ANIMAGINE 4.0 Series Checkpoint Guide

Q: Am I dreaming Cagliostrolab has updated Animagine 4.0 ?
A: No you’re awake! Cagliostro Lab has just updated Animagine/Anime4gine to version 4.0 + two aditional Version 4.0.zero, 4.0.opt.

Q: So, What''s new in Animagine 4.0 compared to version 3.1?

Massive Dataset: V4.0 is trained on a dataset of 8.4 million diverse anime-style images and this includes many popular characters such as Hoshimi Miyabi from Zenless Zone Zero ,Zani from Wuthering Waves and Many other Characters. The model knowledge was cut-off on January 7, 2025, ensuring it can generate the latest popular characters.
Animagine XL 4.0 Zero:, pretrained base model ideal for LoRA training and fine-tuning.The base, pretrained model without additional aesthetic fine-tuning.Ideal as a foundation for further LoRA training and custom fine-tuning.Retains the raw training characteristics, making it more flexible for those who wish to customize further.
Animagine XL 4.0 Opt: (Optimized) This version has been refined with an additional dataset for improved stability, enhanced anatomy accuracy, reduced noise/artifacts, richer color saturation, and overall superior output quality rather than 3.1.
Enhanced Aesthetic Quality
New Prompting System: The updated prompting system now includes special tags such as high score, great score, bad score, average score, and low score along with temporal tags (e.g., "year 2005; year 2010; year 2016; year n") and content rating tags (safe, sensitive, nsfw, explicit).

Q:Hah only that? Many Illustrious Fine tune checkpoint also same has that many datasets

Oh if you think so. Let me tell you something Illustrious does not have, namely anime aesthetic styling (I mean the style more looks like anime ), Animagine 4.0 and before previously having a very fine aesthetic character quality unlike most illustrious or pony models you can see the comparison of the image below Also Animagine can do more styling rather than illustrious checkpoint example can do flat color,3d pixel art, and many other creative style.

Q: What about the differences In settings Between Version 4.0 and 3.1?
A:

Sampling & CFG: For V4.0 Series, the recommended sampling method is either Euler Ancestral or DPM 2 M Karras. Use 25–28 sampling steps (25 being the official recommendation) and a CFG scale of 5–8 (5 is recommended).
Prompt Examples: Use Danbooru Prompting style If you still dont know (danbooru style) yet, go check my guide.
Score related (masterpiece, high score, great score, absurdres) should be put in the last of your final Prompt

Positive : 1girl, firefly \(honkai: star rail\), honkai \(series\), honkai: star rail, casual, solo, looking at viewer, outdoors, smile, reaching towards viewer, night,masterpiece, high score, great score, absurdres,
Negative (What you dont want into your IMG Generation): lowres, bad anatomy, bad hands, text, error, missing finger, extra digits, fewer digits, cropped, worst quality, low quality, low score, bad score, average score, signature, watermark, username, blurry

Additional Insights on the Checkpoint (4.0 doest use aesthtic tags)

Quality Modifier tags masterpiece = 95% Best quality =85% Great quality =75% Good quality =50% of the IMG

Animagine 3.1 Aesthetic Ratio Tags Very aesthetic= 0.75 % Aesthetic = 0.50% (Correct Me IF Im Wrong)

RECOMMENDED RESOLUTION

Orientation	Dimensions	Aspect Ratio
Square	1024 x 1024	1:1
Landscape	1152 x 896	9:7
	1216 x 832	3:2
	1344 x 768	7:4
	1536 x 640	12:5
Portrait	896 x 1152	7:9
	832 x 1216	2:3
	768 x 1344	4:7
	640 x 1536	5:12

Pros and Cons

Pros:

Updated Character Roster: Thanks to the massive, recent dataset, Animagine 4.0 Series can generate many of the latest anime characters.
Improved Prompting System: The new system with special, temporal, and rating tags offers finer control over the output style and quality.
Versatile Aesthetic Control: It supports a variety of illustration styles whether you need flat colors or 3D pixel art and many other style.

Potential Cons:

Prompt Format: Limited to tag-based text prompts; natural language input may not be effective
Anatomy: May struggle with complex anatomical details, particularly hand poses and finger counting
Text Generation: Text rendering in images is currently not supported and not recommended
New Characters: Recent characters may have lower accuracy due to limited training data availability
Multiple Characters: Scenes with multiple characters may require careful prompt engineering
Resolution: Higher resolutions (e.g., 1536x1536) may show degradation as training used original SDXL resolution
Style Consistency: May require specific style tags as training focused more on identity preservation than style consistency