Same parameters, different images - command line vs webgui - how come?

[nigeljohnson73]

Basically, if you get a nice pic in the web UI, then try and recreate it on the command line you get a similar, but different image.

Passing the same prompt, DDIM sampler, Width, Height, CFG scale and seed does not yield the same image, also, the web UI version is also nicer in my opinion. What else is it setting in the web UI version, and how can I re-create that in the command line version?

I'm not quite sure where to start my journey, so apologies if things don't hang together, but, I started in collab with the huggingface stuff and was happy producing batches there. I am unable to run stable diffusion locally because of the 4 computers I have, none are powerful enough or NVIDIA-enabled.

I did find a collab version that git clones the automatic1111 stuff and downloads the runwayml/stable-diffusion-v1-5 model (Which is what I was using before). I started playing with the WebUI, and I got a corker of an image. In my old stuff, I collect and add meta-data so I can upres and do all the PNG stuff later (not on a google GPU), so I set it up and generated it... but I got a different image.

I've gone back to the automatic1111 collab and built a code block that runs the command line, and I get an image that's basically identical to the huggingface code and not the web-ui.

The web UI does do something different. In the webUI version, I can generate a single image that is 512x1536 using the standard GPU profile. However, to generate the same image in the command line, I need to use the Premium GPU profile. Also, the images tend to fit within the boundaries of the image, whereas the command line version seems to want to crop the top off mostly.

Any help much appreciated.