Great to see that work is still being done on this model!

Thank you CagliostroLab for continuing to show the community how a model release should be done.

This release is very clear with no conflicting licenses which you would think would be the bare minimum but alas, no. There is also very good documentation of the model available at launch!

I hope our community takes note of what Cagliostro is doing. Many other outfits in our community have tried various things in efforts to extract money while Cagliostro does the right thing and releases their model for free and under reasonable conditions. There's a ko-fi link on their profile and compute is not free. CagliostroLabs deserves to recoup their training costs so they can continue to use their compute to benefit us!!

This might sound like I'm affiliated. I'm really not. I'm just so frustrated with how other models have been released lately that this release is a breath of fresh air. Thank you.

still censored?

when on site generation so i can do proper tests?

Oh, the heavenly model 🤲

I'm so happy you listened to the community when they had constructive criticism on the initial v4. This model is incredible. Thanks so much for your hard work!

The opt/zero version cannot get the sharpening or the colors of the 1st version, but is truthly better. Now it don't gets crazy with non portrait gens. Maybe I could prefer the zero version for make fine-tuning, but I have the question.

The zero version lacks in something? Like missing characters or styles? Thank you so much for your hard work!

It's much cleaner now. Definitely better for general use. Somehow I really miss the style of v4.0, although little bit messy, but very unique, colorful and artistic. No other model can do that.

A huge improvement in consistency and anatomy now. I really like this version! I can't say a lot on styles, I need to do more testing on those, but in characters database, it has a lot of new characters, which is awesome!

But I do notice that it lacks knowledge on old characters that NoobAI has, maybe it's because it also uses E621 in their database, so rarer cartoon or series are in the database if you prompt it correctly.

I did notice it tends to whitewash some black characters every now and then, not sure if it's because of the way I prompt it or I need to add "dark skin" every time. Either way, small gripe that can be fixed with more descriptive prompting.

Either way, I hope this keeps getting updated, I can definitely see this model surpassing NoobAI in knowledge of styles and characters. I just need to do more testing to know if it already surpassed it or not.

This model is so amazing and I am excited every day. However, I wonder if a list that was in 3.1 would allow for a wider range of expression, do you have plans to create a list? Sorry for being so busy.

The art style of this model is great, but I have a question in use, how to generate the art style of this model stably? Because sometimes/very rarely, there will be a very old style of painting, like printmaking, but the colors are very bright, like lasers. How can I avoid this?

AnimaGen redemption arc is here? this model defintly blow v3 out of the water, and in my opinion way batter than illust, noob and all vpred models. finally there is new model to play with. hopefully we see good finetunes of it. espically a fine tune with batter hands becouse the hands is a little embarrassingly bad with this model.

can i now latest animagine 4.0 is running on epsilon prediction method or v prediction method?

can i use resolution of 1360x1024?? so far this resolution have working fine for me in any model

So at this stage, can we call Animagine a standalone variation model like Illustrious and Pony?

After some testing, I think I like the regular 4.0 more than the optimized version.

It requires a bit more inpainting to fix anatomy issues, but it seems to be a lot more creative in terms of composition.

Ive been testing 4.0Opt for some hours, and i like the level of creativity in term of compositions that can create. But in my personal experience so far it has severe problems in terms of anatomy, even using the proper prompt it seems to struggle a lot. I still like it and it has potential, but the anatomy its the biggest issue so far.

I recommend ppl to try adding it with a hand LoRa, and using these quality tags:

Positive Prompt (attach to end):
(best quality:0.7),

(highres:0.95), (anti-alias:0.6)
Negative prompt (Attach to end):

worst quality, bad score, low score,

bad anatomy, bad hands, text, error, missing finger, extra digits, fewer digits,

logo, text,

(blurry:0.05),

This way you get less of a erotic girl bias, and instead just get a sharp picture of whatever you are trying to prompt. If you don't trust me, try comparing best quality:0.7 to masterpiece alone.

I've had issues making Pony and Illustrous LoRas work with it well, but it does work without burning.

I like this model. It is pretty creative, while also sticking to my prompt. Some tags are less developed, but if ppl build on it then it should work fine.

Is the a file that states which artist this model supports?

are 3.1 loras compatible with animaginexl 4.0?

Overall maintains the unique Animagine 3 art style while reducing artifacts.

Opt improves on 4 with better coloring and less weirdness/asymmetry, but seems to suffer from a bit of catastrophic forgetting with skin tones and expressions. I had to crank up the strength of "tan" to make my dark elf look like a dark elf. Facial expressions also swing between extremes.

amazing model, on par with illust and noob imo

Found this on Dreamerland. Can this model be used for commercial purposes?

animaging我的超人！最喜欢的低模没有之一！！！

Any new character list?

