A Practical Prompt Review Checklist for Free AI Image Generators

A prompt written for a blog header is different from a prompt written for a product mockup or a social post. Before writing, answer three questions:

• Where will it be shown? A wide hero banner, a square feed post, a vertical story, a thumbnail?
• What is the one thing a viewer must notice? A product, a mood, a person, a piece of text?
• What will sit on top of it? Headlines and buttons need calm negative space; a standalone illustration does not.

Writing these down takes thirty seconds and saves a lot of regeneration later. If you cannot answer the second question, the image will almost always feel busy.

A lot of weak prompts mix everything into one long sentence. I find it easier to review a prompt when it has three clear parts:

1. Subject – what is in the frame, described concretely. "A ceramic coffee cup with a chipped rim" beats "a nice cup".
2. Setting – where it is and what light falls on it. "On a wooden windowsill, early morning side light, light rain outside."
3. Style – the medium and finish. "35mm film photo, soft grain, muted colours" or "flat vector illustration, two-tone palette".

When an output looks wrong, this split tells you which part to change. If the cup is right but the mood is off, you only touch the setting line.

Free generators usually default to a square. Square is fine for feeds but wastes space in a blog header and crops awkwardly in a vertical story. Choose the ratio first, then compose for it: for 16:9 put the subject on a third and leave room for text; for 9:16 stack the interest vertically.

The temptation is to pile on adjectives: "cinematic, ultra-detailed, 8k, award-winning, dramatic, trending". In my tests these stacks rarely help and often make every image look like the same glossy stock photo. Instead, add one meaningful constraint per iteration:

• iteration 1: subject + setting only
• iteration 2: add a lighting direction
• iteration 3: add a lens or medium
• iteration 4: add a colour palette

Compare each round with the previous one. Keep the change only if it clearly helped.

Negative instructions ("no text, no watermark, no extra fingers") are useful for fixing a specific recurring problem. They are not useful as a long defensive list added to every prompt. Add a negative only after you have seen the problem twice.

When a batch comes back, it is easy to pick the "prettiest" image. Instead, go back to the answers from step 1:

• Does the important thing read clearly at the size it will be shown?
• Is there space for the headline or button?
• Would it still work in grayscale? (A quick test for contrast and hierarchy.)
• Are there artefacts – warped hands, melted text, repeated objects – that a viewer would notice?

Mark each image pass or fail on these questions before deciding which one you like.

The single habit that improved my results most was keeping a plain text log: the prompt, the ratio, the tool, and one line on what worked. After a few weeks you have your own library of reliable starting points. When a client or teammate asks for "something like last month's header", you can reproduce it instead of guessing.

A minimal log entry looks like this:

date: 2026-10-15
tool: Yeri – https://yeri.ai
ratio: 16:9
prompt: ceramic cup, chipped rim / windowsill, morning side light, rain outside / 35mm film, muted
result: good mood, cup slightly too centred – next time "subject on left third"

Some generators, Yeri included, can turn a still into a short clip. Motion is great for social posts and landing page loops, but it also exposes flaws: a hand that looked fine in a still may wobble in motion. Only animate stills that already pass step 6, and keep the motion simple – a slow push-in, drifting steam, moving rain – rather than asking for complex action.

Before publishing, read the terms of the tool you used. Most free generators allow personal and commercial use of outputs, but details vary. For editorial contexts, it is good practice to note that an image is AI-generated, especially if it depicts something that looks like a real event or person.

This is why I wrote the checklist on Draftmark: images benefit from a second pair of eyes just as much as text does. Before a set of generated images goes into a campaign, share the prompt log and a few candidates with a teammate and ask two questions: what do you notice first, and what feels off? The answers are usually more useful than another ten regenerations.

If you have a step you always use that is missing here, leave a comment on this document – I will fold good suggestions into the next version. And if you want a free place to practise the checklist, Yeri is at https://yeri.ai.