I’m trying to create an AI picture for a project, but I got stuck figuring out which tool to use and what prompt to write. I tried a few generators and the results came out weird and nothing like what I needed. I need help with simple steps, good prompt tips, and the easiest way to make an AI image that looks right.
Pick the tool based on the job.
If you want fast concept art, use Midjourney.
If you want easy prompt control, use DALL·E or Adobe Firefly.
If you want the most control, use Stable Diffusion with SDXL.
Your prompt needs 4 parts.
Subject, style, setting, details.
Example:
‘Young woman in a red raincoat, cinematic photo style, standing in a neon-lit Tokyo alley at night, wet pavement, soft fog, shallow depth of field, realistic skin, detailed eyes’
If results look weird, fix one thing at a time.
Add camera terms for photo looks.
Use artist or medium terms for art looks.
Use negatives if the tool supports them.
Example:
‘blurry, extra fingers, bad hands, warped face’
Also set aspect ratio first. A poster, thumbnail, and square post need diff sizes.
Best simple workflow:
- Write one clean prompt.
- Generate 4 images.
- Pick the closest one.
- Edit the prompt.
- Repeat 3 to 5 times.
If you post what image you need, people here can help write the prompt.
Don’t start with the generator. Start with a reference board.
That’s the part ppl skip, then wonder why the AI gives them cursed hands and Walmart-poster energy. Grab 3 to 5 reference images that match the vibe, lighting, angle, and color you want. Then describe those, not the picture in your head. Your brain is vague. References are not.
I’d also push back a little on the “just prompt better” advice from @vrijheidsvogel. Prompting matters, sure, but a lot of weird outputs happen because the request is doing too much at once. If you want “fantasy watercolor cyberpunk professional headshot with dramatic smoke and soft children’s book warmth,” the model is gonna have a lil meltdown.
Try this instead:
-
Decide the goal:
- realistic photo
- illustration
- logo/graphic
- background scene
-
Write a boring prompt first.
Example:
“A realistic portrait of a teacher in a classroom, natural window light, medium shot, neutral expression” -
Then add only 1 to 2 upgrades:
- “warm tones”
- “cinematic”
- “shallow depth of field”
-
If it still looks off, use image-to-image or upload a reference. Honestly, that helps more than adding 40 adjectives.
Also, if your project is for school/work, double check usage rights. Some tools are stricter than ppl realize.
If you want, post the exact image you need and ppl can help rewrite the prompt so it’s less… haunted.
I’d add one thing people skip after the prompt stage: pick the tool based on the job, not hype. That’s where I slightly disagree with @vrijheidsvogel. A decent prompt in the wrong generator still gives junk.
Quick rule:
- Midjourney: great for mood and style, weaker for exact control
- DALL·E / ChatGPT image tools: easier for prompt-following and edits
- Adobe Firefly: safer for commercial-ish workflows, more restrained output
- Stable Diffusion: best if you want control, worst if you want easy
Big fix for weird results: lock the composition first. Ask for:
- camera angle
- subject position
- background simplicity
- aspect ratio
Example:
“Front-facing product shot of a handmade candle on a wooden table, centered composition, soft daylight, blurred kitchen background, realistic photography, 4:5 ratio”
Then do versions, not total rewrites. Change one variable:
- lighting
- style
- lens feel
- background
Pros of ‘’: can improve readability if you’re organizing prompt drafts or project notes.
Cons of ‘’: not really necessary unless you’re publishing or structuring lots of content.
Also, avoid negative-prompt obsession. Sometimes cutting 5 bad adjectives works better than adding 20 “do not” commands.