| ### AUTOMATIC1111 WebUI Stable Diffusion | |
| `WebUIStableDiffusionPipeline` 是与 [AUTOMATIC1111/stable-diffusion-webui](https://github.com/AUTOMATIC1111/stable-diffusion-webui) 基本对齐的一个pipeline。 | |
| 该自定义Pipeline支持如下的功能： | |
| * 输入的 token 没有长度限制，可以超过77； | |
| * 支持clip_skip，即可以使用不同层text_encoder的输出； | |
| * 支持直接加载webui中的textual_inversion权重； | |
| ```python | |
| import paddle | |
| from ppdiffusers.utils import image_grid | |
| from ppdiffusers import DiffusionPipeline | |
| from pathlib import Path | |
| pipe = DiffusionPipeline.from_pretrained("TASUKU2023/Chilloutmix", paddle_dtype=paddle.float16, custom_pipeline="junnyu/webui_ppdiffusers") | |
| # 自动下载civitai的lora及ti文件（请注意自己的网络。） | |
| # 介绍网页，程序将自动搜索介绍网页的下载链接 | |
| pipe.download_civitai_lora_file('https://civitai.com/models/15365/hanfu') | |
| pipe.download_civitai_lora_file('https://civitai.com/models/12597/moxin') | |
| pipe.download_civitai_ti_file('https://civitai.com/models/1998/autumn-style') | |
| pipe.download_civitai_ti_file('https://civitai.com/models/21131/daisy-ridley-embedding') | |
| # 纯下载链接 | |
| pipe.download_civitai_lora_file('https://civitai.com/api/download/models/21656') | |
| print("Supported Lora: " + "、 ".join([p.stem for p in Path(pipe.LORA_DIR).glob("*.safetensors")])) | |
| # 我们需要安装develop版的paddle才可以使用xformers | |
| # pipe.enable_xformers_memory_efficient_attention() | |
| scheduler_name = ["ddim", "pndm", "euler", "dpm-multi"] | |
| for enable_lora in [True, False]: | |
| images = [] | |
| for sc in scheduler_name: | |
| # 切换scheduler | |
| pipe.switch_scheduler(sc) | |
| # 指定clip_skip | |
| clip_skip = 1 | |
| # 指定seed | |
| generator = paddle.Generator().manual_seed(0) | |
| # guidance_scale | |
| guidance_scale = 3.5 | |
| prompt = "# shukezouma, negative space, , shuimobysim , portrait of a woman standing , willow branches, (masterpiece, best quality:1.2), traditional chinese ink painting, <lora:Moxin_10:1.0>, modelshoot style, peaceful, (smile), looking at viewer, wearing long hanfu, hanfu, song, willow tree in background, wuchangshuo," | |
| negative_prompt = "(worst quality:2), (low quality:2), (normal quality:2), lowres, normal quality, skin spots, acnes, skin blemishes, age spot, glans, (watermark:2)," | |
| img = pipe(prompt, negative_prompt=negative_prompt, num_inference_steps=50, height=768, width=512, clip_skip=clip_skip, guidance_scale=guidance_scale, generator=generator, enable_lora=enable_lora).images[0] | |
| images.append(img) | |
| if enable_lora: | |
| image_grid(images, 2, 2).save(f"lora_enable.png") | |
| else: | |
| image_grid(images, 2, 2).save(f"lora_disable.png") | |
| ``` | |
| 生成的图片如下所示： | |
| | lora_disable.png | lora_enable.png | | |
| |:----------:|:--------------:| | |
| |<center class="half"><img src="https://user-images.githubusercontent.com/50394665/230832029-c06a1367-1f2c-4206-9666-99854fcee240.png" width=50%></center> | <center class="half"><img src="https://user-images.githubusercontent.com/50394665/230832028-730ce442-dd34-4e36-afd0-81d40843359a.png" width=50%></center> | | |