The era of text-to-image prompts filled with dozens of descriptors is over. These days, it's better to start with simple prompts and build them up gradually.
I wonder how many folks (since this was written) have switched from Team Splatterprompt to Team Just Have a Conversation Already. For folks who use these things every day, that is consistent advice I hear, and I can vouch for how much more effective the results are if I just commit to "talking" for a minute.
I'd imagine the common wisdom has generally shifted toward "Just write normally," especially after DALL-E 3 came out, which was even better at natural language and you could click on images in ChatGPT to see the exact prompts it wrote.
Still, I come across splatteprompts in the wild often enough to know it's not a solved issue yet.
Bit late to the party but this was a great read. The splatterprompts you see elsewhere make the generative models seem far less accessible than they are. "Ghee if I have to remember all those weird incantations, why even bother?"
And I couldn't agree more. That's precisely what bugs me about all the "secret prompts" chatter. It makes something that's extremely accessible seem like it isn't. This article is over a year old, and it's only gotten more relevant now.
These days, text-to-image models are so good at natural language understanding that it's usually enough to just explain what you need in simple terms. And if you're stuck, you can even turn to a chabot for help, as I outlined here: https://www.whytryai.com/p/ai-images-chatbots
Thanks! This was helpful and fun to read. I find that Midjourney is really responsive to the names of famous artists. A one line description along with ‘Jeremy Lipking’ always makes beautiful oil painting ideas.
Ha, same! I used to copy-paste long shared prompts without thinking much about it. It's a learning process. And thanks, happy you find the articles useful. Now go out there and prompt away...without splattering!
It's especially great for Nightcafe since they're using Stable Diffusion behind the scenes, and Urania specifically catalogues SD's artist influences.
These days I mostly use Midjourney, where the artist correlation isn't always as pronounced. I generally try to describe scenes in a nuanced way rather than trying to rely on arstist modifiers nowadays, but they're still a nice tool to have in the arsenal!
I wonder how many folks (since this was written) have switched from Team Splatterprompt to Team Just Have a Conversation Already. For folks who use these things every day, that is consistent advice I hear, and I can vouch for how much more effective the results are if I just commit to "talking" for a minute.
I'd imagine the common wisdom has generally shifted toward "Just write normally," especially after DALL-E 3 came out, which was even better at natural language and you could click on images in ChatGPT to see the exact prompts it wrote.
Still, I come across splatteprompts in the wild often enough to know it's not a solved issue yet.
It's definitely still out there, and I think it's still very common... it might even be the dominant paradigm, even today.
Bit late to the party but this was a great read. The splatterprompts you see elsewhere make the generative models seem far less accessible than they are. "Ghee if I have to remember all those weird incantations, why even bother?"
Happy to hear you enjoyed it!
And I couldn't agree more. That's precisely what bugs me about all the "secret prompts" chatter. It makes something that's extremely accessible seem like it isn't. This article is over a year old, and it's only gotten more relevant now.
These days, text-to-image models are so good at natural language understanding that it's usually enough to just explain what you need in simple terms. And if you're stuck, you can even turn to a chabot for help, as I outlined here: https://www.whytryai.com/p/ai-images-chatbots
Thanks! This was helpful and fun to read. I find that Midjourney is really responsive to the names of famous artists. A one line description along with ‘Jeremy Lipking’ always makes beautiful oil painting ideas.
Glad you found it helpful! And yeah, artist tags are the most powerful. I've done a test back in the day with Stable Diffusion and reached the same conclusion: https://www.whytryai.com/p/ai-art-experiment-5-artists-3-prompts
And yes, Midjourney is just as responsive, especially to famous artists.
Guilty as charged! Great site btw with some fab articles... I promise to stop splatterprompting!
Ha, same! I used to copy-paste long shared prompts without thinking much about it. It's a learning process. And thanks, happy you find the articles useful. Now go out there and prompt away...without splattering!
And this is a godsend! https://www.urania.ai/top-sd-artists I use Nightcafe mostly, its very easy and quick!
Yeah it's a great resource! I used it for my artists experiment back in September last year: https://www.whytryai.com/p/ai-art-experiment-5-artists-3-prompts (see the "Methodology" section).
It's especially great for Nightcafe since they're using Stable Diffusion behind the scenes, and Urania specifically catalogues SD's artist influences.
These days I mostly use Midjourney, where the artist correlation isn't always as pronounced. I generally try to describe scenes in a nuanced way rather than trying to rely on arstist modifiers nowadays, but they're still a nice tool to have in the arsenal!