<h3 id="please-read-our-in-depth-guideline-for-prompting-at-cagliostrolab-blog-coiueh249">Please read our in-depth Guideline for prompting at <a target="_blank" rel="ugc" href="https://cagliostrolab.net/posts/optimizing-animagine-xl-40-in-depth-guideline-and-update">Cagliostrolab Blog</a></h3><h2 id="overview-4jy2w95xn">Overview</h2>Animagine XL 4.0, also stylized as Anim4gine, is the ultimate anime-themed finetuned SDXL model and the latest installment of <a target="_blank" rel="ugc" href="https://huggingface.co/collections/Linaqruf/animagine-xl-669888c0add5adaf09754aca">Animagine XL series</a>. Despite being a continuation, the model was retrained from <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0">Stable Diffusion XL 1.0</a> with a massive dataset of 8.4M diverse anime-style images from various sources with the knowledge cut-off of January 7th 2025 and finetuned for approximately 2650 GPU hours. Similar to the previous version, this model was trained using tag ordering method for the identity and style training.With the release of Animagine XL 4.0 Opt (Optimized), the model has been further refined with an additional dataset, improving stability, anatomy accuracy, noise reduction, color saturation, and overall color accuracy. These enhancements make Animagine XL 4.0 Opt more consistent and visually appealing while maintaining the signature quality of the series.<h2 id="changelog-gqeo5zabv">Changelog</h2>- 2025-02-13 – Added Animagine XL 4.0 Opt and Animagine XL 4.0 Zero<ul><li>Better stability for more consistent outputs</li><li>Enhanced anatomy with more accurate proportions</li><li>Reduced noise and artifacts in generations</li><li>Fixed low saturation issues, resulting in richer colors</li><li>Improved color accuracy for more visually appealing results</li></ul>- 2025-01-24 – Initial release<h2 id="model-details-ldqhi7qps">Model Details</h2><ul><li>Developed by: <a target="_blank" rel="ugc" href="https://github.com/cagliostrolab">Cagliostro Research Lab</a></li><li>Model type: Diffusion-based text-to-image generative model</li><li>License: <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/LICENSE.md">CreativeML Open RAIL++-M</a></li><li>Model Description: This is a model that can be used to generate and modify specifically anime-themed images based on text prompt</li><li>Fine-tuned from: <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0">Stable Diffusion XL 1.0</a></li></ul><h2 id="usage-guidelines-oepvp8jeo">Usage Guidelines</h2>The summary can be seen in the image for the prompt guideline.<img src="https://cdn-uploads.huggingface.co/production/uploads/64c088660a4d02f37a965f6c/YPe3MCnQAHM7nCZ1vQ7vI.png" /><h3 id="1.-prompt-structure-s2bolxh5w">1. Prompt Structure</h3>The model was trained with tag-based captions and the tag-ordering method. Use this structured template:<pre><code>1girl/1boy/1other, character name, from which series, rating, everything else in any order and end with quality enhancement
</code></pre><h3 id="2.-quality-enhancement-tags-wy9kaqsld">2. Quality Enhancement Tags</h3>Add these tags at the end of your prompt:<pre><code>masterpiece, high score, great score, absurdres
</code></pre><h3 id="3.-recommended-negative-prompt-2jegnk272">3. Recommended Negative Prompt</h3><pre><code>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
</code></pre><h3 id="4.-optimal-settings-qh99nn1i6">4. Optimal Settings</h3><ul><li>CFG Scale: 4-7 (5 Recommended)</li><li>Sampling Steps: 25-28 (28 Recommended)</li><li>Preferred Sampler: Euler Ancestral (Euler a)</li></ul><h3 id="5.-recommended-resolutions-7zj90ju2b">5. Recommended Resolutions</h3><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/bf19bf0e-0fbb-4d6d-9ce3-df52c9f4052c/width=525/bf19bf0e-0fbb-4d6d-9ce3-df52c9f4052c.jpeg" /><h3 id="6.-final-prompt-structure-example-vokz3nmia">6. Final Prompt Structure Example</h3><pre><code>1girl, firefly \(honkai: star rail\), honkai \(series\), honkai: star rail, safe, casual, solo, looking at viewer, outdoors, smile, reaching towards viewer, night, masterpiece, high score, great score, absurdres
