然后,base模型的那一套输出给第一个采样器节点,refiner模型的那一套输出给第二个采样器节点。 最后,我们能可以输出两个图像节点,第一个链接base模型的vae,可以设置为预览图像;第二个链接一个VAE加载器的节点,加载sdxl自带的vae,设置为保存图像,也就是我们最终输出的图像。 使用这个工作流我们来跑一张sdxl模型的...
Negative prompt: (deformed iris, deformed pupils), text, worst quality, low quality, ugly Steps: 40, Seed: 17748028598465, Model: DreamShaperXL1.0Alpha2_fixedVae_half_00001_, width: 768, height: 1024, Sampler: DPM++ 2S a Karras, CFG scale: 8 3、画物体 ultra detailed 8k cg,japanese ram...
因为这个最主要是让你用 i2i 的方式来做 upscale 的,所以你可以选择在这里接上一个 load image 方块,然后直接上传一张你要 upscale 的图片。 如果你想用 t2i 的方式来做,往上接一小段基本 workflow 的尾巴就可以,叫出 Ksampler 和 VAE decoder,前面的模型啊提示词啊都可以共用,最后把 image 接到 upscale ...
Steps: 40, Seed: 17748028598465, Model: DreamShaperXL1.0Alpha2_fixedVae_half_00001_, width: 768, height: 1024, Sampler: DPM++ 2S a Karras, CFG scale: 8 3、画物体 ultra detailed 8k cg,japanese ramen,chopsticks,egg,steam, Negative prompt: embedding:BadDream,embedding:UnrealisticDream, Steps:...
Fixed FP16 VAE It is currently recommended to use aFixed FP16 VAErather than the ones built into the SD-XL base and refiner for significant reductions in VRAM (from 6GB of VRAM to <1GB VRAM) and a doubling of VAE processing speed. ...
(instead of using the VAE that's embedded in SDXL 1.0, this one has been fixed to work in fp16 and should fix the issue with generating black images) (optional) download SDXL Offset Noise LoRA (50 MB) and copy it into ComfyUI/models/loras (the example lora that was released alongsid...
"Node name for S&R": "VAEDecode" }, "widgets_values": [], "color": "#323", "bgcolor": "#535" }, { "id": 3, "type": "KSampler", "pos": [ 1190, 30 ], "size": [ 300, 262 ], "flags": {}, "order": 11, "mode": 0, "inputs": ...
生成预置包将当前出图环境参数打包保存为新的预置包,将预置包文件存入presets目录下,自动加入顶部导航。 扩展预置参数扩展主线的预置包参数范围,补充开发者模式的参数,以及风格样式的定义和通配符的定义。支持的预置包参数见预置包ReadMe 统一模型ID和下载对接模型信息库,使用以模型文件哈希为基础的统一模型MUID。可自动检...
Bake in VAE When saving the model, the selected VAE is incorporated into the model. Save current Merge Saves the currently loaded model. This is effective when merging takes time due to PC specifications or other issues. Merge Block Weight This is a merging technique where you can set the ...
最后,我们能可以输出两个图像节点,第一个链接base模型的vae,可以设置为预览图像;第二个链接一个VAE加载器的节点,加载sdxl自带的vae,设置为保存图像,也就是我们最终输出的图像。 使用这个工作流我们来跑一张sdxl模型的图片,设置好两个模型和提示词,点击生成。