Extended Outpaint（XL-Update）
5.0
2 reviews
Description
扩图工作流的一个升级版，基于XL的模型。
针对以下问题作出了改进：
色彩不统一，与原始图有色彩偏差，有明显边界，有多余的物体和人物，构图不够合理。。。。
其中我利用了差异扩散控制主体和背景的融合，来解决了色彩偏差和接缝明显等问题，没有主体的风景图一样适用，我做了自动识别。
这个升级版本更好的适用了真实拍摄的图片
或许你不需要管那么多 点击生成一键全自动化
--------------------------------------------------------------------------------------------------------------------
使用的模型：
CHECKPOINT
RealVisX V4.0
https://civitai.com/models/139562?modelVersionId=344487
Fooocus
https://huggingface.co/lllyasviel/fooocus_inpaint/tree/main
下载所有模型放置在ComfyUI/models/inpaint/路径下
IPAdapter
https://github.com/cubiq/ComfyUI_IPAdapter_plus
按照以上链接中的说明，下载和安装所有模型。
https://huggingface.co/ViperYX/BiRefNet/tree/main
在以上链接下载以下5个模型，把模型放置在ComfyUI/models/BiRefNet/路径下
BiRefNet-ep480.pth
pvt_v2_b2.pth
pvt_v2_b5.pth
swin_base_patch4_window12_384_22kto1k.pth
swin_large_patch4_window12_384_22kto1k.pth
Lama
https://drive.google.com/drive/folders/1Aq0a4sybb3SRxi7j1e1_ZbBRjaWDdP9e
在以上链接下载 big-lama.pt模型 ，把模型放置在ComfyUI/models/lama/路径下
--------------------------------------------------------------------------------------------------------------------
https://space.bilibili.com/30671205
WeChat（微信）:knowknow0
Node Diagram
Discussion
Error occurred when executing KSampler: too many values to unpack (expected 3)
?
thx
sorry ~ Maybe it's because I supplied the wrong CHECKPOINT, but now it's been changed.RealVisX V4.0
thx
The above list of mask calculation nodes looks awesome, but the effect is limited. Most of the time, the main body will deform because it is too close to the edge. More deadly is that if the original image is of high resolution, the final output will blur the entire image.
There are two Brightness & Contrast nodes, which can be adjusted to solve the issue of subject edge distortion.
The blurring problem can be addressed by adding a magnification correction function in the final output blend node.
Additionally, the edge distortion issue can be resolved by adjusting the extended parameters of the LAMA node.
(Edited)Error occurred when executing LayerMask: BiRefNetUltra:
We couldn't connect to 'https://huggingface.co' to load this file, couldn't find it in the cached files and it looks like hustvl/vitmatte-small-composition-1k is not the path to a directory containing a file named config.json.
Checkout your internet connection or see how to run the library in offline mode at 'https://huggingface.co/docs/transformers/installation#offline-mode'.
download
https://huggingface.co/hustvl/vitmatte-base-composition-1k/tree/main
and put these files in \models\vitmatte
I am keep getting error
Error occurred when executing LayerMask: BiRefNetUltra:
We couldn't connect to 'https://huggingface.co' to load this file, couldn't find it in the cached files and it looks like hustvl/vitmatte-small-composition-1k is not the path to a directory containing a file named config.json.
Checkout your internet connection or see how to run the library in offline mode at 'https://huggingface.co/docs/transformers/installation#offline-mode'.
download
https://huggingface.co/hustvl/vitmatte-base-composition-1k/tree/main
and put these files in \models\vitmatte
效果不错，一次跑通无报错
Thanks you i have successfully Render.
and i prefer this CHKPT model https://civitai.com/models/277058?modelVersionId=489217
for better composition.
How long did it take you to finish the rendering?
The way this workflow handles inpainting is very clever ! To first outpaint the pixels using another LaMa model to ensure color consistency then with masked attention for the IP adapter, it works really well even with higher resolutions. I still don't understand the need for that many nodes later on though, in any case, thanks for sharing probably the best inpainting workflow using SDXL !
This is insanely good
Just make sure you have downloaded everything, yes every single file so you will get a result just like the above
also, you can try to add steps to 40 or more will also help to generate a better image
Node Details
Primitive Nodes (12)
DF_Divide (1)
DF_Sum (2)
DF_To_text_(Debug) (5)
Reroute (4)
Custom Nodes (76)
ComfyUI
- MaskToImage (8)
- ImageScaleBy (3)
- ImageToMask (6)
- ImageScale (4)
- GrowMask (3)
- InvertMask (3)
- PreviewImage (3)
- CheckpointLoaderSimple (1)
- CLIPTextEncode (2)
- VAEDecode (1)
- LoadImage (1)
- ImagePadForOutpaint (1)
- SplitImageWithAlpha (1)
- KSampler (1)
- DifferentialDiffusion (1)
- RepeatLatentBatch (1)
- SaveImage (1)
- easy compare (1)
- easy int (2)
- easy if (1)
- MaskBlur+ (1)
- ImpactSwitch (1)
- INPAINT_VAEEncodeInpaintConditioning (1)
- INPAINT_LoadFooocusInpaint (1)
- INPAINT_ApplyFooocusInpaint (1)
- GlobalSeed //Inspire (1)
- LayerUtility: ImageScaleRestore V2 (2)
- LayerUtility: GetColorTone (1)
- LayerUtility: TextBox (1)
- LayerUtility: GetImageSize (4)
- LayerMask: MaskBoxDetect (1)
- LayerUtility: ImageBlend (5)
- LayerColor: Brightness & Contrast (2)
- LayerUtility: LaMa (1)
- LayerMask: BiRefNetUltra (1)
- LayerMask: MaskPreview (1)
- IPAdapterUnifiedLoader (1)
- IPAdapterAdvanced (1)
- IPAdapterNoise (1)
- Paste By Mask (1)
- Image Blank (2)
Model Details
Checkpoints (1)
realvisxlV40_v40Bakedvae.safetensors
LoRAs (0)