</code></pre><h2 id="special-tags-kit0e1jzn">Special Tags</h2>The model supports various special tags that can be used to control different aspects of the image generation process. These tags are carefully weighted and tested to provide consistent results across different prompts.<h3 id="quality-tags-6f9nx9zlc">Quality Tags</h3>Quality tags are fundamental controls that directly influence the overall image quality and detail level. Available quality tags:<ul><li><code>masterpiece</code></li><li><code>best quality</code></li><li><code>low quality</code></li><li><code>worst quality</code></li></ul>Score TagsScore tags provide a more nuanced control over image quality compared to basic quality tags. They have a stronger impact on steering output quality in this model. Available score tags:<ul><li><code>high score</code></li><li><code>great score</code></li><li><code>good score</code></li><li><code>average score</code></li><li><code>bad score</code></li><li><code>low score</code></li></ul><h3 id="temporal-tags-8eivwxggr">Temporal Tags</h3>Temporal tags allow you to influence the artistic style based on specific time periods or years. This can be useful for generating images with era-specific artistic characteristics. Supported year tags:<ul><li><code>year 2005</code></li><li><code>year {n}</code></li><li><code>year 2025</code></li></ul><h3 id="rating-tags-qjdv8wcse">Rating Tags</h3>Rating tags help control the content safety level of generated images. These tags should be used responsibly and in accordance with applicable laws and platform policies. Supported ratings:<ul><li><code>safe</code></li><li><code>sensitive</code></li><li><code>nsfw</code></li><li><code>explicit</code></li></ul><h2 id="training-information-twmrly5xa">Training Information</h2>The model was trained using state-of-the-art hardware and optimized hyperparameters to ensure the highest quality output. Below are the detailed technical specifications and parameters used during the training process:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/b86b2598-11ec-4c18-9c45-71e84b0f1d36/width=525/b86b2598-11ec-4c18-9c45-71e84b0f1d36.jpeg" /><h2 id="acknowledgement-bulem1yaa">Acknowledgement</h2>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:<ol><li><a target="_blank" rel="ugc" href="https://moescape.ai/">Moescape AI</a>: Our invaluable collaboration partner in model distribution and testing</li><li>Lesser Rabbit: For providing essential computing and research grants</li><li><a target="_blank" rel="ugc" href="https://github.com/kohya-ss">Kohya SS</a>: For developing the comprehensive open-source training framework</li><li><a target="_blank" rel="ugc" href="https://github.com/discus0434">discus0434</a>: For creating the industry-leading open-source Aesthetic Predictor 2.5</li><li>Early testers: For their dedication in providing critical feedback and thorough quality assurance</li></ol><h2 id="contributors-m7l3lm6zm">Contributors</h2>We extend our heartfelt appreciation to our dedicated team members who have contributed significantly to this project, including but not limited to:<h3 id="model-vei5wpijv">Model</h3><ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/kayfahaarukku">KayfaHaarukku</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Raelina">Raelina</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Linaqruf">Linaqruf</a></li></ul><h3 id="gradio-xzuggxl0r">Gradio</h3><ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/DamarJati">Damar Jati</a></li></ul><h3 id="relations-finance-and-quality-assurance-oh4x2zukk">Relations, finance, and quality