Modelpublic

lordjia/stereoscopic-portrait-3d-li-ti-xiao-xiang

sourceHugging Faceotherupdated 3y agoView on Hugging Face

4likes31downloads

Model Card

Stereoscopic Portrait - 3D 立体肖像

Model description

This is an experimental project designed to produce Stereoscopic Portraits. It supports both Parallel-view and Cross-view modes, corresponding to two downloadable versions.Performance:When used as recommended, the subject of the portrait achieves a relatively high success rate. The background, in contrast, might require multiple attempts and might achieve a lower success rate.User Guide: For optimal results and image quality, strictly follow the procedure outlined below:<ul><li>Checkpoint: Realistic Vision V5.1</li><li>Textual Inversion: ngdeepnegativev175t</li><li>Positive prompts: {prompt}, masterpiece, high detail, 8k, high detailed skin, 8k uhd, high quality</li><li>Negative prompts: ugly, disfigured, deformed, worst quality, low quality, ngdeepnegativev175t</li><li>LoRA Weight: 0.8-1.0</li><li>Image Dimensions: Width: 832, Height: 560</li><li>Optional: Use Hires. fix to enhance image quality with a Denoising strength of 0.3.</li></ul>Note:<ul><li>Using the Realistic Vision V5.1 in tandem with ngdeepnegativev175t can significantly improve the success rate and generated image quality. See the "Suggested Resources" section below for more details.</li><li>This LoRA is specifically designed for generating Stereoscopic Portraits with a human subject. Without a human subject, a 3D effect cannot be achieved.</li><li>For detailed information on Stereoscopy, refer to the Wikipedia entry titled "<a target="blank" rel="ugc" href="https://en.wikipedia.org/wiki/Stereoscopy">Stereoscopy</a>".</li></ul><hr />这是一个实验性项目，用来生成 立体视觉肖像（Stereoscopic Portrait），支持 Parallel-view 与 Cross-view 两种模式（对应两个下载版本）。在按照推荐方法使用的前提下，人物主体可以获得比较高的成功率，背景的成功率相较人物主体要低一些，有时需要多次尝试。使用说明：（严格按照以下流程生成图片，可以获得比较高的成功率与图像质量。）<ul><li>模型（checkpoint）：Realistic Vision V5.1</li><li>嵌入式（Textual Inversion）：ngdeepnegativev175t</li><li>正面提示词：{prompt}, masterpiece, high detail, 8k, high detailed skin, 8k uhd, high quality</li><li>负面提示词：ugly, disfigured, deformed, worst quality, low quality, ngdeepnegativev175t</li><li>LoRA 权重：0.8-1.0</li><li>图像尺寸：宽度：832，高度：560</li><li>可选：使用高清修复（Hires. fix）提升图像质量，重绘幅度（Denoising strength）0.3</li></ul>注意：<ul><li>配合使用 Realistic Vision V5.1 与 ngdeepnegativev175t 可以极大提高成功率与生成图像质量。具体见下方“Suggested Resources”。</li><li>此 LoRA 只用于生成带有人物主体的立体视觉肖像图，如果不带有人物主体则无法产生立体效果。</li><li>关于立体视觉（Stereoscopy）的具体说明，参见维基百科“<a target="blank" rel="ugc" href="https://zh.wikipedia.org/zh-cn/%E7%AB%8B%E4%BD%93%E5%9B%BE">立体图</a>”词条。</li></ul>

Download model

Weights for this model are available in Safetensors format.

Download them in the Files & versions tab.

Use it with the 🧨 diffusers library

from diffusers import AutoPipelineForText2Image
import torch

pipeline = AutoPipelineForText2Image.from_pretrained('runwayml/stable-diffusion-v1-5', torch_dtype=torch.float16).to('cuda')
pipeline.load_lora_weights('lordjia/stereoscopic-portrait-3d-li-ti-xiao-xiang', weight_name='stereoscopic_parallel_v1.0.safetensors')
image = pipeline('a princess in front of a castle, half body
masterpiece, high detail, 8k, high detailed skin, 8k uhd, high quality
').images[0]

For more details, including weighting, merging and fusing LoRAs, check the documentation on loading LoRAs in diffusers