<h1 id="full-report:-7ryhjg9ab">Full Report:</h1><h1 id="https:nieta-art.feishu.cnwikippwqwvdzjine5kkuhrtcesn6nmh-u7n1bcsgi"><a target="_blank" rel="ugc" href="https://nieta-art.feishu.cn/wiki/PpwqwVDzjiNE5kkUhRtcEsn6nmh">https://nieta-art.feishu.cn/wiki/PpwqwVDzjiNE5kkUhRtcEsn6nmh</a></h1><h3 id="first-release-address:-e3rcieloz">First release address:</h3>Civitai - <a target="_blank" rel="ugc" href="https://civarchive.com/models/410737">https://civarchive.com/models/410737</a>Liblib - <a target="_blank" rel="ugc" href="https://www.liblib.art/modelinfo/55b06e35dd724862b3524ff00b069fe8">https://www.liblib.art/modelinfo/55b06e35dd724862b3524ff00b069fe8</a><img src="https://nieta-art.feishu.cn/space/api/box/stream/download/asynccode/?code=OTQ5MTg2MmE2YzI0MTM2YWQwMzkyMjMwODliYjY4NTJfcmx3cEJwUkJDVmgzNmN4YXd1NU5xWG5UcWxPRVNzTTlfVG9rZW46UWtiSmIzajVjb0N6aTV4bUlScmM5dzJObmRHXzE3MjE2OTk2OTk6MTcyMTcwMzI5OV9WNA" /><h1 id="i.-overview-06qjsecyx">I. Overview</h1><ul><li>We have updated Neta Art XL V2.0. Comparing to version 1.0, we have optimized the character's posture dynamics, further strengthened hand stability, and are very good at telling stories with atmospheric epic scenes.</li><li>Main motivation:<ul><li>In V1.0, the default poses generated by our characters are relatively fixed, while the new version enhances the sense of dynamics, diverse character camera expressions, and more precise prompt control.</li><li>On the basis of increasing more stylistic diversity, more stable anatomy has been maintained, especially the probability of having five fingers (instead of four or six) on the hands is now higher!</li><li>Further training methods for issues such as Rectified Flow and Noise Offset have been better deployed. The current picture has a strong sense of epic, and can now show very bright and dark scenery (as shown in the figure below).</li></ul></li></ul><img src="https://nieta-art.feishu.cn/space/api/box/stream/download/asynccode/?code=N2VjYzUwNmY4N2Y4NThiNWU2ZmFlYjNhYzk5NjliODFfNlR4UHZwaENUVzNMQkFPaEJGOEd1YUo0dzBISjB0eTNfVG9rZW46WlZCVWJaQVJVb1EwOHZ4YWhLOGNydzlFblJoXzE3MjE2OTk2OTk6MTcyMTcwMzI5OV9WNA" />{ "prompt": "1girl, full moon, moonlight shadow, cinematic lighting, battle field, fighting armies, castle, warriors around, waving, looking at viewer, Heterochromatic pupil, sitting on a cliff, very dark, epic scene, inferno, cowboy shot, depressed, angel wings, solo, white long hair, wings bangs, beautiful color, amazing quality,", "negative_prompt": "nsfw, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name, rating: sensitive, low contrast, signature, flexible deformity, abstract, low contrast, ", "resolution": "1344 x 768", "guidance_scale": 8, "num_inference_steps": 28, "sampler": "Euler a", "use_lora": null, "use_upscaler": { "upscale_method": "bislerp", "upscaler_strength": 0.65, "upscale_by": 1.5, "upscale_cfg": 11, "new_resolution": "2016 x 1152" }}<h3 id="prompt-guide-ti65joji7">Prompt guide</h3>In Neta V2, we re-trained based on ArtiWaifu Diffusion V1.0 (AWA1). Therefore, we used the prompt order summarized in ArtiWaifu:Tag order: art style ( by xxx ) - &gt; character ( 1 frieren (sousou no frieren) ) - &gt; race (elf) - &gt; composition (cowboy shot) - &gt; painting style ( impasto ) - &gt; theme (fantasy theme) - &gt; main environment (in the forest, at day) - &gt; background (gradient background) - &gt; action (sitting on ground) - &gt; expression (expressionless) - &gt; main characteristics (white hair) - &gt; other characteristics (Twintails, green eyes, parted lips) - &gt; clothing (wearing a white dress) - &gt; clothing accessories (frills) - &gt; other items (holding a magic wand) - &gt; secondary environment (grass, sunshine) - &gt; aesthetics ( beautiful color , detailed ) - &gt; quality ( best quality) - &gt; secondary description (birds, clouds, butterflies)<h1 id="v.