assurance</h3><ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Scipius2121">Scipius</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Asahina2K">Asahina</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/ItsMeBell">Bell</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Boboiazumi">BoboiAzumi</a></li></ul><h3 id="data-xpygvykjf">Data</h3><ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/paripi">Pomegranata</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Kr1SsSzz">Kr1SsSzz</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/saikanov">Fiqi</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/williamsoeherman">William Adams Soeherman</a></li></ul><h2 id="-9nkmsuswj"></h2><h2 id="fundraising-are-now-open-again!-3d7xjdwru">Fundraising Are Now Open Again!</h2>We're excited to introduce new fundraising methods through GitHub Sponsors to support training, research, and model development. Your support helps us push the boundaries of what's possible with AI.You can help us with:<ul><li>Donate: Contribute via ETH or USDT to the address below.</li><li>Share: Spread the word about our models and share your creations!</li><li>Feedback: Let us know how we can improve.</li></ul>Donation Address:ETH/USDT/USDC(e): <code>0xd8A1dA94BA7E6feCe8CfEacc1327f498fCcBFC0C</code>Github Sponsor: <a target="_blank" rel="ugc" href="https://github.com/sponsors/cagliostrolab/">https://github.com/sponsors/cagliostrolab/</a> Why do we use Cryptocurrency?: When we initially opened fundraising through Ko-fi and using PayPal as withdrawal methods, our PayPal account was flagged and eventually banned, despite our efforts to explain the purpose of our project. Unfortunately, this forced us to refund all donations and left us without a reliable way to receive support. To avoid such issues and ensure transparency, we have now switched to cryptocurrency as the way to raise the fund. Want to Donate in Non-Crypto Currency? Although we had a bad experience with Paypal, and you’d like to support us but prefer not to use cryptocurrency, feel free to contact us via <a target="_blank" rel="ugc" href="https://discord.gg/cqh9tZgbGc">Discord Server</a> for alternative donation methods. Join Our Discord Server Feel free to join our discord server: <a target="_blank" rel="ugc" href="https://discord.gg/cqh9tZgbGc">https://discord.gg/cqh9tZgbGc</a> <h2 id="limitations-ocg1xafco">Limitations</h2><ul><li>Prompt Format: Limited to tag-based text prompts; natural language input may not be effective</li><li>Anatomy: May struggle with complex anatomical details, particularly hand poses and finger counting</li><li>Text Generation: Text rendering in images is currently not supported and not recommended</li><li>New Characters: Recent characters may have lower accuracy due to limited training data availability</li><li>Multiple Characters: Scenes with multiple characters may require careful prompt engineering</li><li>Resolution: Higher resolutions (e.g., 1536x1536) may show degradation as training used original SDXL resolution</li><li>Style Consistency: May require specific style tags as training focused more on identity preservation than style consistency</li></ul><h2 id="license-3m7yoboxx">License</h2>This model adopts the original <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/LICENSE.md">CreativeML Open RAIL++-M License</a> from Stability AI without any modifications or additional restrictions. The license terms remain exactly as specified in the original SDXL license, which includes:<ul><li>✅ Permitted: Commercial use, modifications, distributions, private use</li><li>❌ Prohibited: Illegal activities, harmful content generation, discrimination, exploitation</li><li>⚠️ Requirements: Include license copy, state changes, preserve notices</li><li>📝 Warranty: Provided "AS IS" without warranties</li></ul>Please refer to the <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/LICENSE.md">original SDXL license</a> for the complete and authoritative terms and conditions.

