Hassaku aims to be a anime model with a bright and distinct anime style.
You can run Hassaku XL and use its API on SinkIn: https://sinkin.ai/m/JWknjgr
My Discord for everything related to anime models and art.
____________________________________________________________
Supporters:
Thanks to my supporters Riyu, SETI and Kodokuna
____________________________________________________________
Using the model:
The model is trained using images with minimal disruptive elements such as floating text, logos, speech bubbles, and signatures. If any of these elements are present in an image, please include the prompt "signature" as a negative prompt
Metadata and franchise tags are excluded—please do not use them. Tags such as "highres" and franchise-related tags like "re:zero kara hajimeru isekai seikatsu" are not used in training
Look at the example images, to see how the model should be used
Model is trained to be not complicated in use, tag order:
First are the number of person, then character names, rest.
Example: 1girl, rem \(re:zero\), standing, masterpiece, upper bodyNoobAIs Loras working for the most part better in compare to Illustrious loras
Here are some resolution options for SDXL:
1536 x 640
1344 x 768
1216 x 832
1152 x 896
1024 x 1024
896 x 1152
832 x 1216 (most recommended)
768 x 1344
640 x 1536
______________________________________________________
Version and License info:
Below V1 merges ANIMAGINE XL 3.0
Version V1 use Illustrious-XL & WAI-NSFW-illustrious-SDXL with additional training
V2 is trained on its own and don't include any extra merge. Base was V1, it is trained to include newer or missing characters. It was also used for further training tests
V3 is a merge of Illustrious-XL and V2, to fix some issues and was also trained to include newer or missing characters
All Models using the Fair AI Public License 1.0-SD license.
Description
Be aware that this is a WIP model, WIP means mostly the knowleghe of new characters, outsite of that, it is a full working model, but there is still work on it to do.
The model occasionally performs better when using more than 75 tokens and it also creates a bit to detailed/realistic shiny/glossy bodys currently, both will change in a later update.
Model is training on changing datasets, ranging from 120k to around 760k images. The primary learning dataset (70k images) included characters from the past year (01.06.2024 of last year to now), with at least 100 images on Danbooru. The learning dataset was heavily filtered by score and quality, compressing it to a relative small size of 70k images. Note: While the most recent characters from the last week are included, they are not yet trained well enough to be reliable (like Sameko Saba).
In Detail:
Over the past two months, I ran into several issues. After about two weeks of training, the model began to deteriorate rapidly for no apparent reason.
As a result, I had to do one hard reset and two soft resets during the process, effectively throwing away a lot of time, effort, and money. I only discovered the root cause at the beginning of this month. So, this current version is still mid-training. It mainly includes faces and hairstyles of new characters, while clothing is still in the process to get learned for the most part. I'm sharing it now despite the state it's in, as I'm already 2 months over my schedule now.
This model has a style merge, which can sometimes lead to artifacts on the eyes, especially viewed from a distance. This aligns what other models are doing.