"The Shroomsons" by Dan Witz, Junji Ito, Benoit Mandelbrot, Beeple, James Gurney, Gerald Brom and Tom Bagshaw

😁

Hahaha I'm sorry 😅

No no it's not bad, but you can't deny this could be kind of spooky irl

lol thanks 😆

It's okay, not a masterpiece but it's a short read

Haha ikr, by far my favorite part!

What do you have in mind? I can tell you what prompts I would use for it.

I would love that! Thank you. Feel free to completely change it into something else if needed.

Maybe this one? /creation/ax65fQTX0V0jwiOImFtZ

OR /creation/G4KatZiDxd4uCjJHfS6E

OR /creation/vLG8CgFNXfDk32nXLKtX

Alright so a few tips to start: don't be afraid to use multiple prompts. I usually have the first prompt with the title and the artists, then repeat the title on the second prompt to make sure the AI really follows what I have in mind (might be placebo tbh but it doesn't seem to have any negative effect). Then I use a prompt with lesser weight (0.8 for instance) with a few details. One of my favorite is "deep colors" although it doesn't work well if you want to make something super bright and colorful. In that case I use "bright/vivid/vibrant colors" instead. Finally I use two prompts, one with 0.8/0.9 weight to increase the quality with tags like "detailed, unreal engine, sharp" and another prompt with negative weight (-1) with tags like "blur, blurry, unfinished, watermark". They're really effective. If there's something else you don't want the AI to add in your generation, add it there as well (like "human, man, woman, people" if you want to generate a scene without people in it).

Then I use these settings with a cheap artistic generation (short, thumbnail), with or without a starter, for 1 credit. This is enough to give me a good idea of how the result will look. Don't be afraid to try it with different prompts or starter pictures until you get what you want. Finally, that's where the magic happens. I use this artistic generation and evolve it with coherent mode. Coherent mode is definitely the supperior algorithm but I found it's way too random without a starter picture, and it works really, really well with artistic generations, especially if you use the same prompts for both. Then it's all about the settings. I usually always crank prompt weight to the max, and adjust the other two depending on how close I want the result to look compared to the artistic generation. I always do a short, medium res gen with accuracy boost at 2x or 3x. Accuracy costs a lot but it honestly gives much better results than raising the runtime.

What I think is the most important is finding the right artists and tags to use. Obviously, some artists are much better at generating different things, so don't be afraid to experiment. A few of my favorites are Beeple (adds a lot of colors and works wonderful with anything cyberpunk/with neons), Ferdinand Knab (adds a lot of details and works really well with buildings), HR Giger/Zdzisław Beksiński (for some alien/otherworldy theme), Tom Bagshaw (adds details and works well with portraits) and Bastien Lecouffe Deharme (adds a moody ambiance and works well with gothic horror), but I'm always experimenting with new ones to find what works with the AI. That's about it honestly. I hope those tips will help! 😃

This detailed explanation is so helpful! I feel like I needed to hear it from someone personally. Thank you! I will explore more with the -1 feature and the coherent mode too, specifically, I didn't think it worked very well...but it definitely will not work if I don't try it. I don't know what was holding me back from the multiple-prompts, but I'll explore more.

Thank you again for explaining all of these nuances, I will try to come back and experiment more over the next few months. Keep on creating!