My results don’t look like what I asked for

Add specifics, not adjectives

"Beautiful", "amazing" and "high quality" give a model almost nothing. Concrete detail does: the lens, the lighting direction, the time of day, the material, the camera angle, the mood. Every specific you supply is a decision the model no longer makes randomly.

Check you are on the right model

• Prompt being partly ignored? GPT Image 2 is the strongest instruction follower.
• Text in the image coming out garbled? Use Nano Banana Pro or GPT Image 2 — most models cannot render text reliably.
• A person’s face changing between generations? Use the Image Editor with Nano Banana Pro, which is built for likeness.
• Style not landing? Seedream 4.5 and Dreamina 3.1 are the artistic models; Recraft 3 has an explicit style picker.

Iterate cheaply

Test on Z Image at 2 credits and change one thing per attempt. If you change three things at once you will not know which one helped. Once the wording works, run it on the expensive model.

For video specifically

If the subject is wrong, do not fight it in text-to-video. Generate the exact frame you want in the Image Generator, where iterations cost single-digit credits, then use image-to-video to animate it. Describe the motion and camera movement explicitly — models default to very little movement if you do not.