Finally got a decent AI piece after 40 tries, and the secret was research
I kept prompting generic stuff like 'cyberpunk city' and getting the same mush, but when I actually studied 12 reference photos of a real alley in Chicago and fed specific lighting notes into the prompt, the output finally had that gritty authenticity people keep claiming AI can't do, so has anyone else found that better input skills beat better models?
Honestly that tracks hard. Most people type two words and expect magic, but AI just averages everything it's seen. The real game changer is feeding it specific details like "wet asphalt, steam from a vent, sodium lamp glare" instead of vibes. I've had way better luck with prompts that sound like a photography assignment than any fancy model upgrade. It's like cooking, the ingredients matter more than the pan.