps
CFG 8

About DPO (Direct Preference Optimization).

DPO fine-tune on datasets scored as highly visually appealing by an aesthetics classifier, to bias the model to visually appealing generations, while maintaining distributional guarantees and improving generic text-image alignment in addition to visual appeal. The training is via gradient descent, automatically validate checkpoints with the 500 unique prompts of the Pick-a-Pic validation set: measuring median PickScore reward of generated images. DPO scheme implicitly learns the differences in between images by taking an expectation over the inner term of the prompt.

So, DPO pay more attention on your prompt and try to generate a better image from it.

According to the Illustrious licence, here is the recipe for the model, if you use my model as an ingredient, you are obliged to provide the recipe publicly:

model A: OFA_Anime_Base_XL-AY
model B: sd_illustriousPencilXL_v200x0.5+raehoshiIllustXL_v21x0.5x0.5+waiNSFWIllustrious_v70x0.5+featurelessMix_v10x0.5x0.5_x0.67+molKeunFurryMix_molKeunMixNoobXLx0.33
model C: Illustrious-XL-v0.1x0.25

alpha,beta : (0.5, 0.25)

mode : sum Twice

CalcMode : Cosine_A

I'm very grateful to the authors of the Checkpoints and LoRAs used in the ONE FOR ALL series and I recommend everyone to visit them and appreciate their work (if your name isn't here, it's because any of your assets were been used):

And to all those who in some way collaborated or encouraged this work, I affectionately offer this simple but sincere Special Thanks:

Special Thanks to those creators:

Special Thanks to my amazing followers YOU GUYS ARE AMAZING!! (995 followers!!):

AND YOU WHO IS AMAZING TO READ SO FAR HERE!! WOW!! 😱🤩💓

YOU ARE REALLY AMAZING!! THANK YOU VERY MUCH!! 🥰🥰🥰

0 comment