-training-strategies-7ea3lvyd6">V. Training strategies</h1>Model Training is divided into three stages.<ol><li>Rough practice: Teach models basic knowledge and concepts of the second dimension. Learning Rate: Constants 1e-5 (unet), 7.5e-6 (text encoder) Equivalent batch size: 96 Optimizer: AdaFactor (False relative_step, scale_parameter and warmup_init) Noise-free offset Time step sampling: LogitNormal (ln3,1) Minimum signal to noise ratio: 5 Note: Since the basic model has most of the prior knowledge, this stage is skipped in actual training.</li></ol><ol><li>Concept Reinforcement: This stage teaches and reinforces all the concepts that the model needs to learn, including artists and characters with less data. The core strategy is data weighting - increasing the weight of data with less data, that is, the number of times it is repeatedly trained in a single epoch, and supplemented with other strategies to help concept learning, such as randomly removing core features of characters. The training parameters for the concept reinforcement stage are exactly the same as those for rough practice.</li><li>Refinement: This stage uses data with quality ratings of best and amazing to continue fine-tuning the model, freezes the text encoder, enables label randomization algorithm, adds noise offset of 0.0357, and uniformly samples time steps during training. Learning Rate: Constants 2e-6 (unet), 1e-6 (text encoder) Equivalent batch size: 48 Optimizer: AdaFactor (False relative_step, scale_parameter and warmup_init) Noise offset: 0.0357 Time step sampling: uniform distribution (same as normal training) Minimum signal to noise ratio: 5</li></ol><h1 id="vi.-terms-of-use-ywuas7vk2">VI. Terms of Use</h1><ul><li>The model was developed by Nieta Lab : <a target="_blank" rel="ugc" href="http://Neta.art">Neta.art</a> Lab - <a target="_blank" rel="ugc" href="https://civarchive.com/user/neta_art">https://civarchive.com/user/neta_art</a></li><li>Collaborative participants:<ul><li>Euge: <a target="_blank" rel="ugc" href="https://civarchive.com/user/Euge_">https://civarchive.com/user/Euge_</a></li><li>汤人烂: <a target="_blank" rel="ugc" href="https://space.bilibili.com/8594480">https://space.bilibili.com/8594480</a></li><li>Chenkin: <a target="_blank" rel="ugc" href="https://civarchive.com/user/Chenkin">https://civarchive.com/user/Chenkin</a></li><li>Bo Dai: <a target="_blank" rel="ugc" href="https://daibo.info/">https://daibo.info/</a></li></ul></li><li>Terms of Use:<ul><li>The model was developed from ArtiWaifu and AAM XL using <a target="_blank" rel="ugc" href="https://freedevproject.org/faipl-1.0-sd/">Fair AI Public License 1.0-SD</a></li><li>If you later modify, merge, or develop the model again, you need to open source the subsequent derived model .</li></ul></li></ul><h1 id="vii.-summary-and-outlook-fwa029p8k">VII. Summary and Outlook</h1>Neta Art DiT follow-up training. Stay tuned and test our product for free: <a target="_blank" rel="ugc" href="https://neta.art">https://neta.art</a> .Bilibili: <a target="_blank" rel="ugc" href="https://space.bilibili.com/505727005">https://space.bilibili.com/505727005</a>RED: <a target="_blank" rel="ugc" href="https://www.xiaohongshu.com/user/profile/63f2ebf2000000001001e206">https://www.xiaohongshu.com/user/profile/63f2ebf2000000001001e206</a>Twitter: <a target="_blank" rel="ugc" href="https://twitter.com/netaart_ai">https://twitter.com/netaart_ai</a>Civitai：<a target="_blank" rel="ugc" href="https://civarchive.com/user/neta_art">https://civarchive.com/user/neta_art</a>

