Found a free way to tweak AI art prompts using image descriptions

I was stuck getting the same generic fantasy landscapes from Midjourney no matter how I rewrote my prompts. Then I tried feeding a photo I liked into a free image description tool online and pasted that text back as my prompt. The results were way more specific and matched the vibe I wanted after just 3 tries. It felt weird using a machine to talk to another machine, but it worked better than my guesswork. Has anyone else tried using alt text or descriptions as a shortcut for better AI art prompts?

3 comments

3 Comments

vera1951mo ago

Yeah, "using a machine to talk to another machine" sums it up perfectly. I started doing this with Clip Interrogator for images I found on art stations. The key is you gotta edit that description down a bit first. It spits out way too many details like "soft volumetric lighting" and "octane render" that confuse the process. I strip it to just the subject, the main colors, and the general mood. Got a specific picture of a rainy back alley that way instead of the usual generic city streets.

drew_jones311mo ago

Yep, you nailed it. Gotta strip it down to the bones. I do the same thing with any description I feed in. Like, I'll keep "woodland path, mossy stones, soft morning light" but cut all the "sharper image standard" and "raw photo" tags. Those just tell the AI to copy a style instead of the actual scene. It's like the difference between asking for "a photo of a forest" and "a photo of a forest that looks exactly like this other photo you've seen.

aaronsullivan1mo ago

Whoa hold on, you edited the CLIP output? That's wild. I never even thought to strip out all that technical junk. I just fed it the raw mess and wondered why my results looked like every other generic synthwave wallpaper. So you literally cut out "octane render" and "soft volumetric" and bam, you get a specific rainy alley? That feels like cheating in the best way. I gotta try this right now with a moody forest pic I've been fighting with.