Found this and been very distracted.
https://github.com/vitoplantamura/OnnxStream
Stable Diffusion running on a Pi5.
Makes 512x512 images in 3minutes, not too good with people etc but landscapes are good enough for model train backdrops.
Except they are only square images, while a model train background is usually ultra, ultra wide.
Background1a was the first one that I did.
Background7a was after learning to prompt better.
Background13a was to see if I could make it look a bit like my Australian country side.
Background17a is very similar to what I have seen flying over Oz country at a thousand feet or so.
XL makes 1024x1024 but takes 30minutes so takes longer to tweak the prompts.
I believe it was trained on different formats but I cannot figure out how to do 1024x512 etc.
Somewhere buried in Stable Diffusion is a LLM maybe.
Not sure which one, maybe other LLM's I now have can make images?
I have no idea why the square format was used, maybe for general use with figures?
For panoramic landscapes an ultra wide image would be fantastic.
Can SD do different countries landscapes?
Probably, but being able to do stuff this on a Pi5 is still amazing![:D]( "Very Happy")
Expect to see the Pi5 get redlined.
And I wonder if it would run a bit faster with a NVMe drive?
Somewhere I have an old paid for Bryce 2.0? CDROM, this is so much better, quicker and easier.
https://github.com/vitoplantamura/OnnxStream
Stable Diffusion running on a Pi5.
Makes 512x512 images in 3minutes, not too good with people etc but landscapes are good enough for model train backdrops.
Except they are only square images, while a model train background is usually ultra, ultra wide.
Background1a was the first one that I did.
Background7a was after learning to prompt better.
Background13a was to see if I could make it look a bit like my Australian country side.
Background17a is very similar to what I have seen flying over Oz country at a thousand feet or so.
XL makes 1024x1024 but takes 30minutes so takes longer to tweak the prompts.
I believe it was trained on different formats but I cannot figure out how to do 1024x512 etc.
Somewhere buried in Stable Diffusion is a LLM maybe.
Not sure which one, maybe other LLM's I now have can make images?
I have no idea why the square format was used, maybe for general use with figures?
For panoramic landscapes an ultra wide image would be fantastic.
Can SD do different countries landscapes?
Probably, but being able to do stuff this on a Pi5 is still amazing
Expect to see the Pi5 get redlined.
And I wonder if it would run a bit faster with a NVMe drive?
Somewhere I have an old paid for Bryce 2.0? CDROM, this is so much better, quicker and easier.
Statistics: Posted by Gavinmc42 — Sun Mar 10, 2024 8:39 am