AWPortrait

After five months, AWPortrait has finally released version 1.4. This update, based on 1.3 training, enhances the realism of character generation, bringing the realism of SD1.5 portraits to a new level. It includes additional elements such as qipao and hanfu, improved studio and outdoor effects, and optimized close-up features to reduce facial distortion. The model can generate characters of varying ages, from 10 to 80 years old. Using "1girl" tends to produce a consistent average face, while using natural language descriptions and adding more details can yield a wider variety of facial features. Describing a character as "a woman" can also produce more diverse results. For more examples, please refer to the sample prompts. To achieve the best image quality, please follow these parameter settings: - Recommended sampling: DPM++ 2M Karras or Restart, Step 30 - Resolution: 512x768, 576x864 - Upscaling algorithm: Lanczos, redraw amplitude 0.35-0.45 - Negative prompts: ng_deepnegative_v1_75t, (badhandv4:1.2), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, ((monochrome)), ((grayscale)), watermark, moles, large breast, big breast Adetailer: - Face model: face_yolov8n - For close-up portraits, you can disable or enable redraw amplitude at 0.25-0.3. - For upper body portraits, you can disable or enable redraw amplitude at 0.3-0.35. - For full body portraits, enable redraw amplitude at 0.35-0.4. Model Usage Agreement (Please read carefully) By downloading this model or using online image generation services based on this model, you agree to comply with this agreement. - Individual Users: You may use this model for image generation (including commercial use) and for research and study purposes. If you create derivative models by merging with AWPortrait, you must open source the derivative model, clearly state in the model description that it was merged with AWPortrait, and disclose the merging weights. The derivative models must not be sold or used commercially in any form (including but not limited to text-to-image, image-to-image, face swapping, integration into any third-party software or services, etc.). The author reserves the right to claim any commercial income obtained in violation of this agreement and any losses incurred. - Corporate Users: If you need to use this model (including but not limited to text-to-image, image-to-image, face swapping, integration into any third-party software or services, etc.), you must purchase a commercial license for this model. To purchase a license, please visit Xiaohongshu. The author reserves the right to claim any commercial income obtained in violation of this agreement and any losses incurred. - Disclaimer: You are responsible for any text content generated based on this model and must use it in compliance with applicable laws and this agreement. You must handle any intellectual property issues arising from the generated content, and the author is not responsible for any losses incurred. Except as required by law, without the author's written consent, you may not integrate this model service into any third-party software or services, nor disclose any direct or indirect, explicit or implicit association of the generated content with this service to any third party. Without the author's written consent, non-individual users (including but not limited to enterprises and profit-making organizations) may not use the model for commercial purposes (including but not limited to text-to-image, image-to-image, face swapping, etc.). The author reserves the right to claim any commercial income obtained in violation of this agreement and any losses incurred. AWPortrait1.4 经过5个月，AWPortrait终于迎来了1.4。本次更新基于1.3训练，使得人物的生成更加趋近真实感，将SD1.5人像的真实感提升到了一个新的高度，增加了旗袍、汉服等东方元素，更好的棚拍效果及户外效果；优化了close up下的特写人物五官容易造成畸形的情况；可更改人物年龄，比较稳定的输出效果是10-80岁；使用1girl比较容易生成一致性的平均脸，如需获得更多面相，建议使用自然语言及加入更多的人物描述，使用a woman的描述词也会获得更多的可能性；更多模型效果请参考例图Prompts。 为了达到最好的生图品质，请遵照以下参数配置来进行设置。 采样推荐：DPM++ 2M Karras或Restart ， Step 30 分辨率：512*768，576*864 放大算法：Lanczos， 重绘幅度0.35-0.45 负面提示词：ng_deepnegative_v1_75t,(badhandv4:1.2),(worst quality:2),(low quality:2),(normal quality:2),lowres,bad anatomy,bad hands,((monochrome)),((grayscale)) watermark,moles,large breast,big breast Adetailer： 脸部模型 face_yolov8n Close-up 特写情况下，可不开启或开启0.25-0.3的重绘幅度。 UpperBody 半身像情况下，可不开启或0.3-0.35的重绘幅度。 Fullbody 全身像情况下，请开启0.35-0.4的重绘幅度。 模型使用协议（请认真阅读） 您下载本模型或使用基于加载本模型的在线生图服务，即表示同意遵守本协议。 一. 个人用户，可使用本模型进行生图（包含商用），以及基于研究和学习为目的融合模型，融合AWPortrait的衍生模型如再发布，必须开源且在模型说明中明确注明融合AWPortrait并公布融合权重，融合后的衍生模型不得以任何形式售卖及进行商业使用（包括但不限于文生图、图生图、换脸，集成在任何第三方软件或服务等）。您违反本协议所获得的商业收入所得及给作者本人所造成的损失等，作者有权向您追偿。 二. 企业用户（包括但不限于企业、盈利性组织），如需使用本模型（包括但不限于文生图、图生图、换脸，集成在任何第三方软件或服务等），均需购买本模型的商业授权，购买授权请移步小红书购买。您违反本协议所获得的商业收入所得及给作者本人所造成的损失等，作者有权向您追偿。 三. 免责声明： 您基于本模型服务生成的文本内容由您自行维护并对其独立判断后使用，您需在遵守适用法律法规和本服务条款的条件下使用，基于本服务生成内容产生的任何知识产权问题由您自行处理，您需对生成内容负责，本人对由此造成的任何损失不负责任。除法律法规另有规定外，未经本人书面授权同意，不得将本模型服务集成在任何第三方软件或服务中，亦不得以任何直接间接、明示或暗示方式向其他任何第三方披露生成内容与本服务的关联性；未经本人书面授权同意，非个人用户（包括但不限于企业、盈利性组织）不得将模型用于商业用途（包括但不限于文生图、图生图、换脸等），您违反本协议所获得的商业收入所得及给作者本人所造成的损失等，作者有权向您追偿。 ———————————————————————————————————————— AWPortrait1.3 1.3继续优化棚拍的质感及皮肤肌理，且增加了户外场景的优化，使得户外场景的直出兼具了1.2的写真感，以及更加有“网感”的拍摄构图及模特造型，1.3整体照片的色彩呈现一种兼具宽容度的灰度和柔和色彩感受的直出效果。 增加了更多的对于表情理解，可以做出微笑、大笑、痛苦、哭泣等面部表情，最好结合Adetailer，将权重设置到0.35-0.4为宜。 本模型非常适合制作摄影风格化的LoRA，可以达到非常真实的照片效果。 更多Prompts参考，可看直出例图。 采样推荐Restart, DPM++ 2M Karras Version 1.3 continues to optimize the texture and skin details in studio photography, and has added enhancements for outdoor scenes. As a result, direct output from outdoor settings now combines the photogenic quality of version 1.2 with a more 'internet-trendy' composition and model styling. Overall, the photos in version 1.3 display colors that strike a balance between a tolerant grayscale and a gentle color feel. The update includes improved understanding of facial expressions, allowing for more accurate depictions of smiles, laughter, pain, and crying. It