LoRA (Low-Rank Adaptation) is a technique used in Stable Diffusion to fine-tune models efficiently without requiring extensive computational resources. It allows users to adapt a pre-trained diffusion model to generate specific styles, characters, or themes by injecting a smaller set of parameters into the original model. This approach maintains the quality of the original model while enabling the creation of customized outputs tailored to specific creative needs. Within days of the release of the Flux.1 models, LoRAs became available to help you generate images that weren’t possible with the original models. In this article, I’ll walk you through the steps to use LoRA with the Flux.1 model using ComfyUI.
Requirement
- A working installation of ComfyUI with the latest updates.
- Nvidia GPU – 3090, 4090 You might be able to get this to work with a GPU with 16GB VRAM or less, but 24GB is recommended.
Workflow
- Download this lora and put it in ComfyUI\models\loras folder as an example.
- Drag the full size png file to ComfyUI’s canva.
- The workflow is like this:
- If you see red boxes, that means you have missing custom nodes. Use ComfyUI Manager to install the missing nodes.
- Restart ComfyUI
- Note that this workflow use Load Lora node to load a single Lora. If you need to load multiple Loras, you can use Power Load Lora node (part of rgthree-comfy custom node).
Examples
The prompt is the same for these images:
A stunning young woman stands gracefully in the center of a grand hotel ballroom. She wears an elegant, flowing dress that shimmers under the soft, golden chandeliers. Her hair is styled in loose waves, framing her face with a natural beauty. The ballroom is lavishly decorated, with tall, ornate mirrors, rich drapes, and polished marble floors that reflect the warm glow of the lights. The atmosphere is filled with sophistication and charm, making her the focal point in this luxurious setting.
w/o Lora
w/ Lora
w/o Lora
w/ Lora
w/o Lora
w/ Lora
Further reading:
How to Use ChatGPT to Write Flux.1 Prompts in T5 Style with an Input Image
This post may contain affiliated links. When you click on the link and purchase a product, we receive a small commision to keep us running. Thanks.
Leave a Reply