Cassandra Is Found - This random seed is the one!

Cassandra Is Found - This random seed is the one! | Ecency

I think I've found my winner.

This was the first new iteration of a new work session on my mythological theme of Cassandra. Sure, there's not people running around in the background, and there's a bit of a boring composition, but boy does her facial expression tell a story.

Needs a tiny bit of work, but gosh, this is close to my prompt, and something I can experiment with further by using some different aspect ratios.

This was generated at 1024*512, with a guidance scale of 7.5, and 50 inference steps.

Seed: 3531576

highly detailed cinematic portrait color photograph of a beautiful ancient greek woman perfect facial symmetry, flames behind, shot in the burning hallway of a ruined temple on fire in flames, figures running around in the background, ultra realistic, depth, beautiful lighting, by richard avedon and annie leibovitz and arnold newman, photorealistic, hyperrealistic, octane, epic composition, hasselblad camera, 35 mm, sharp focus

I want to scale it up to 1024 x 1024 to see what happens...

Well, I accidentally moved the guidance scale... and got this. Whoops. Still an interesting image with the same seed, but not the same. Attempting to address the error of my ways now, and it is an agonising minute and a half wait.

I feel as though through Stable Diffusion I have a reason to upgrade my GPU beyond my 3080, and I'm not angry about that. My wallet will probably be angry about that, though.

I can deal with it though, if it keeps me happy making art.

Hmm. Apparently seeds don't work the way I thought they did. Let me try to get the original image back again...

Same seed, different guidance, same res, different result to the above. Hoping I can get back the oiriginal image with the same settings as before!

These waits are truly agonising!

I upped the iterations to 200 in order to get a sharper image, and this one, I will be upscaling properly. Once that was done, I tried to break my GPU by giving some instructions for in-painting to add further detail.

This is not the expected result of inpainting. Not blank space...

Before settling on the base image, I want to see ten more images on this theme be generated. I'm greedy. I'm feeling lucky. I could be so very wrong about this. Let me see what happens. I'll find out in three and a half minutes or so, which means about thirty seconds per image generation.

The first two are strange, so I do not mind one bit. I'll keep going. Interesting second one though, with such detail in the faces of a relief on a wall. That is very, very impressive.

The first lot of faces are amazing. The second had some NSFW events which are self censored in the below screenshot, it didn't ad anything artistic to the image, but it is interesting to see just how much the same prompt can vary from run to run, based purely on a random noise generator.

Here are the next three:

Cool faces, but well away from my prompt.

Interesting flames in the background, but non-corporeal faces are not welcome :D

... Speaking of non-corporeal faces...

I'll just hang my mask up on the wall.

The final three results from the last batch of ten aren't even worth copy/pasting. They were awful renditions. I'm lucky that these were interesting.

In the next post, I'll be running off to Photoshop to work on the output image that I am happy with, and see what I can do to make it be ever so slightly more tweaked. That will be on another day though, as I am incredibly tired at this stage.