Neta Art XL

netaArtXL_v10.safetensors

v1.0

v2.0

v1.0 DPO

Neta Art XL 2D Characters

<h3 style="text-align: start;">⚠️ Note: Please use CFG = 11 for model effect testing</h3>Complete release report: <a href="https://nieta-art.feishu.cn/wiki/ARWLw99w7ikjaSkoV99cxmsOn5g" target="_blank">https://nieta-art.feishu.cn/wiki/ARWLw99w7ikjaSkoV99cxmsOn5g</a> Supported first-step image(before highres fix）generation ratios: '1:1': [1024, 1024], &nbsp;'3:4': [896, 1152], &nbsp;'4:3': [1152, 896], &nbsp;'16:9': [1344, 768], &nbsp;'9:16': [768, 1344], <h1>1. Overview</h1><ul><li>Introducing Neta Art XL V1.0, the easiest-to-use SDXL model to date.</li><li>Keywords: Best character coverage, vivid storytelling, diverse styles, stable human anatomy.</li><li>Main motivation: Better multi-style stability and human anatomy for character visual storytelling purposes: providing prompt sorting rules, easier to follow Prompts; striking a good balance between better knowledge and stability. Maintaining the aesthetic upper standards of multiple anime art styles while keeping the output baseline attractive to the general user. Less use of character/style/artist Lora, making it easier to leverage static model acceleration technologies (like TensorRT).</li></ul> Contains 13500+ character reference list: please refer to A3.1 character list and training data information in the report<img src="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/d6409f95f60ae751f9f154a4324a6f7bd48787f6b9616a0da4e9ae25224dc864.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" alt="" data-href="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/d6409f95f60ae751f9f154a4324a6f7bd48787f6b9616a0da4e9ae25224dc864.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" style=""/><h3>Prompt Guide</h3>To avoid ambiguity in text prompts and to leave room for very complex scenes (such as multi-character scenes), we found that enforcing order in prompts can bring better instruction-following behavior (learning from NAI3 / Animagine3 / AIDXL!). Specifically, we recommend the following order in Neta Art XL: Tag order: &nbsp;Subject (1boy / 1girl) -&gt; Character name (a girl named frieren from sousou no frieren series) -&gt; Artist (by xxx) -&gt; race (elf) -&gt; Camera combo (cowboy shot) -&gt; Style (impasto style) -&gt; Theme (fantasy theme) -&gt; Main environment description (in the forest, at day) -&gt; Background description (gradient background) -&gt; Action(sitting on ground) -&gt; Expression (is expressionless) -&gt; Main character features (white hair) -&gt; Body features (twintails, green eyes, parted lip) -&gt; Clothing features (wearing a white dress) -&gt; Accessories (frills) -&gt; Other appendages (a cat) -&gt; Secondary scene (grass, sunshine) -&gt; Aesthetics (beautiful color, detailed, aesthetic) -&gt; Quality words ((best quality:1.3)) Negative prompt words: (worst quality:1.3), low quality, lowres, messy, abstract, ugly, disfigured, bad anatomy, draft, deformed hands, fused fingers, signature, text, multi viewsSampler parameters: Eular a normal as default, 28+ steps recommended. <img src="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/2ac922c11656bf16c25888c17dcf634e4c355a89a204fbcdc061c248e0fa9db8.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" alt="" data-href="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/2ac922c11656bf16c25888c17dcf634e4c355a89a204fbcdc061c248e0fa9db8.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" style=""/>Prompt guidance coefficient CFG scale Another advantage of Neta Art XL is that it supports a very wide range of prompt guidance coefficients CFG scale (supports 5 - 20, while the general model is 7 - 9). Although we have found through experience that higher CFG brings more details and higher contrast, usually using CFG 9 - 14（important！）can achieve the best results. <h1>2. Highlights: Coexistence of Multiple Styles</h1>We have carefully selected 13 styles with good orthogonality and commonly used in many scenarios, and their rationality is proven by over 30 million generations by Neta AI users.Having orthogonal styles means each style is actually different from the others, allowing you to easily combine and createnew mixed styleswithout interference.<img src="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/757b75891e001f3897fc23e463ac718708d0d0843e2d9f0f9950874d3abeeac7.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" alt="" data-href="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/757b75891e001f3897fc23e463ac718708d0d0843e2d9f0f9950874d3abeeac7.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" style=""/><img src="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/3dd52db38411ebc82570aaee66dcf20f5e62856432aa790a1ef178a02db89af1.