Two ways to write prompts: the freeform approach vs. the precise description approach. Here's what I learned after messing up a bunch of times.

I switched from coding to AI art, and at first I kept trying to write prompts as precisely as function parameters, locking down every detail. But a lot of the time the images came out stiff. Then I came across this way of splitting it that really clicked: one is the vibe approach—you give a vague direction, a mood, or a few style cues, let the model do its thing, good for when you don’t know what you want yet and are just collecting ideas.

The other is the descriptive approach—spell out the lighting, composition, lens, all that, chasing control. My workflow now is to cast a wide net with the vibe approach early on, then once I see something close to what I had in mind, switch to descriptive and nitpick every detail.

The tools themselves—Midjourney, Firefly, SD, Ideogram—are pretty much the same; what really makes the difference is the words you feed in. Pure trial and error is way too inefficient, you gotta consciously switch between these two modes. Which one do you guys lean towards more?

I totally agree with the idea of going wild in the early stages and then fine-tuning later.

Classic programmer disease—trying to parameterize everything, lol.

Yeah, going with inspiration-based generation definitely leads to some nice surprises.

Writing the prompt too densely just boxes the model in.

I usually just go straight with a descriptive prompt, can’t be bothered to cast a wide net.

Different tools have different tolerance levels for vague terms.

I’ve been burned by that too—overloading the prompt just boxes the model in. Leaving some breathing room for it to improvise sometimes gives you way more surprising results.