Animagine XL 4.0

animagineXL40_v40.safetensors

v4.0

v4 Zero

v4 Opt

OverviewAnimagine XL 4.0, also stylized as Anim4gine, is the ultimate anime-themed finetuned SDXL model and the latest installment of <a target="_blank" rel="ugc" href="https://huggingface.co/collections/Linaqruf/animagine-xl-669888c0add5adaf09754aca">Animagine XL series</a>. Despite being a continuation, the model was retrained from <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0">Stable Diffusion XL 1.0</a> with a massive dataset of 8.4M diverse anime-style images from various sources with the knowledge cut-off of January 7th 2025 and finetuned for approximately 2650 GPU hours. Similar to the previous version, this model was trained using tag ordering method for the identity and style training.Model Details<ul><li>Developed by: <a target="_blank" rel="ugc" href="https://github.com/cagliostrolab">Cagliostro Research Lab</a></li><li>Model type: Diffusion-based text-to-image generative model</li><li>License: <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/LICENSE.md">CreativeML Open RAIL++-M</a></li><li>Model Description: This is a model that can be used to generate and modify specifically anime-themed images based on text prompt</li><li>Fine-tuned from: <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0">Stable Diffusion XL 1.0</a></li></ul>Downstream Use<ol><li>Use this model in our Hugging Face Spaces</li><li>Use it in ComfyUI or Stable Diffusion Webui</li><li>Use it with 🧨 diffusers</li></ol>🧨 Diffusers Installation1. Install Required Librariespip install diffusers transformers accelerate safetensors --upgrade2. Example CodeThe example below uses lpw_stable_diffusion_xl pipeline which enables better handling of long, weighted and detailed prompts. The model is already uploaded in FP16 format, so there's no need to specify variant="fp16" in the from_pretrained call.import torch from diffusers import StableDiffusionXLPipeline pipe = StableDiffusionXLPipeline.from_pretrained( "cagliostrolab/animagine-xl-4.0", torch_dtype=torch.float16, use_safetensors=True, custom_pipeline="lpw_stable_diffusion_xl", add_watermarker=False ) <a target="_blank" rel="noopener noreferrer nofollow" href="http://pipe.to">pipe.to</a>('cuda') prompt = "1girl, arima kana, oshi no ko, hoshimachi suisei, hoshimachi suisei \(1st costume\), cosplay, looking at viewer, smile, outdoors, night, v, masterpiece, high score, great score, absurdres" negative_prompt = "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" image = pipe( prompt, negative_prompt=negative_prompt, width=832, height=1216, guidance_scale=6, num_inference_steps=25 ).images[0] <a target="_blank" rel="noopener noreferrer nofollow" href="http://image.save">image.save</a>("./arima_kana.png")Usage Guidelines1. Prompt StructureThe model was trained with tag-based captions and the tag-ordering method. Use this structured template:1girl/1boy/1other, character name, from which series, everything else in any order.2. Quality Enhancement TagsAdd these tags at the start or end of your prompt:masterpiece, high score, great score, absurdres3. Recommended Negative Promptlowres, 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, blurry4. Optimal Settings<ul><li>CFG Scale: 5-7 (6 Recommended)</li><li>Sampling Steps: 25-28 (25 Recommended)</li><li>Preferred Sampler: Euler Ancestral (Euler a)</li></ul>5. Recommended ResolutionsOrientation Dimensions Aspect Ratio6. Final Prompt Structure Examplemasterpiece, high score, great score, absurdres, 1girl, firefly \(honkai: star rail\), honkai \(series\), honkai: star rail, casual, solo, looking at viewer, outdoors, smile, reaching towards viewer, nightSpecial TagsThe model supports various special tags that can be used to control different aspects of the image generation process. These tags are carefully weighted and tested to provide consistent results across different prompts.Quality TagsQuality tags are fundamental controls that directly influence the overall image quality and detail level. Available quality tags:<ul><li>masterpiece</li><li>best quality</li><li>low quality</li><li>worst quality</li></ul>Score TagsScore tags provide a more nuanced control over image quality compared to basic quality tags. They have a stronger impact on steering output quality in this model. Available score tags:<ul><li>high score</li><li>great score</li><li>good score</li><li>average score</li><li>bad score</li><li>low score</li></ul>Temporal TagsTemporal tags allow you to influence the artistic style based on specific time periods or years. This can be useful for generating images with era-specific artistic characteristics. Supported year tags:<ul><li>year 2005</li><li>year {n}</li><li>year 2025</li></ul>Rating TagsRating tags help control the content safety level of generated images. These tags should be used responsibly and in accordance with applicable laws and platform policies. Supported ratings:<ul><li>safe</li><li>sensitive</li><li>nsfw</li><li>explicit</li></ul>Training InformationThe model was trained using state-of-the-art hardware and