This workflow is using flux redux to replicate style or character without training LORA.
You can add up to 6 images. If you want to use less, remove them from Select Images section. Remember that you have to count from 0. You will still have to attach a image but it will not be used.
Available on SeaArt: https://www.seaart.ai/workFlowAppDetail/cvh6csde878c73cdlsh0
Available on TensorArt (as a workflow): https://tensor.art/workflows/823027897252986921
Description
FAQ
Comments (3)
Do you have any documentation or a video for this workflow? I added 3 images (a t-shirt, a person, and a car), and bypassed the other load image nodes that I was not using. I created a prompt along the lines of the person wearing a t-shirt standing next to the car. Typically, what got generated was a variation in the first image and nothing related to the other images. Changing add 1 image to latent made no difference. I would like to see how this is supposed to work, because what I think it does, and what is happening, are not the same thing.
Unfortunately, I'm afraid that this workflow won't work in this situation. Flux redux is used to create images very similar to the ones given. It is designed to do things that you would normally use LORA for, meaning you have to use multiple images of the same character to make more images of that character.
To do what you want you need to use different technology. You can try to use search for a workflow that uses Flux UNO, but it requires a very powerful GPU
@LodWKobku Ok, thank you for the response. I misunderstood what this workflow does. I have set up the gradio app for UNO, but it does a poor job of keeping people objects similar. It typically does a good job with objects but not with people. I will have to go back and revisit this workflow.