如果您需要更好的效果，请去寻找基于Z-IMAGE的I2I工作流，但是目前Z-IMAGE在生成NSFW内容方面尚不如IL系列模型好，本模型可以还原角色的着装，姿势，在还原人物外貌方面需要用到另一个ipadapter节点，这个内容会在未来的v1.0 preview版本中推出，如果你有任何疑惑或者使用问题，请联系我。If you require better results, please look for the I2I workflow based on Z-IMAGE. However, Z-IMAGE is currently not as good as the IL series models in generating NSFW content. This model can reproduce the character's clothing and pose. To reproduce the character's appearance, another iPadapter node is needed. This feature will be released in the future v1.0 preview version. If you have any questions or usage issues, please contact me.<h3>请注意使用规范，在符合社区规定和法律的情况下使用。</h3><h3 id="please-be-aware-of-usage-guidelines-and-use-the-product-in-accordance-with-community-rules-and-laws.">Please be aware of usage guidelines and use the product in accordance with community rules and laws.</h3><h3></h3><h2>请仔细阅读注意事项，谢谢！</h2><h2 id="please-read-the-precautions-carefully-thank-you!">Please read the precautions carefully, thank you!</h2><h2></h2>这是结合基于 IPAdapter 的参考引导、人体分割和基于 FaceDetailer 的局部增强，以实现参考一致性、视觉质量和结构稳定性之间的平衡的工作流。A workflow that balances reference consistency, visual quality, and structural stability is achieved by combining IPAdapter-based reference guidance, human segmentation, and FaceDetailer-based local enhancement.注意：该工作流仍然在开发中，它有一些不稳定因素，如果你有任何建议或者想法，可以给我发私信，模型适用SDXL模型，其他模型仍在开发中。Note: This workflow is still under development and has some uncertainties. If you have any suggestions or ideas, please send me a private message. The model is applicable to SDXL; other models are still under development.该模型还没有达到我最希望的工作效果，我仍然在改进！The model has not yet achieved the desired working effect, and I am still working on improving it!FaceDetailer 及相关的检测和优化节点来自 ComfyUI-Impact-Pack，可从 <a target="_blank" rel="ugc" href="https://github.com/ltdrdata/ComfyUI-Impact-Pack">https://github.com/ltdrdata/ComfyUI-Impact-Pack</a> 下载。IPAdapter 和高级 weight_type 控制由 ComfyUI_IPAdapter_plus 提供，可从 <a target="_blank" rel="ugc" href="https://github.com/cubiq/ComfyUI_IPAdapter_plus">https://github.com/cubiq/ComfyUI_IPAdapter_plus</a> 获取。人体分割由 ComfyUI-Easy-Use 处理，可从 <a target="_blank" rel="ugc" href="https://github.com/yolain/ComfyUI-Easy-Use">https://github.com/yolain/ComfyUI-Easy-Use</a> 下载。人脸检测依赖于 Ultralytics YOLO 模型，这些模型是 Ultralytics 项目的一部分，位于 <a target="_blank" rel="ugc" href="https://github.com/ultralytics/ultralytics。">https://github.com/ultralytics/ultralytics。</a>这是我使用的一些模型：SDXL 所需的 IPAdapter 模型是 ip-adapter-plus_sdxl_vit-h.bin，可从官方 IP-Adapter 仓库下载，网址为 <a target="_blank" rel="ugc" href="https://huggingface.co/h94/IP-Adapter">https://huggingface.co/h94/IP-Adapter</a>FaceDetailer 使用 YOLO 人脸检测模型 face_<a target="_blank" rel="ugc" href="http://yolov8m.pt">yolov8m.pt</a>，可从 Ultralytics 模型发布页面获取，网址为 <a target="_blank" rel="ugc" href="https://github.com/ultralytics/assets/releases">https://github.com/ultralytics/assets/releases</a>为了保证解码和编码的稳定性，此工作流程使用 fixFP16ErrorsSDXLLowerMemoryUse_v10.safetensors 作为 VAE，该 VAE 可在 Hugging Face 模型中心（例如 <a target="_blank" rel="ugc" href="https://huggingface.co）找到。">https://huggingface.co）找到。</a>FaceDetailer and related detection and optimization nodes are from ComfyUI-Impact-Pack, available for download at <a target="_blank" rel="ugc" href="https://github.com/ltdrdata/ComfyUI-Impact-Pack">https://github.com/ltdrdata/ComfyUI-Impact-Pack</a>.IPAdapter and advanced weight_type control are provided by ComfyUI_IPAdapter_plus, available at <a target="_blank" rel="ugc" href="https://github.com/cubiq/ComfyUI_IPAdapter_plus">https://github.com/cubiq/ComfyUI_IPAdapter_plus</a>.Human segmentation is handled by ComfyUI-Easy-Use, available for download at <a target="_blank" rel="ugc" href="https://github.com/yolain/ComfyUI-Easy-Use">https://github.com/yolain/ComfyUI-Easy-Use</a>.Face detection relies on Ultralytics YOLO models, which are part of the Ultralytics project, located at <a target="_blank" rel="ugc" href="https://github.com/ultralytics/ultralytics">https://github.com/ultralytics/ultralytics</a>.The IPAdapter model required for SDXL is ip-adapter-plus_sdxl_vit-h.bin, which can be downloaded from the official IP-Adapter repository at <a target="_blank" rel="ugc" href="https://huggingface.co/h94/IP-Adapter">https://huggingface.co/h94/IP-Adapter</a>FaceDetailer uses the YOLO face detection model face_<a target="_blank" rel="ugc" href="http://yolov8m.pt">yolov8m.pt</a>, which can be obtained from the Ultralytics model release page at <a target="_blank" rel="ugc" href="https://github.com/ultralytics/assets/releases">https://github.com/ultralytics/assets/releases</a>To ensure the stability of decoding and encoding, this workflow uses fixFP16ErrorsSDXLLowerMemoryUse_v10.safetensors as the VAE, which can be found in the Hugging Face Model Hub (e.g., <a target="_blank" rel="ugc" href="https://huggingface.co">https://huggingface.co</a>).

workflow：I2I base on upload image(without controlnet)

此版本着重提升整体稳定性和参考控制，同时工作流程仍在积极优化中。IPAdapter 的行为已得到改进，以减少过度参考并提高不同<code>weight_type</code>模式下的视觉一致性。FaceDetailer 设置已进行调整，以最大限度地减少画中画伪影和意外的面部重建。 人体分割和注意力掩蔽已得到更好的协调，以减少背景污染。 参数默认值已针对更安全的通用用途进行了调整，但计划在未来的版本中进行进一步的改进和结构调整。 IPAdapter behavior has been refined to reduce over-referencing and improve visual consistency across different <code>weight_type</code> modes. FaceDetailer settings have been adjusted to minimize picture-in-picture artifacts and unintended facial reconstruction. Human segmentation and attention masking have been better aligned to reduce background contamination. Parameter defaults are tuned for safer general use, though further refinements and structural adjustments are planned in future versions.

Files with SHA256: 63df3f56d78987fca57ed0547529b3a15eb95ca3a062a1082c3e99083d0d207f