63 lines
2.6 KiB
Markdown
63 lines
2.6 KiB
Markdown
# Documentation
|
|
- Class name: InpaintWithModel
|
|
- Category: inpaint
|
|
- Output node: False
|
|
- Repo Ref: https://github.com/Acly/comfyui-inpaint-nodes
|
|
|
|
该节点使用深度学习模型执行图像修复，有效地填充图像中缺失或被遮罩的区域，并生成与周围内容相适应的内容。它适应不同的模型架构，并确保输出与原始图像无缝集成。
|
|
|
|
# Input types
|
|
## Required
|
|
- inpaint_model
|
|
- 修复模型对于节点的功能至关重要，它定义了用于生成修复内容的具体深度学习架构。模型的架构直接影响修复结果的质量和准确性。
|
|
- Comfy dtype: INPAINT_MODEL
|
|
- Python dtype: PyTorchModel
|
|
- image
|
|
- 输入图像是节点操作的主要数据，修复过程旨在在填充缺失部分的同时保持其整体结构和美感。图像的质量和尺寸直接影响修复结果。
|
|
- Comfy dtype: IMAGE
|
|
- Python dtype: torch.Tensor
|
|
- mask
|
|
- 遮罩定义了图像中需要修复的区域。它是一个关键参数，因为它指导模型专注于特定区域，确保修复是有针对性和相关的。
|
|
- Comfy dtype: MASK
|
|
- Python dtype: torch.Tensor
|
|
|
|
# Output types
|
|
- output_image
|
|
- 输出图像是修复过程的结果，最初被遮罩或缺失的区域现在填充了与周围内容相匹配的内容。它代表节点的主要输出，对于进一步的图像分析或显示至关重要。
|
|
- Comfy dtype: IMAGE
|
|
- Python dtype: torch.Tensor
|
|
|
|
# Usage tips
|
|
- Infra type: GPU
|
|
|
|
# Source code
|
|
```
|
|
class InpaintWithModel:
|
|
|
|
@classmethod
|
|
def INPUT_TYPES(s):
|
|
return {'required': {'inpaint_model': ('INPAINT_MODEL',), 'image': ('IMAGE',), 'mask': ('MASK',)}}
|
|
RETURN_TYPES = ('IMAGE',)
|
|
CATEGORY = 'inpaint'
|
|
FUNCTION = 'inpaint'
|
|
|
|
def inpaint(self, inpaint_model: PyTorchModel, image: Tensor, mask: Tensor):
|
|
if inpaint_model.model_arch == 'MAT':
|
|
required_size = 512
|
|
elif inpaint_model.model_arch == 'LaMa':
|
|
required_size = 256
|
|
else:
|
|
raise ValueError(f'Unknown model_arch {inpaint_model.model_arch}')
|
|
(image, mask) = to_torch(image, mask)
|
|
image_device = image.device
|
|
(original_image, original_mask) = (image, mask)
|
|
(image, mask, original_size) = resize_square(image, mask, required_size)
|
|
mask = mask.floor()
|
|
device = get_torch_device()
|
|
inpaint_model.to(device)
|
|
image = inpaint_model(image.to(device), mask.to(device))
|
|
inpaint_model.cpu()
|
|
image = undo_resize_square(image.to(image_device), original_size)
|
|
image = original_image + (image - original_image) * original_mask.floor()
|
|
return (to_comfy(image),)
|
|
``` |