Greg rutkowski, dark, surreal scary swamp, terrifying, horror, poorly lit Like if I was imagining a spooky piece and thought the results of the above prompt weren't scary enough I might change it to: The AI also pays attention to emphasis! If you have something in your prompt that's important to you, be annoyingly repetitive. Your piece is a little TOO dark? Move " dark" in your prompt to the very end. If your faces aren't detailed enough? Add something like "highly-detailed symmetric faces" to the front. Greg rutkowski, dark, scary swamp, terrifyingĮssentially, each chunk of your prompt is a slider you can move around by physically moving it through the prompt. If I had the prompt above and decided I wanted to get a little more greg influence, I could reorder it: The things near the front of your prompt are weighted more heavily than the things in the back of your prompt. These types of variations can have a massive impact on your generations. Or " character concept art by greg rutkowski". "moody greg rutkowski piece" instead of "greg rutkowski" is cool and valid too. See how I'm separating descriptions from moods and artists with commas? You can do it this way, but you don't have to. This prompt is an example of one possible way to tokenize a prompt. Scary swamp, dark, terrifying, greg rutkowski Find out what fans and art critics say about an artist. Studying artists' techniques also yields great prompt phrases. There's always a reason in your prompt (although sometimes that reason can be utterly inscrutable).Īllow me a quick aside on using artist names in prompts: use them. Look at the image and try to see why the AI made the choices it did. If it looks bad, add or remove words and phrases until it doesn't look bad anymore. At the early stages of prompt engineering, you're mainly looking toward mood, composition (how the subjects are laid out in the scene), and color. Go back to the drawing board on your prompt. If your image looks straight up bad (or nowhere near what you're imagining) at k_euler_a, step 15, CFG 8 (I'll explain these settings in depth later), messing with other settings isn't going to help you very much. Get yourself a GUI that tells you when you've hit this limit, or you might be banging your head against your desk: some GUIs will happily let you add as much as you want to your prompt while silently truncating the end. Just keep in mind every prompt has a token limit of (I believe) 75. The AI is weirdly intuitively trained to see "Wow this person has a lot to say about this piece!" as "quality image". My theory for this is that people don't waste their time describing in detail images that they don't like. Typically, the longer and more detailed your prompt is, the better your results will be. Don't feel the need to add phrases one at a time to see how the model reacts. You can add as much as you want at once to your prompts. Ultimately, the process of creating images in Stable Diffusion is self-driven. Steal their prompt verbatim and then take out an artist. Take note of phrases used in prompts that generate good images. Flip through here and look for things similar to what you want. Sites that have repositories of AI imagery with included prompts and settings like are your god. But I can go over some good practices and broad brush stuff here. Prompting could easily be its own post (let me know if you like this post and want me to work on that). Save this post and come back to this guide when you feel ready for it. These settings are completely fine for a wide variety of prompts. Note: if the thought of reading this long post is giving you a throbbing migraine, just use the If anything I say sounds totally different than your experience, please comment and show me with examples! Let's share information and learn together in the comments! I'm going to go over basic prompting theory, what different settings do, and in what situations you might want to tweak the settings.ĭisclaimer: Ultimately we are ALL beginners at this, including me. In this guide, I'm going to talk about how to generate text2image artwork using Stable Diffusion. These settings will give you a better experience once you get comfortable with them. There's sliders everywhere, different diffusers, seeds. So you've taken the dive and installed Stable Diffusion. Beginner/Intermediate Guide to Getting Cool Images from Stable Diffusion
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |