I don't know how you made it, but when I run this workflow, 2 of the 6 images always end up with weird text.
Flux1KontextDevGenerates_v10.zip
[flux1-kontext-dev] generates a comic from input image.zip