Moss Image vs Qwen 2.1 ( Qwen wins)
retained text
So i played around with qwen 2.1 with image in, prompt, image out. (NO LORAS) At first it was okay but things would mess up or didnt look real So i heard everyone raving about MOSS IMAGE and some loras So i download and test it out. It was HORRIBLE and following directions. Same prompt in QWEN, QWEN got 1st. MOSS was just a mess the output image didnt look ANYTHING like the first. Or if i could do it, it couldn't figure out positioning. It was just a mess and i tried multiple things working with chatgpt and even chatgpt gave up in the end and said just use QWEN So i go back to QWEN, this time i up CFG from 1.0 to 4.0 (boom most of the issues with poses, anatomy) vanished. It looks way more realistic. I also upped steps from 25 to 30. Qwen just follows imagery better in my opinion. Now maybe moss is okay with text to image but not image to image. Its trash in that department.
document fetched 2026-11-19 00:42 · canonical https://reddit.com/r/comfyui/comments/1xzljc6/moss_image_vs_qwen_21_qwen_wins