Been playin around with #ComfyUI and #SDXL models for a few months now, what a fun tool. Lately I've been rather obsessed with using it to creating some ""paintings"" that I have floating around in my head. I learned most of the technique from a fella on YouTube called Rob Adams, with a bit of side reading to figure out what some of the things were I needed to play with to get what I want from his method.
I start with a basic two-tone mask that I crudely draw to give the model a main direction for the things I describe, then add a noise layer on top of it. From there it moves on to the first pass which takes an overly (#ollama) complicated series of concatenated input prompts a
... Show more...
Been playin around with #ComfyUI and #SDXL models for a few months now, what a fun tool. Lately I've been rather obsessed with using it to creating some ""paintings"" that I have floating around in my head. I learned most of the technique from a fella on YouTube called Rob Adams, with a bit of side reading to figure out what some of the things were I needed to play with to get what I want from his method.
I start with a basic two-tone mask that I crudely draw to give the model a main direction for the things I describe, then add a noise layer on top of it. From there it moves on to the first pass which takes an overly (#ollama) complicated series of concatenated input prompts and gives me a very rough (1024²) idea of where I'm headed. Then I crop a 16:9 section from that and scale it up a bit; this is also where the most of the color tweaking is done. Once I'm happy with this ""starting point"", I pass it on to another run through a different set of models to get my final image. Once I adjust the lighting on that one, it get's upscaled again for use as merely a wallpaper, maybe I'll print em out one day ...maybe.
Finally, today I've gotten my image composition layout to work the way I want, so I figure I'd share a few samples (1/4 scale).
Mathias
in reply to eshep • • •They look great. I obviously get curious about the prompts too.
As a sidenote: your images are 25-30mb. Each. In most of the services where I can see your post I get served a lo-fi representation of the images. Only when I click the separate images do I get the full resolution and detail, however they are much larger than the screen size I have access to.
It would probably be a good thing to resize/compress images for social sharing in a controlled fashion, that does the images justice, and perhaps link off to the full-size images if that is what you wish to share. Just a thought.
eshep likes this.
eshep
in reply to Mathias • • •Thanks, it's been loads'a fun so far! The prompts for txt2img can be quite finicky, that's why I've added in the LLM to serve as a translator from human to robot.
And thanks for the heads up on the size, I hadn't bothered to ever check if images get autocompressed on upload to here ...guess they don't.
Mathias likes this.
eshep
in reply to Mathias • • •