Modifies the first image to apply the color palette supplied in the second image. Prompting is capable of targeting specific elements of the image.
To generally impose the color palette, use this prompt:
Apply this palette to the entire imageTo only apply the palette to specific elements, use this prompt:
Apply this palette to <object>
Apply this palette to the their outfits and the trees.
Apply this palette to the women's dressesThe model is trained to generally avoid affecting the ambient light and skin tones, but if you wish to add a baseline color cast, changes the ambient light, use:
Add a <color> hue
Add a blue hueTo remove a color cast from an image that already has one:
Remove the <color> hue
Remove the tan hueDescription
FAQ
Comments (3)
What kind of data is needed to train a LORA like this, using multiple images, and make it follow the prompt using multi image correctly?
I train it using Ostris AI Toolkit. That platform allows you to define multiple datasets, and Image Edit LoRAs like this one take an input of three datasets: the control (original image/Image 1), the secondary (color palette/Image 2), and Target (output).
I created this particular LoRA by doing the process in reverse with synthetic data. I generated an image, then used Qwen/Klein to tell it things like "change the color of outfit/hair/etc to X." From there I had the before/after pairing, then I used my Extract Color Palette LoRA to generate Image 2.
I got started by following this tutorial; it's for Qwen, but the dataset construction also applies to Klein: https://youtu.be/d_b3GFFaui0?si=q2JoL5F3DVjTF1JD
@ReltivlyObjectv Thank you, sir.