A designer drops three visual options into the team channel. One person wants “more energy.” Someone else likes the quiet one. A third suggests starting over. By the next afternoon, the folder has nine images instead of three, and nobody is any closer to a decision.
Effort isn’t the problem. The review never had a clear question, a shared vocabulary, or anyone in charge of pulling the comments together. AI generation makes that easy to miss, because clicking “generate” again feels like progress. On a shared AI creative workspace, a new take arrives in minutes, so it’s tempting to answer every vague comment with another file instead of a conversation.
A better workflow flips the order. Decide what’s being reviewed, ask for specific observations, and turn the discussion into a short list of actions. Only then make the next version.
Start With One Question
Before anyone opens a file, write down the decision you need. Maybe it’s a close crop versus a wider scene. Maybe the direction is settled, and you only need someone to check the product details. Those are two different reviews. Run them together, and a comment about a button color can reopen the whole concept.
Add a few lines of context: who the piece is for, where it’ll run, and any fixed requirements. Then say what stage it’s at. An early video test shouldn’t be judged like a finished ad, and a final check shouldn’t quietly turn into a new brainstorm.
Name an Owner Before Opinions Arrive
Pick one person to gather feedback and write up what happens next. Small teams often need a named approver too, someone who can break a tie when two good preferences collide. One person can hold both jobs, as long as everyone knows it. Without that, designers tend to treat the latest comment as the decision, even when it contradicts what the group agreed yesterday.
Give specialists a clear job. The person who knows the product checks how it’s shown. Someone else checks whether the visual works in its placement. They can still flag anything they notice, but every essential check has an owner. A big group won’t reliably cover everything just because it’s big.
Share Fewer Options, Clearly Labeled
Only show options that are genuinely different. If two images make the same point, either explain what separates them or drop the weaker one. Label each file the same way and add a line on what question it explores. Reviewers shouldn’t have to guess why four near-identical pictures are in the folder.
Tools can help here. LookRoll’s team plans, for example, include shared projects, roles, comments and review in the same place the images are generated. Check what your plan actually includes and who has which permissions before you set the process up. Even so, the software won’t write your labels or agree on the question for you.
Ask What People See Before What They’d Change
Good feedback says what the reviewer noticed, why it matters, and where. “The background pulls attention away from the product in the last two seconds” is something a designer can act on. “Make it pop” isn’t. For video, note the moment. For an image, point to the area instead of calling the whole thing wrong.
Keep the observation separate from the fix. A reviewer can spot a real problem, like a subject that blends into the background, and still suggest a fix that causes a new one. Writing down the observation preserves what’s useful and leaves the designer room to solve it their own way. It also stops every suggested technique from getting crammed into one overloaded prompt.
Sort Comments Before You Act on Them
Once the comments are in, split them into required corrections, direction calls, and personal preferences. A wrong product detail has to be fixed. Choosing between two workable compositions is a direction call. Wanting a teal background instead of a grey one is usually a preference, unless the brief says otherwise. If a comment doesn’t fit cleanly into one group, ask the reviewer.
Settle conflicts before generating anything. Say one person wants a tighter crop and another needs room for a headline. Go back to the layout. Often the answer is practical, and no amount of new images will produce it. The owner writes down which priority won, so the designer isn’t left guessing.
Write a Revision Note, Then Generate
Before the next attempt, write a short note covering what’s changing, what must stay the same, and which comments are parked for later. That keeps the new version tied to the discussion. Every new file should answer a question the team agreed on. Busywork doesn’t count.
AI adds a wrinkle here. Generation results vary from take to take, so a “small” change can shift things nobody asked to touch, like a face, a hand, or the lighting. Some edits are hard to isolate, and models differ in how precisely you can steer them. Spelling out what to avoid helps, and it’s one reason why AI ignores instructions when prompts only say what to add. If a change can’t be controlled reliably, tell the team before promising it.
Check the cost too. Many platforms show estimated credits before you submit, and a completed generation usually uses credits even if you end up discarding it. Ten “quick tries” add up.
Judge the Revision Against the Note
Put the new version next to the old one and the revision note. Then ask two narrow questions: did it fix the issue, and did it break anything else? That’s a smaller ask than “which one do you like now?” It protects decisions you’ve already made and still catches surprises.
If the revision didn’t help, write down why. Maybe the source image was weak, the scene was too busy, or the placement requirements were never clear. Another attempt is fine, but it needs a reason. Re-running the same request without discussing the result just fills the folder.
Close It Out
Once the team picks a direction, mark the approved file and note where it’ll be used. Move old drafts out of the delivery folder, but keep whatever working records you need. Do the last check in the real layout, at normal viewing size, not zoomed in on a big monitor.
Before anything goes public, confirm you have the rights to every reference image, face, logo, and piece of music in it. Ownership of AI output is still unsettled in many places, and the questions around who owns AI designs don’t disappear just because the team approved the file.
A good review leaves a short trail: the question, the observations, the agreed changes, the final pick. The next person can follow it without digging through a week of messages. And the next time someone reaches for the generate button, they’ll know what the new version is supposed to fix.
Related: AI-Designed Stand-Up Pouches Look Perfect—But Can You Legally Sell Them?
