Cagliostro Research Lab

[Announcing Animagine XL 4.0 Updates] Greetings, Alchemists ! We hope you're doing well! 😊 We have some exciting updates to share with you today! ✨ After extensive research and testing, we are thrilled to announce several important changes to the Animagine-XL 4.0. We're releasing a new model update: Anim4gine: Opt and Zero and a full in-depth guideline. Here's what’s new: - Opt (short for Optimized) has been trained with an expanded dataset and fine-tuned for optimal performance, making it the ideal choice for general use. - Zero is the pretrained base model, designed specifically for LoRA training and fine-tuning, offering a solid foundation for further customization. 🚀 Read more details on our blog: cagliostrolab.net/posts/optimizi… 🔗 Links: Huggingface: huggingface.co/cagliostrolab/… huggingface.co/cagliostrolab/… Anifusion: anifusion.ai CivitAI: civitai.com/models/1188071… Moescape: moescape.ai/models/39711ae… Seaart : seaart.ai/models/detail/… Tensor : tensor.art/models/8247821… Shakker : shakker.ai/modelinfo/9ccd… #Anim4gine #SDXL #AIart

✨ Moescape Diffusion 2.0 is live! Built on Illustrious with knowledge up to Aug 2025 — generate the newest characters even without LoRA 💫 🎨 Try it now: moescape.ai/models/6b171d5… #AIart #aiイラスト #AI绘画 #StableDiffusion #Illustrious #MoescapeAI #MoescapeDiffusion

🔗Link: moescape.ai/models/6b171d5…

Hi, Alchemists~ We’re helping @MoescapeAi fine-tune the second iteration of Moescape Diffusion! This version is based on Illustrious XL and can utilize any available Illustrious XL–hosted LoRA models in the platform. Dataset cutoff is August 2025. Try it out on the link below👇

AnimaTensor Exclusive Available NOW! After a year, @CagliostroLab has formed an exclusive global collaboration with @TensorArt ! Introducing ✨AnimaTensor✨ ⭐️Regular: tensor.art/models/8759520… 🌟Pro: tensor.art/models/8759682… Online Generation Available Now! Online Training available on 22th June! More Info⏩ tensor.art/blackboard/Ani…

[Introducing AnimaTensor! 🎉] Hello, Alchemists! We hope this message finds you well because we are thrilled to share a big news! After a year of back-and-forth with "cooking together", we're excited to present our new model iteration in collaboration with @TensorArt: AnimaTensor! It's truly an honor to get the chance to do research and development together to train SDXL vPred, Zero Terminal SNR, and EQ-SDXL-VAE models together. This journey has been both challenging and inspiring, and we’re proud to finally share the results with you. We're launching two exclusive versions: 🌟 AnimaTensor-Pro (For Pro Members) 🔗tensor.art/models/8759682… 🌟 AnimaTensor-Regular (For All Users) 🔗tensor.art/models/8759520… ✨ Key Features: ✨ Zero-Terminal SNR¹ A noise schedule that fixes the model's inability to generate full-range brightness by ensuring the final noising step completely erases the original image signal. This corrects the training-inference discrepancy where standard schedules "leak" information, enabling generation of true blacks, whites, and high-contrast imagery. ✨ V-Prediction² A model parameterization that ensures mathematical stability when sampling at zero SNR, making it the required counterpart to a ZTSNR schedule. The model predicts the data "velocity" (v) instead of noise (ε), avoiding division-by-zero issues that occur with standard ε-prediction at the final timestep. ✨ EQ-SDXL-VAE (Equivariance-Regularized VAE)³⁴ A VAE fine-tuning method that simplifies the latent space by enforcing equivariance, which dramatically speeds up training of the subsequent generative model. It applies regularization that ensures spatial transformations (rotation, scaling) result in linear changes in latent space, accelerating convergence by 4x-7x. ✨ Contributor This Iterations: Trainer: @kayfahaarukku Co Trainer: @raelina_ai Public Relations and IT Support: @Asahina2k, @ItsMeBell20, @Scipius2121 For all the details, check out the blackboard: ➡️tensor.art/blackboard/Ani… Go try it now and show us what you make! We can't wait to see your creations! ✨ Footnote: ¹ Shanchuan Lin, Bingchen Liu, Jiashi Li, and Xiao Yang, "Common Diffusion Noise Schedules and Sample Steps are Flawed," arXiv preprint arXiv:2305.08891, 2023. ² Alex Nichol and Prafulla Dhariwal, "Improved Denoising Diffusion Probabilistic Models," arXiv preprint arXiv:2102.09672, 2021. ³ Theodoros Kouzelis, Ioannis Kakogeorgiou, Spyros Gidaris, and Nikos Komodakis, "EQ-VAE: Equivariance Regularized Latent Space for Improved Generative Image Modeling," arXiv preprint arXiv:2502.09509, 2025. ⁴ KBlueLeaf, "EQ-SDXL-VAE," Hugging Face model repository, 2024. huggingface.co/KBlueLeaf/EQ-S…

