Chapter 3 doesn't need to be read in sequence!
Think of it as a reference document that you can consult as needed.
Learning Goals
- Recognize external services such as CivitAI, Openart, Hugging Face, and GitHub, and get a basic understanding of how to use them.
- Provide a brief guide on how to switch to a different node if the custom node imported from an external source is not supported.
Plain Text
복사
CivitAI
It is a sharing site for checkpoints frequently used by StableDiffusion users.
There are various models available, so feel free to explore if you need new checkpoints.
Model Information
Let me explain using the DreamShaperXL model that we frequently use.
First, let's access the following CivitAI link:
•
1: Information about the version is listed. A single model can have multiple versions.
•
2: Details about the base model and when it was trained are provided.
•
3: There is a 'description' left by the author. Reading this description carefully can be very helpful. (For example, if you want to use DreamShaper effectively, it is more efficient to read the text written by the creator of DreamShaper rather than not reading it.)
Generation Data
CivitAI has many images attached with generation data.
For example, if information is provided as follows, you can set up the workflow like this. Of course, you also need to adjust the image size accordingly.
If the output from the model you created is not coming out as well as the one created by someone else, there could be several reasons for this.
There can be multiple reasons for this. Therefore, it's recommended to approach it from a 'reference' perspective rather than a 'replication' perspective.
•
You might not have installed the 'negative embedding' or other specific elements used by that person.
•
Previously, WebUI was more commonly used before ComfyUI, and these two programs are not identical.
◦
For example, ComfyUI may work well with fewer steps compared to WebUI.
•
Additional work like upscaling or post-processing with traditional tools like Photoshop may have been done, resulting in cherry-picked examples.
HuggingFace
It is a model/code sharing site frequently used by machine learning researchers/engineers.
For example, let's visit the page of Stability, which developed the Stable Diffusion model.
You can see that 88 models have been uploaded and made publicly available below.
Model installed in nordy
Currently, the following models are installed on Nordy:
You can request to add models in the following way. (Details to be provided)
Openart
It is a workflow sharing site frequently used by ComfyUI users.
However, some workflows can be very complex and challenging, which might be difficult for beginners.
It is recommended to build a solid foundation with this document while also utilizing the workflows.
By thoroughly reading the content in this document alone, you will be able to create workflows at the level found on OpenArt.
Custom node installed in nordy
Currently, the following custom nodes are installed on Nordi:
You can request to add models in the following way. (Details to be provided)
GitHub
It is a code-sharing site frequently used by developers.
For example, to access the ‘image resize’ node provided in the ‘comfyui essential’ project, you can follow this path:
Then, you will see a complex screen that might look intimidating to developers. Don’t be afraid.
In the document labeled "Readme" below, there is often documentation on how to use the project effectively. So, if additional information or examples about the node are needed, you can make good use of GitHub.
Import Trouble Shooting
When trying to run JSON files downloaded from external sources on Nordi/local, it often doesn't work smoothly.
This is due to the complex ecosystem of custom nodes in ComfyUI.
For example, there are five or more custom nodes with the same name, such as 'Set'. Some users create variants like Load LoRA Simple or Load LoRA Extremely Simple because they find it inconvenient to connect Clip to Load LoRA, leading to a quite cumbersome experience.
When encountering such 'red boxes', you might need to manually switch them out. Here’s an example:
For instance, although there are nodes with different names performing the same Resize operation, they still work well when connected.