optimized hyperparameters to ensure the highest quality output. Below are the detailed technical specifications and parameters used during the training process:AcknowledgementThis 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:<ol><li><a target="_blank" rel="ugc" href="https://moescape.ai/">Moescape AI</a>: Our invaluable collaboration partner in model distribution and testing</li><li>Lesser Rabbit: For providing essential computing and research grants</li><li><a target="_blank" rel="ugc" href="https://github.com/kohya-ss">Kohya SS</a>: For developing the comprehensive open-source training framework</li><li><a target="_blank" rel="ugc" href="https://github.com/discus0434">discus0434</a>: For creating the industry-leading open-source Aesthetic Predictor 2.5</li><li>Early testers: For their dedication in providing critical feedback and thorough quality assurance</li></ol>ContributorsWe extend our heartfelt appreciation to our dedicated team members who have contributed significantly to this project, including but not limited to:Model<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/kayfahaarukku">KayfaHaarukku</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Raelina">Raelina</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Linaqruf">Linaqruf</a></li></ul>Gradio<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/DamarJati">Damar Jati</a></li></ul>Relations, finance, and quality assurance<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Scipius2121">Scipius</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Asahina2K">Asahina</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/ItsMeBell">Bell</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Boboiazumi">BoboiAzumi</a></li></ul>Data<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/paripi">Pomegranata</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Kr1SsSzz">Kr1SsSzz</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/saikanov">Fiqi</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/williamsoeherman">William Adams Soeherman</a></li></ul>Fundraising Are Now Open Again!We’re excited to reopen Fundraising to fund new training, research, and modeldevelopment. Your support helps us push the boundaries of what’s possible with AI.You can help us with:<ul><li>Donate: Contribute via ETH or USDT to the address below.</li><li>Share: Spread the word about our models and share your creations!</li><li>Feedback: Let us know how we can improve.</li></ul>Donation Address:ETH/USDT/USDC(e):&nbsp;0xd8A1dA94BA7E6feCe8CfEacc1327f498fCcBFC0C Why do we use Cryptocurrency?: When we initially opened fundraising through Ko-fi and using PayPal as withdrawal methods, our PayPal account was flagged and eventually banned, despite our efforts to explain the purpose of our project. Unfortunately, this forced us to refund all donations and left us without a reliable way to receive support. To avoid such issues and ensure transparency, we have now switched to cryptocurrency as the way to raise the fund. Want to Donate in Non-Crypto Currency? Although we had a bad experience with Paypal, and you’d like to support us but prefer not to use cryptocurrency, feel free to contact us via <a target="_blank" rel="ugc" href="https://discord.gg/cqh9tZgbGc">Discord Server</a> for alternative donation methods. Join Our Discord Server Feel free to join our discord server: <a target="_blank" rel="noopener noreferrer nofollow" href="https://discord.gg/cqh9tZgbGc">https://discord.gg/cqh9tZgbGc</a>Limitations<ul><li>Prompt Format: Limited to tag-based text prompts; natural language input may not be effective</li><li>Anatomy: May struggle with complex anatomical details, particularly hand poses and finger counting</li><li>Text Generation: Text rendering in images is currently not supported and not recommended</li><li>New Characters: Recent characters may have lower accuracy due to limited training data availability</li><li>Multiple Characters: Scenes with multiple characters may require careful prompt engineering</li><li>Resolution: Higher resolutions (e.g., 1536x1536) may show degradation as training used original SDXL resolution</li><li>Style Consistency: May require specific style tags as training focused more on identity preservation than style consistency</li></ul>LicenseThis model adopts the original <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/LICENSE.md">CreativeML Open RAIL++-M License</a> from Stability AI without any modifications or additional restrictions. The license terms remain exactly as specified in the original SDXL license, which includes:<ul><li>✅ Permitted: Commercial use, modifications, distributions, private use</li><li>❌ Prohibited: Illegal activities, harmful content generation, discrimination, exploitation</li><li>⚠️ Requirements: Include license copy, state changes, preserve notices</li><li>📝 Warranty: Provided "AS IS" without warranties</li></ul>Please refer to the <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/LICENSE.md">original SDXL license</a> for the complete and authoritative terms and conditions.

