Recently akuma.ai seems famous (sorry if information is old). I was watching explanation, and felt similar thing could be reproduced in ComfyUI so I try it.
Work Flow
Usage Procedure
1. Introduction of ComfyUI
Introduce ComfyUI by following method.
>-
2. Installation of Extension
Please install following extension to ComfyUI. (Via Manager is easy)
Custom nodes that extend the capabilities of Comfyui - AlekPet/ComfyUI_Custom_Nodes_AlekPet
3. Download and Placement of Model
Download LCM-LoRA and place in directory for LoRA.
LCM-LoRA - Acceleration Module! Tested with ComfyUI, although I hear it's working with Auto1111 now! Step 1) Download LoRA Step 2) Add LoRA alongsi...
4. Import of Workflow
Download from link destination below and import.
5. Setting of Model
Set LCM-LoRA to Load LoRA node, and set favorite base model to Load Checkpoint node.
6. Setting of Auto Queue
Set so that generation process runs automatically when paint or parameter is updated.
Specifically set Extra Options as below.
Setting of Auto Queue
7. Generate
When Queue Prompt, result generated from scribble comes to be reflected in Preview Image node.
Tips
- Please tweak prompt according to what you want to generate. Good result might come out easily if using embedding like EasyNegative.
- Since it is trouble if image becomes different at every generation, seed is fixed. Since it can be changed at Primitive node, please change manually if seed was miss.
- Since it is fine if it can generate in real time even if not LCM-LoRA, I think it is good to replace with Lightning, Turbo, Hyper models.
Generation Result
Please don’t mind my drawing is poor…
Conclusion
Isn’t it generated relatively nicely? It is hard to understand since there is no video, but if you actually try, result painted is reflected almost in real time.
Since I think trying freely for free is strength of open source, those who thought interesting please try.
