I'm getting mixed results with Kontext (FP8 scaled). I'm inputting two images and trying to put Keanu Reeves on the surf board. Kontext only accurately recreates the Keanu's face when the resolution is 1024 x 1024 but then the rest of the image is changed. Using similar or exact resolutions as the surfboard image yields poorer results but the composition of the background is better?

Does attention/emphasis (prompt:1.5) work? I've been elbows deep in Automatic1111 since Spring and having a blast. What of the bird? But I'm always exploring better ways to do this. Ever since I saw that little ComfyUI avatar in the SDXL live chat last week, I had an interest in checking out the app, especially since it was so stable at launch but I must be doing something wrong?