Overview
Animagine XL 4.0, also stylized as Anim4gine, is the ultimate anime-themed finetuned SDXL model and the latest installment of Animagine XL series. Despite being a continuation, the model was retrained from Stable Diffusion XL 1.0 with a massive dataset of 8.4M diverse anime-style images from various sources with the knowledge cut-off of January 7th 2025 and finetuned for approximately 2650 GPU hours. Similar to the previous version, this model was trained using tag ordering method for the identity and style training.

Model Details
Developed by: Cagliostro Research Lab

Model type: Diffusion-based text-to-image generative model

License: CreativeML Open RAIL++-M

Model Description: This is a model that can be used to generate and modify specifically anime-themed images based on text prompt

Fine-tuned from: Stable Diffusion XL 1.0

Downstream Use
Use this model in our Hugging Face Spaces

Use it in ComfyUI or Stable Diffusion Webui

Use it with 🧨 diffusers

🧨 Diffusers Installation
1. Install Required Libraries
pip install diffusers transformers accelerate safetensors --upgrade
2. Example Code
The example below uses lpw_stable_diffusion_xl pipeline which enables better handling of long, weighted and detailed prompts. The model is already uploaded in FP16 format, so there's no need to specify variant="fp16" in the from_pretrained call.

import torch
from diffusers import StableDiffusionXLPipeline

pipe = StableDiffusionXLPipeline.from_pretrained(
    "cagliostrolab/animagine-xl-4.0",
    torch_dtype=torch.float16,
    use_safetensors=True,
    custom_pipeline="lpw_stable_diffusion_xl",
    add_watermarker=False
)
pipe.to('cuda')

prompt = "1girl, arima kana, oshi no ko, hoshimachi suisei, hoshimachi suisei \(1st costume\), cosplay, looking at viewer, smile, outdoors, night, v, masterpiece, high score, great score, absurdres"
negative_prompt = "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"

image = pipe(
    prompt,
    negative_prompt=negative_prompt,
    width=832,
    height=1216,
    guidance_scale=6,
    num_inference_steps=25
).images[0]

image.save("./arima_kana.png")
Usage Guidelines
1. Prompt Structure
The model was trained with tag-based captions and the tag-ordering method. Use this structured template:

1girl/1boy/1other, character name, from which series, everything else in any order.
2. Quality Enhancement Tags
Add these tags at the start or end of your prompt:

masterpiece, high score, great score, absurdres
3. Recommended Negative Prompt
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
4. Optimal Settings
CFG Scale: 5-7 (6 Recommended)

Sampling Steps: 25-28 (25 Recommended)

Preferred Sampler: Euler Ancestral (Euler a)

5. Recommended Resolutions
Orientation Dimensions Aspect Ratio



6. Final Prompt Structure Example
masterpiece, high score, great score, absurdres, 1girl, firefly \(honkai: star rail\), honkai \(series\), honkai: star rail, casual, solo, looking at viewer, outdoors, smile, reaching towards viewer, night
Special Tags
The model supports various special tags that can be used to control different aspects of the image generation process. These tags are carefully weighted and tested to provide consistent results across different prompts.

Quality Tags
Quality tags are fundamental controls that directly influence the overall image quality and detail level. Available quality tags:

masterpiece

best quality

low quality

worst quality

Score Tags

Score tags provide a more nuanced control over image quality compared to basic quality tags. They have a stronger impact on steering output quality in this model. Available score tags:

high score

great score

good score

average score

bad score

low score

Temporal Tags
Temporal tags allow you to influence the artistic style based on specific time periods or years. This can be useful for generating images with era-specific artistic characteristics. Supported year tags:

year 2005

year {n}

year 2025

Rating Tags
Rating tags help control the content safety level of generated images. These tags should be used responsibly and in accordance with applicable laws and platform policies. Supported ratings:

safe

sensitive

nsfw

explicit

Training Information
The model was trained using state-of-the-art hardware and optimized hyperparameters to ensure the highest quality output. Below are the detailed technical specifications and parameters used during the training process:



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 testing

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

Contributors
We extend our heartfelt appreciation to our dedicated team members who have contributed significantly to this project, including but not limited to:

Model
KayfaHaarukku

Raelina

Linaqruf

Gradio
Damar Jati

Relations, finance, and quality assurance
Scipius

Asahina

Bell

BoboiAzumi

Data
Pomegranata

Kr1SsSzz

Fiqi

William Adams Soeherman

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/USDC(e): 0xd8A1dA94BA7E6feCe8CfEacc1327f498fCcBFC0C

Why do we use Cryptocurrency?:
When we initially opened fundraising through Ko-fi and using PayPal as withdrawal methods, our PayPal account was flagged and eventually banned, despite our efforts to explain the purpose of our project. Unfortunately, this forced us to refund all donations and left us without a reliable way to receive support. To avoid such issues and ensure transparency, we have now switched to cryptocurrency as the way to raise the fund.

Want to Donate in Non-Crypto Currency?
Although we had a bad experience with Paypal, and you’d like to support us but prefer not to use cryptocurrency, feel free to contact us via Discord Server for alternative donation methods.

Join Our Discord Server

Feel free to join our discord server: https://discord.gg/cqh9tZgbGc

Limitations
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

License
This model adopts the original CreativeML Open RAIL++-M License from Stability AI without any modifications or additional restrictions. The license terms remain exactly as specified in the original SDXL license, which includes:

✅ Permitted: Commercial use, modifications, distributions, private use

❌ Prohibited: Illegal activities, harmful content generation, discrimination, exploitation

⚠️ Requirements: Include license copy, state changes, preserve notices

📝 Warranty: Provided "AS IS" without warranties

Please refer to the original SDXL license for the complete and authoritative terms and conditions.

Files with SHA256: 1d5b43ff75b6ab598502d4c779d2fbfa3dceca51c60c3b609640a60772333916