Hello folks,
inspired by the image from the hidream github, i made this comparison of qwen, kontext, hidream editing models.
I chose to use the models that fit my 12 GB of vram, they are Q4_K for hidream, kontext FP8 scaled and qwen Q5_0 (because with Q4 i have problem with lora).
Only for qwen edit i used the 4 step lora that is available, all the workflow are the same for each models and for each image.
I picked the first generation with random seed for every image in the list.
Qwen is the winner with a good understanding of the prompt and precision on editing part of image.
All the image, original and output with metadata, prompt, workflow are attached in the article.
