Hello, Alchemists, CagliostroLab DevRel here~ 👋​

Today’s a good day for us, and hopefully yours, too. “Why is it?”, you might ask. 🤔​ After months of lengthy training processes, it’s finally coming to this day. Today, it is officially marked as the day Animagine-XL v4.0 a.k.a. Anim4gine is released to the public!🥳🎉 We cannot hold up our excitement to see this model can finally be used globally. So, what makes this model different from the previous iterations? What makes it special? Let’s take a look, shall we?~👀

Back to the Origin#

As you may have noticed, previous iterations were built on top of its predecessor. We’ve got Animagine-XL v3.1 that was based on v3.0, which based on v2.0. This time, we are shifting back to the root: Anim4gine was retrained directly from SDXL. With the whopping amount of 8.4 millions anime-styled images dataset, we are aiming to not only retain many characters that was already available on v3.1, but also expand the character roster further. Anim4gine has this knowledge up to January 7th, 2025, ensuring most of the recent popular characters from games and animes can be generated with this model.

Crazy Mita from Miside

Hoshimi Miyabi from Zenless Zone Zero

Zani from Wuthering Waves

Arima Kana cosplays as Hoshimachi Suisei

Besides that, this model went through lengthy processes in order to improve the quality. Curating the dataset itself took several months. On top of that, the fine-tuning process took about 2650 GPU hours. So, hopefully it will turn out great! Now that we all know the differences and its features, where and how can we use this model?

Anim4gine’s 404 101#

First of all, and as usual, we are uploading the model to several sites that are accessible to the public. We will provide the links to the sites right below. 👇​

Now that we know where to access it, next up is, of course, how to use this model properly! In order to achieve better generation results, you may follow the guideline provided on the model card. But don’t worry, we’ll provide the brief explanation here~

There are additional tags for improving the odds on achieving better results, namely quality tags and score tags. These following tags can be put before or after the prompting order above.

Besides quality-enhancing tags, you may want to insert these following special tags to achieve a specific style. You may place these in general tags within the order.

With all of these combined, the finalized, recommended prompting order is something like this, as an example.

We also recommend the following tags to be put on the negative prompts.

Moving on to the settings⚙️​

As Animagine models were based on SDXL, this rendition is also following SDXL’s recommended resolution. This model optimally works using Euler Ancestral as its sampling method with 5–7 CFG Scale (6 recommended) and 25–28 sampling steps (25 recommended). Still not too different from Animagine-XL v3.1, right? Hope this brief explanation will be useful~ 😉​

Acknowledgement#

This long-term project would not have been possible without the groundbreaking work, innovative contributions, and comprehensive documentation provided by Stability AI, Novel AI, and Waifu Diffusion Team. We are especially grateful for the kickstarter grant from Main that enabled us to progress beyond v2. For this iteration, we would like to express our sincere gratitude to everyone in the community for their continuous support, particularly:

• Moescape AI: Our invaluable collaboration partner in model distribution and giving the opportunity for the model’s early access;
• Lesser Rabbit: For providing essential computing and research grants;
• Kohya SS: For developing the comprehensive open-source training framework;
• discus0434: For creating the industry-leading open-source Aesthetic Predictor 2.5;
• Early testers: For their dedication in providing critical feedback and thorough quality assurance.

But Wait, There’s More#

You heard (or saw) that right! 😳​

There are more things that we plan to do over the span of this year. We have a rundown list for our projects in 2025, but of course it is still Subject to Change!

Fundraising Are Now Open Again!#

We’re excited to reopen Fundraising to fund new training, research, and model development. Your support helps us push the boundaries of what’s possible with AI.

You can help us with::

• Donate: Contribute via ETH or USDT to the address below.

• Share: Spread the word about our models and share your creations!

• Feedback: Let us know how we can improve.

Donation Address:

ETH/USDT(e): 0xd8A1dA94BA7E6feCe8CfEacc1327f498fCcBFC0C

Join Our Discord Server#

Feel free to join our discord server

Afterwords#

Thank you for supporting us in various ways. We are so happy and excited to see how all of these things unfold in near future. See you later, Alchemists~ 🔮​