Last Update: 06.03.2026
For this article I focus on a different aspect of models: The character knowledge and it's knowledge cutoff. Every model is trained on some kind of database so the model is only able to generate characters that existed (with sufficient exemplary images) when the database was created.
The most popular database comes from danbooru.donmai.us, a website with millions of images to train on (NSFW warning). This database grew over the last years (Danbooru2021, Danbooru2022, Danbooru2023, Danbooru2024) to ~8.000.000 images. But there are also other sets like e621-2024 and e621_newest from e621.net, yande_full from yande.re etc. Models usually learn to generate all characters with sufficient images within the trained datasets.
But even if a model is generally great at generating characters, it cannot know how to create a character that is newer than the dataset it is trained on. This is where the knowledge cutoff becomes important. In this article I want to illustrate the different knowledge cutoffs using test images.
For these tests I'm trying out different characters from Genshin Impact. With ~267.000 images on Danbooru at the moment, it is one of the largest franchise in the database. Additionally new characters are being released every few weeks, so it serves as a great example for knowledge cutoff.
For everyone not familiar with Genshin Impact characters, here is an overview what the characters originally look like. The dates are the time when the 100th image of the character was uploaded to Danbooru, since newer models are generally able to generate characters with at least ~100 training images. All dates are in the format DD.MM.YYYY.
________________________________________________________________________________________________________
Overview:
________________________________________________________________________________________________________
Pony Diffusion V6 XL released 08.01.2024, knowledge cutoff ~ 2022/2023
________________________________________________________________________________________________________
Illustrious-XL v0.1 released 25.09.2024, knowledge cutoff ~ 2023/2024
________________________________________________________________________________________________________
Illustrious XL 2.0 released 18.03.2025, knowledge cutoff ~ 2023/2024
________________________________________________________________________________________________________
NoobAI-XL Epsilon-pred 1.1 released 01.12.2024, knowledge cutoff ~ Summer/Autumn 2024
________________________________________________________________________________________________________
Animagine XL 4.0 released 13.02.2025, knowledge cutoff ~ Winter 2024/2025
________________________________________________________________________________________________________
RouWei v0.8.0 vpred released 09.06.2025, knowledge cutoff ~ early Spring 2025
________________________________________________________________________________________________________
WAI-illustrious v15.0 released 31.08.2025, knowledge cutoff ~ late Spring 2025
________________________________________________________________________________________________________
Hassaku XL v3.2 released 24.10.2025, knowledge cutoff ~ late Spring 2025
________________________________________________________________________________________________________
JANKU v6.0 released 14.11.2025, knowledge cutoff ~ Spring/Summer 2025
________________________________________________________________________________________________________
Raehoshi illust XL v8.0 released 25.11.2025, knowledge cutoff ~ Autumn 2025 but only for few selected series including Genshin Impact
________________________________________________________________________________________________________
Chenkin Noob XL v0.2 released 25.12.2025, knowledge cutoff ~ Spring 2025, seemingly undertrained
________________________________________________________________________________________________________
Anima - Preview released 02.02.2025, knowledge cutoff September 2025
________________________________________________________________________________________________________
I will update this article from time to time.