Creating people who create! We are AICU media from Japan!

Animagine XL 3.1 Release

A new version, Animagine XL 3.1, has been released from the image generation model ‘Animagine XL 3.0’, which was published on January 16, 2024 based on Stable Diffusion XL (SDXL)!

This article will compare the differences in visual style and other aspects between Animagine XL 3.0 and Animagine XL 3.1, as well as introduce the new tags introduced from Animagine XL 3.1.

【Related Articles】

Which Model is Suitable for Manga Production? I Conducted a Detailed Comparison of Animagine 3.1 vs. Animagine 3.0!

https://note.com/aicu/n/n393f2cebfc75

In Animagine XL 3.1, you can continue using the important prompts that were available in Animagine XL 3.0!

https://note.com/aicu/n/n44085d702617

Here are the release notes for Animagine XL 3.0.

https://note.com/aicu/n/nd05f3a656dd8

Here is Hugging Face.

https://huggingface.co/cagliostrolab/animagine-xl-3.1

Official Release(Parcial Excerpt）

Animagine XL 3.1 is an enhancement over the previous version, Animagine XL 3.0, designed to generate anime-style images of higher quality. It includes a wider range of characters from famous anime series, optimized datasets, and new tags for better image creation. Aimed at providing valuable resources for anime fans, artists, and content creators through accurate and detailed representation of anime characters.

That is, “Animagine XL 3.0 has been improved for higher quality!”

Comparison with Animagine XL 3.0

Now, let’s compare with Animagine XL 3.0.The environment used for this generation was a GALLERIA laptop with Windows 10, equipped with NVIDIA GeForce RTX 3060 with 6GB VRAM, directly installed with ComfyUI.

Art Style

Initially, generate using a simple prompt and compare it with Animagine XL 3.0.

Prompt : best quality, 1girl
Negative Prompt : worst quality, low quality, normal quality, bad anatomy, NSFW

Animagine XL 3.0

Animagine XL 3.1

The significant features such as a strong light source,a hairstyle with visible ears, and a sense of backlight remain unchanged. However, distinctive features of Animagine XL 3.0, such as slanted eyes and pink cheeks, are less prominent now, and the overall exposure has been reduced. Also, while Animagine XL 3.0 had a glossy skin tone, that too has been toned down, resulting in a more subdued overall appearance.

Additionally, compared to Animagine XL 3.0, Animagine XL 3.1 has clearer and finer line art, with reduced thickness and depth, giving a slightly more two-dimensional impression.

Quality tags

Source: Animagine XL 3.1 Official Hugging Face

The numerical values of the traditional quality tags have been slightly adjusted, and some tag names have been changed. However, commonly used tags like “masterpiece” and “best / worst / low quality” remain unchanged, so they can still be used in the same way as before.

Year Tags

The main changes include the addition of “2024” and slight adjustments to the years, and the change of “late” to “recent”. The era tags have not changed significantly.

Comparison of actual generated images

newest (2021–2024)

Images similar to Animagine XL 3.0.

recent (2018–2020)

It has a slightly refined atmosphere compared to the newest version. It’s the heyday of VTubers, isn’t it?

mid (2015–2017)

The decadent atmosphere is cute. The art style has clearly moved away from the VTuber style.

early (2011–2014)

It has a sepia-toned atmosphere.

https://www.amazon.co.jp/AICU-%E3%83%9E%E3%82%AC%E3%82%B8%E3%83%B3-Vol-2-2024%E5%B9%B45%E6%9C%88%E5%8F%B7-%E6%AC%A1%E4%B8%96%E4%BB%A3%E3%81%BE%E3%82%93%E3%81%8C%E5%88%B6%E4%BD%9C%E8%A1%93%EF%BC%81-ebook/dp/B0CZ8WGXVD/ref=sr_1_2?dib=eyJ2IjoiMSJ9.efN-e7JILU64z5-f93zeMOOrdGL_HdLlvbZs_QbtITrGjHj071QN20LucGBJIEps.vaRpuH9_6nr2X5LUYre2XGxrq7dkKFXJwbt2e69Owtc&dib_tag=se&qid=1716525384&s=digital-text&sr=1-2&text=AICU+media+%E7%B7%A8%E9%9B%86%E9%83%A8

Oldest (2005–2010)

It’s the era when Haruhi Suzumiya and Yuki Nagato were popular. There’s a taste of the 90s style, and the distinction in artistic approach is clearly noticeable.

Rating tags (Age restriction tags)

The rating tags used to adjust sensitive content haven’t changed significantly from Animagine XL 3.0. Use prompts and negative prompts selectively to adjust the exposure level.

Aesthetic Tags

It’s a tag used to generate visually appealing content using a ViT (Vision Transformer) image classification model specialized in anime data, along with the model shadowlilac/aesthetic-shadow-v2, which evaluates the aesthetic value of content.

aesthetic

The prompt “best quality, 1girl, very aesthetic”

The camera angles and poses have become more dynamic!

The ‘aesthetic’ prompt works effectively on platforms like Nijijourney and Novel AI, so give it a try!

Recommended settings

I will introduce recommended settings and prompts as presented on the official Hugging Face platform.

‘Absurdres’ is a term coined from ‘absurd resolution,’ meaning an incredibly high resolution.

Prompts

masterpiece, best quality, very aesthetic, absurdres

Negative

nsfw, lowres, (bad), text, error, fewer, extra, missing, worst quality, jpeg artifacts, low quality, watermark, unfinished, displeasing, oldest, early, chromatic aberration, signature, extra digits, artistic error, username, scan, [abstract]