Image size changes affect the final result
I've noticed that when I change the image size (for example, from 16:9 to 9:16), the result changes even with the same settings. It doesn't seem to be a crop.
Personally, I'd like all horizontal outputs to be 4:3 and vertical outputs to be 3:4, then cropped to the desired size, for example: 4:3 » 3:2 for horizontal and 3:4 » 2:3 for vertical.
Real example
Model: Stable Diffusion 3.5
Seed: 80707671
Dimensions: 848x1264 (2:3)
Image Style: None
Adherence: High (Actually, it's Medium; there's an error in the adherence scale)
Negative Indication: None
Steps: 25
LoRA Strengthening: 75
Secure Venice: Off
Embed Exif Metadata: Off
Blend Strength for Retouching: N/A
Model: Stable Diffusion 3.5
Seed: 80707671
Dimensions: 1264x848 (3:2)
Image Style: None
Adherence: High (Actually, it's Medium; there's an error in the adherence scale)
Negative Indication: None
Steps: 25
LoRA Strengthening: 75
Secure Venice: Off
Embed Exif Metadata: Off
Blend Strength for Retouching: N/A
Log in to comment and vote
Comments2
Silver River
Apr 5, 2025
I think that has to do with how image generation works. It starts with a latent image which is pretty much static and tries to interpret the prompt onto the image. With images of different sizes the outcome would be different simply because the method would be applied differently.
If you’re looking to make the same image larger, as in include into the image what was out of frame before, you might want to look into outpainting with something like comfyui or A1111 or other services that provide it, or make an feature request of it.
JP
Apr 28, 2025
This is correct. Different sizes result in different generation for the same prompt data. We can’t structurally change that.