When do I use which version? Zero for training with kohya_ss and opt for generation with sd?

I am really impressed, first coment after more than a year around here, my congratulations, keep like this guys, my first tips for you <3

can i use illustrious and pony loras with this?

much better!

Love this model !!! Creative, Polished, Detailed, Glossy finish, !!!_______________________________________

As much as i like your aproach and skills to training, i dont think v4opt can match v3.1 tbh, yes it is an improvement but you need to reckon v3.1 is just topnotch !!!

Thank you for the Prompt Guide, it really makes a difference !!! prompt sections can be used as a template and you only change the main tags, cool approach !!! Thank you again !!!

cameras angles are awesome !!!

awesome

Any idea about parameters such as noise offset in LoRA training on v4 zero?

What parameters & upscaler do you recommend me to use for HiResFix?

animagine is always the best model for me!

Does this model support the following syntax?

prompt1 | prompt2 | prompt3,

[prompt1 | prompt2 | prompt3],

should we set clip to 2?

very nice model but hope v5 can fix the bad anatomy...

does the model compaible with XL controlnets?

great model

I have created a realistic version of this model, if you like it you can add it to your page and I can delete my upload. You get 100% credit for the original amazing model! https://civitai.com/models/1378329

Strong in the relationship between people and the background!

what hires setting did you use?
like upscaler model, denoising strength, steps, and upscale by value.

also did this model model support danbooru artist name tag as a prompt?

After fine-tuning Animagine 4 Zero on a quality dataset, hands come out disfigured about 15 out of 25 times. As much as I respect the work on Animagine 3.1, NoobAI is still king for decent anatomy.

Hands down this is the best anime focused model. It even works pretty damn good for furry content.

This honestly is a solid competitor for NoobAI/illustrious.

More Loras should be trained for this model so people can see how incredibly versatile it can truly be

Incredible for textures

我不知道我的lora用什么大模型好，能不能帮帮我

great checkpoint!
is there a list of supported artist tags other than ciloranko used in prompting guide somewhere?

This is a very solid checkpoint, however the only reason i am not using it as much is because having to keep quality tags near the end is a pain if you like experimenting. Hopefully this is changed in a future version.

I still don't understand the "BILD" thing, but I'm happy that this model is available to the public.

Among all the best models aimed at anime/gaming, this is by far, the best database. Not only it has the details for characters be them from gacha games (ala Genshin, ZZZ, or WuWa), but it's also beautiful, and plays well with detailed prompts.

My only wish is that Cagliostro keeps improving on it, or other models increase their databases to match Animagine XL 4.0.

BEST DATABASE OUT THERE!

This was my 1st experience with LoRa and Models ~

This is truly an incredible model! I enjoy using the latest version.

It's genuinely one of the best I've seen for anime art, and the way it handles textures and the relationship between characters and backgrounds is truly amazing.

The database feels very strong, especially for gacha game-style characters.

My only constructive feedback would be to continue improving on the anatomy, particularly the hands. I hope a future version can address this and wish the creator art inspiration.

Thank you so much for your amazing work. This is a solid contender for one of the top anime models out there! 💖

hello i like your model are you going to do an illustrous version ?

I'm new in the generative things. I'm trying things with fooocus. Why i keep having rainbow colors and abstract images with this checkpoint?

I used this model to train a LoRA using the same parameters as NoobAI, but the results are disappointing. The learned style doesn't look much like the training set. Is there anything specific I should look out for? I'm not sure where I went wrong

any rumours on v4.1? or animaestro 2B?

I like it! Ty ^^

🔥🔥🔥 damn, i've tried at least 40 sd, pony, ill, sdxl models
this is by far the BEST model when it comes to poses, very dynamic and varied it's INSANE!!!
every other model has the issue of the characters look as if they're posing for the camera, portrait basically, and prompting them for more dynamic poses would make them cycle between few and very similar fighting poses, they can't even do non action poses
it's also SFW by default, it won't generate NSFW content unless it is prompted to with 99% successes rate, again the only model that i've ever tested with this feature.

the only negative this model has is hands, the anatomy sometimes act funny but that's fine if you want this level of dynamic poses, but the hands kinda ruin it
i know this is an old model, but i really hope for an update that fixes the hands issue without messing with anything else since other than fingers this model is perfect
(maybe a bit of backgrounds too but that's not remotely necessary)