lllyasviel / lllyasviel/ControlNet

关于ControlNet变量控制的问题

Open
#75 7 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
34.1k
Forks
3k
PR merge metrics
No merged PRs in 30d

Description

通过ControlNet的原模型和Auto1111的插件均仔细体验了ControlNet的效果,目前主要使用的方式有Canny和Scribble和 Fake Scribble 三种方式,使用了作者提供的原模型,以及AnythingV3所混合的模型(有模型混合版,也有通过Auto1111实现的非混合直接调用AnythingV3的方式)

**所放置的测试图片内容如下:**

![xiaoxiong](https://user-images.githubusercontent.com/14341380/219328876-596200e9-2a9b-41b0-a5cb-656bdc8a9061.png)

![控笔画画练习102 2](https://user-images.githubusercontent.com/14341380/219329011-1bfbe8ba-5f3e-4f33-9378-4dc0b8002512.jpg)

![image](https://user-images.githubusercontent.com/14341380/219329795-f8405939-5092-4f7e-8fa1-089e932bcac3.png)

**得到的结果中,效果最佳的如下:**

![image](https://user-images.githubusercontent.com/14341380/219329967-22a40e92-c7ec-4f07-9c6c-77d6cc4b28ba.png)

![image](https://user-images.githubusercontent.com/14341380/219330122-57ed09cd-adff-4d7c-a471-18e4c5ad081d.png)

![image](https://user-images.githubusercontent.com/14341380/219330216-1627d3dc-700d-4eef-9167-436c7d5ff218.png)

![image](https://user-images.githubusercontent.com/14341380/219330295-a2f15cd9-71f6-437d-ae60-43a924098c10.png)

因为经过了大量的参数变量的调整和产出结果的对比测试,**目前总结的可控性问题**,主要集中在适配的模型(应该找比AnythingV3风格特征强度更弱化一点的模型),**以及ControNet本身的可控制变量参数上:**

**1、除了能提供底稿图外,请问是否还可以通过某区域标注的方式,标明物体的内容——比如猫,比如脸,比如摆件等?

2、除了提供底稿图外,是否可以通过在封闭区间,或者元素区域标注颜色提示,来实现对应的配色效果?**

**这样的可控性通过不同维度和颗粒度的处理,会让模型更加具备可控性,希望作者大佬采纳,感谢~!**

Contributor guide

No contributing guide indexed for this repository

Research direction

The issue reports experiments with ControlNet using Canny, Scribble, and Fake Scribble, then requests region labels for object content and color hints for enclosed areas. No source files, tests, entry points, or acceptance criteria are identified; first clarify the intended control inputs and expected outputs before locating the relevant model or conditioning code.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
ai, machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.