Help see your character from multiple angles. Qwen Image Edit is pretty good at rotations by itself. This Lora just produces a composited image of character from multiple angles. Recommended: input resolution 600x1080, output resolution 2600x1080 or 3000x1080.
For Qwen 2511 it seems the output characters are always narrower than they should be. I find if you resize the output horizontally by an increase of 20% it works better. I trained the model multiple times but Qwen 2511 always produced narrow characters.
Description
FAQ
Comments (5)
It's very interesting but... (given that I'm a noob), I knew that the Qwen models were trained with the 1024 x 1024 format and outside of this format their quality could decrease, surely there are many exceptions, but please could you explain this issue to me?
1024x1024 is not the actual resolution. Its actually 1 megapixel which is equivalent to 1024x1024. So when I was training it with 3000x1080 images it actually lowered the resolution to 1664x576 which is roughly 1 megapixel. You can train with non square images.
@tarn59
Thanks for the reply!
Qwen Image Edit is wonderful!
@llmsonoio1234356 these are the resolutions that work best and trained on hope this helps
"1:1": (1328, 1328), "16:9": (1664, 928),
"9:16": (928, 1664), "4:3": (1472, 1104)
"3:4": (1104, 1472), "3:2": (1584, 1056),
"2:3": (1056, 1584)
Simple no bs. It works. Thank you very much for this lovely gem.