FLUX-Controlnet-Inpainting的推理

目录

一、前言

二、FLUX-Controlnet-Inpainting的推理

二、区别和联系

一、前言

        局部重绘有三种方式：

1.基于FLUX.1-dev：传统的局部重绘，在每一时间步的去噪操作之后将掩码的上下文信息贴回去，掩码的作用并不是在模型中以直接的方式控制目标区域的去噪范围。

2.FLUX-Controlnet-Inpainting：借助外挂的Controlnet来实现局部重绘

3.FLUX.1-Fill-dev：重新训练了一个端到端的局部重绘模型。

二、FLUX-Controlnet-Inpainting的推理

Github地址：

https://github.com/alimama-creative/FLUX-Controlnet-Inpainting/blob/main/controlnet_flux.py

模型地址：

https://huggingface.co/alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Alpha

https://huggingface.co/alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Beta

在conda环境中安装依赖：

pip install diffusers==0.30.2
pip install accelerate==1.3.0
pip install transformers==4.44.0
pip install peft==0.14.0 safetensors==0.4.3
pip install sentencepiece
pip install protobuf

推理代码：python main.py

https://github.com/alimama-creative/FLUX-Controlnet-Inpainting/blob/main/main.py

import torch
from diffusers.utils import load_image, check_min_version
from controlnet_flux import FluxControlNetModel
from transformer_flux import FluxTransformer2DModel
from pipeline_flux_controlnet_inpaint import FluxControlNetInpaintingPipeline

check_min_version("0.30.2")

# Set image path , mask path and prompt
image_path='2026-05-07ebmvnh2lrtw.png',
mask_path='2026-05-07rdajr4jxu01.jpeg',
prompt='a person wearing a white shoe, carrying a white bucket with text "FLUX" on it'

# Build pipeline
controlnet = FluxControlNetModel.from_pretrained("alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Alpha", torch_dtype=torch.bfloat16)
transformer = FluxTransformer2DModel.from_pretrained(
"black-forest-labs/FLUX.1-dev", subfolder='transformer', torch_dtype=torch.bfloat16
)
pipe = FluxControlNetInpaintingPipeline.from_pretrained(
"black-forest-labs/FLUX.1-dev",
controlnet=controlnet,
transformer=transformer,
torch_dtype=torch.bfloat16
).to("cuda")
pipe.transformer.to(torch.bfloat16)
pipe.controlnet.to(torch.bfloat16)

# Load image and mask
size = (768, 768)
image = load_image(image_path).convert("RGB").resize(size)
mask = load_image(mask_path).convert("RGB").resize(size)
generator = torch.Generator(device="cuda").manual_seed(24)

# Inpaint
result = pipe(
prompt=prompt,
height=size[1],
width=size[0],
control_image=image,
control_mask=mask,
num_inference_steps=28,
generator=generator,
controlnet_conditioning_scale=0.9,
guidance_scale=3.5,
negative_prompt="",
true_guidance_scale=1.0 # default: 3.5 for alpha and 1.0 for beta
).images[0]

result.save('flux_inpaint.png')
print("Successfully inpaint image")

二、区别和联系

比如说我用diffusers官方的flux.1-dev（简称为模型A）的lora训练代码训练出来的lora（简称为模型A-lora），这时候我可以叠加controlnet（简称为模型B）对吧，那diffusers官方应该也有一个训练controlnet的代码，训练出来的模型（简称为模型B-trained），那现在我有一个FLUX.1-dev-Controlnet-Inpainting-Beta（简称为模型C），有一个FLUX.1-Fill-dev（简称为模型 D）。

模型A、B、C、D的区别和联系是什么？