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" alt="" data-href="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/3dd52db38411ebc82570aaee66dcf20f5e62856432aa790a1ef178a02db89af1.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" style=""/><img src="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/0ca582b07b28eb0a55b37eaeaf24554ce19effd8426a29e1ea57563c906dc044.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" alt="" data-href="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/0ca582b07b28eb0a55b37eaeaf24554ce19effd8426a29e1ea57563c906dc044.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" style=""/><img src="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/d871d4549ffa0624bd3e4c7187e34f938c8458c86e6ca96a431fd8c9cc1008f4.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" alt="" data-href="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/d871d4549ffa0624bd3e4c7187e34f938c8458c86e6ca96a431fd8c9cc1008f4.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" style=""/> Neta Art XL also includes a large collection of artist styles, activated using the "by xxx" tag.<img src="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/0b4943ebd7ffaa9d1a344c8535329756bef3a5ee0b9e4014d7a74a62a87f8448.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" alt="" data-href="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/0b4943ebd7ffaa9d1a344c8535329756bef3a5ee0b9e4014d7a74a62a87f8448.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" style=""/><img src="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/955234d101a2b0649bf42e19ae069d5aebd6120895aa8771cad373d5659de9b2.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" alt="" data-href="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/955234d101a2b0649bf42e19ae069d5aebd6120895aa8771cad373d5659de9b2.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" style=""/>For reference, <a href="https://www.liblib.art/modelinfo/23c862f5c9d4492e903f898cd3fb3013" target="_blank">AIDXL</a> you can find more supported artists <h1>3. For other displays, please refer to the report</h1> <a href="https://nieta-art.feishu.cn/wiki/ARWLw99w7ikjaSkoV99cxmsOn5g" target="_blank">https://nieta-art.feishu.cn/wiki/ARWLw99w7ikjaSkoV99cxmsOn5g</a> <img src="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/3c4d251d21e4babfbadcdf778474255617bdeb2aed115b8d7ef520aba8a90165.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" alt="" data-href="https://liblibai-online.liblib.cloud/img/b6b0a317a417452dbea216eb5290a7ef/3c4d251d21e4babfbadcdf778474255617bdeb2aed115b8d7ef520aba8a90165.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" style=""/> <h1>4. Terms of Use</h1><ul><li style="text-align: left;">This model is developed by Neta.art Lab</li><li style="text-align: left;">Collaborators: </li><ul><li style="text-align: left;">Euge: https://civitai.com/user/Euge_</li><li style="text-align: left;">Tang Ren Lan: https://space.bilibili.com/8594480</li><li style="text-align: left;">Chenkin: https://civitai.com/user/Chenkin</li><li style="text-align: left;">Bo Dai: https://daibo.info/</li></ul><li>Acknowledgements and References:</li><ul><li>https://blog.novelai.net/introducing-novelai-diffusion-anime-v3-6d00d1c118c3</li><li>https://cagliostrolab.net/posts/animagine-xl-v3-release</li><li>https://civitai.com/models/269232/aam-xl-anime-mix</li><li>https://civitai.com/models/124189/anime-illust-diffusion-xl</li><li>https://github.com/deepghs/waifuc</li></ul><li style="text-align: left;">Model category: Diffusion-based text-to-image generation model</li><li style="text-align: left;">Usage agreement: This model incorporates 0.05 CLIP and 0.15 UNet input layers from AnimagineXL 3.1, thus following the Fair AI Public License 1.0-SD. If you modify, merge, or further develop this model, you must open source the derived model. </li></ul><h1>9. Summary and Outlook</h1>Drawbacks:<ol><li>Some characters are not well-fitted.</li><li>Style activation is not obvious when prompts are long.</li><li>Some styles appear grayish under low CFG and short prompts. Partial explanations can be found at https://civitai.com/articles/4969</li><li>Future work:</li></ol><ul><li>Prepare a larger training set and more knowledge-based data to improve character, style, and detail handling.</li><li>We welcome others to join the discussion and provide suggestions to contribute to the model's improvement. </li></ul>Neta Art XL 2.0 is coming soon. Please stay tuned and test our products for free: http://nieta.art<ul><li>Bilibili: https://space.bilibili.com/505727005</li><li>Xiaohongshu: https://www.xiaohongshu.com/user/profile/63f2ebf2000000001001e206</li><li>Twitter: https://twitter.com/netaart_ai</li><li>Civitai: https://civitai.com/user/nieta_art</li></ul>

neta_art_xl_v1.safetensors

V1.0

V1.0 DPO

V2.0

Files with SHA256: 5a268fde845f9bce52111cd05b8241cabe1580af4c96ee5f68fb61f422693cb1