We apologize for technical inaccuracies in our AnimaTensor announcement due to internal miscommunication. We're preparing corrected specifications and will share them shortly. Thank you for your understanding. 🙏 #AnimaTensor

Hello, Alchemists! From the bottom of our heart, Happy Eid al-Fitr Mubarak to all muslims and to those whom celebrate it. May Allah Ta'ala blesses us all, forgives our wrongdoings, and grants our wishes. We would like to take this moment to apologize for our mistakes during our time here; as we are not without flaws. Thank you for supporting us, through time and time again. _____ (Image generated and retouched by @Scipius2121) #AIart #AIイラスト #Anim4gine

animagine4.0のyearの使い方を模索している #AIイラスト #anim4gine

[Anim4gine is officially on PixAI!] Hello, Alchemists~ Today, we'd like to share that we recently (re)launched Anim4gine on PixAI! We certainly acknowledged that there were Anim4gine reuploads from third party on the site. Upon further discussion, we decided to make the models available from our own PixAI account in order to make it easier for us to update and maintain the models. You may access the models on the links below. 🔗 4.0 Initial Release: pixai.art/model/18406473… 🔗 4.0 Opt version: pixai.art/model/18492274… 🔗 4.0 Zero version: pixai.art/model/18469648… We would like to apologize for the delayed official launch on this site due to claiming process. We also want to thank @PixAI_Official for reaching us and sorting things out. Hopefully, this official launch can be useful, especially for PixAI users. We are also looking for other platforms that haven't got the official release, so please don't hesitate to reach us out! Thank you and have a good day~ #AIArt #Anim4gine #SDXL

#AIart #AIイラスト #Anim4gine Serenity ——— Kinda feeling unwell since yesterday, so I may just kick back for a bit 🤒

#anim4gine 金髪デコヴァンパイア #AIイケメン部

animagineXL4.0optで色々 #AIイラスト #AIart #anim4gine

New AnimagineXL by @CagliostroLab is 🔥

You can also try our model on Anim4gine's HuggingFace Space below. Have a good day~ huggingface.co/spaces/caglios…

[Announcing Animagine XL 4.0 Updates] Greetings, Alchemists ! We hope you're doing well! 😊 We have some exciting updates to share with you today! ✨ After extensive research and testing, we are thrilled to announce several important changes to the Animagine-XL 4.0. We're releasing a new model update: Anim4gine: Opt and Zero and a full in-depth guideline. Here's what’s new: - Opt (short for Optimized) has been trained with an expanded dataset and fine-tuned for optimal performance, making it the ideal choice for general use. - Zero is the pretrained base model, designed specifically for LoRA training and fine-tuning, offering a solid foundation for further customization. 🚀 Read more details on our blog: cagliostrolab.net/posts/optimizi… 🔗 Links: Huggingface: huggingface.co/cagliostrolab/… huggingface.co/cagliostrolab/… Anifusion: anifusion.ai CivitAI: civitai.com/models/1188071… Moescape: moescape.ai/models/39711ae… Seaart : seaart.ai/models/detail/… Tensor : tensor.art/models/8247821… Shakker : shakker.ai/modelinfo/9ccd… #Anim4gine #SDXL #AIart

これは私の（絵のスタイルの）趣向とスキルの問題なんですが、SDXLで好きなスタイルの男性を生成しようとすると女性の3倍ぐらい労力がかかる……でもAnimagineの繊細さを活かせた様に思います。頑張った！ #Anim4gine #SDXL #AIart

顔をいじるテスト #AIイラスト #AIart #anim4gine

#SDXL1 #Anim4gine 各種情報はちちぷいに掲載。 chichi-pui.com/posts/adacf4e8… AnimagineXL v4.0もだいぶ慣れてきたので #ちちぷい に投稿しておきました！ たぶんそのうち調整版のv4.1とかリリースされるんじゃないかと思うんですけど、現時点でも細部の精彩さ等、v3系よりもメリットな部分も多いですね！

[Issue Notice Regarding Anim4gine] Hello, Alchemists! 🧪 We have received a plenty of feedback on #Anim4gine over this few days after public release. After careful consideration, we are currently investigating the cause of the problems on Anim4gine and fix the issues accordingly. We would like to thank everyone who shared their concerns on the model. We certainly will give an update in near future. Have a great day~ #AIイラスト #SDXL