Creations by GravidaSomnia

()

That's a sound approach and apart from gigapixel it's how I startes trying to improve the face situation, first by just redoing the faces but now with the whole image. It can require more work but I like going with quality over quantity. Occasionally we get both.

I'm keen to get gigapixel, GFPGAN upscaling does a great job on faces but gigapixel seems to do all the details.

()

Finally have my own hardware and it's a much better workflow for creating artistic pieces. Render speed is about the same but viewing and transferring the output files was always slow and clunky for me when using the a1111 webui colab.

()

started out trying to do some more vintage art style pics but ended up with this instead:

()

Another Draft that probably won't go anywhere:

()

Experimenting again with making waifu art photorealistic:

It's still very much trial and error, this is the best result so far. Using weights that are a blend of SD1.5 and the f111 helped immensely, even though the latter is geared towards adult content it's based on photos so works better for this sort of thing than vanilla SD 1.5. I didn't think I'd use it much because I don't aim to replicate porn but it's been great for giving a photographic boost to both nsfw and sfw artistic content. There's a sweet spot for cfg vs. denoising strength vs. interations that improves the likelihood of a good transition but it seems to vary betwen inputs, there's no magic numbers but I can at least say what worked this time. The euler scheduler seems most reliable.

Parameters: Steps: 24, Sampler: Euler, CFG scale: 8, Seed: 1448501382, Size: 704x1024, Model hash: c276019b (SD1.5 x f111, 0.4 ratio IIRC, float16), Batch size: 2, Batch pos: 1, Denoising strength: 0.7

The prompts are too cringe to share sorry but ofc there's (((photo))) etc. in the prompt and ((((drawn)))) etc. in the negative prompt. And even then it's not necessarily a photographic-style output.

I've had success doing this using the vanilla SD 1.5 model in the past and even just using DreamStudio's img2img, but couldn't make it work reliably.

Doing the inverse - transforming a photo into anine styke art - is relatively simple, just use img2img and the Waifu or NovelAI models:

Quite often Stable Diffusion won't reach the desired result in one go, but using a less than perfect output with img2img takes things closer to where we want them to go,

()

Results aren't reliably fantastic but I've made progress in getting decent pics out from a single operation with the a1111 highres fix rather than running img2img separately and then collating the best bits together. It does a lousy job on pics in portrait orientation but square and wider pics can come out ok a lot of the time with the right prompt and setting.

()

I'm really enjoying working on individual pics rather than trying to mass produce them, but these were quick and lazy:

()

This one was a bit more work but still came together really easily:

OliviaCohen · (357)

That last one is cool... it's like you'd almost expect them to be singing a group song. Like that scene from O Brother Where Art Thou, when they're all walking to the river to get baptized.

()

I could almost do a story for each one but for now I'll let people use exercise own imagination.