{"course":{"slug":"ai-images","title":"Image Generation, in Practice","url":"https://addaly.com/learn/ai-images","level":"beginner","track":"create"},"lesson":{"slug":"the-qa-pass","title":"Twelve things to check before you send","url":"https://addaly.com/learn/ai-images/the-qa-pass","position":66,"of":69,"minutes":8,"answer":"A written check performed at 100% on the final file catches the specific faults that generated images produce and that the person who made the image has stopped being able to see.","keyIdea":"A written check performed at 100% on the final file catches the specific faults that generated images produce and that the person who made the image has stopped being able to see."},"body":{"format":"markdown","text":"## Why a list rather than looking\n\nYou have been looking at this image for three hours. You cannot see it any more: your eye skips the areas you have already resolved, and the fault you missed at hour one is now invisible to you permanently.\n\nA written list defeats this, because it directs attention rather than relying on it. It is the same reason surgeons and pilots use checklists, and the same reason it feels unnecessary right up until it catches something.\n\n## The twelve\n\nRun these at 100% on the final flattened file, not on the layered working document.\n\n**1. Hands and fingers.** Count them. Every visible hand.\n\n**2. Eyes.** Both looking the same way, matched in size, catchlights consistent with the light source.\n\n**3. Teeth.** Even but not uniform, correct in number, no extra row.\n\n**4. Ears, glasses, jewellery.** Glasses arms that reach the ears, earrings matching in pairs, watches with plausible faces, necklaces that pass behind rather than through.\n\n**5. Text.** Anything legible in frame. Signage, labels, book spines, screens. Either correct or deliberately illegible, and never accidentally spelling something.\n\n**6. Logos and brand marks.** Against the forbidden list. Including background signage, clothing and packaging.\n\n**7. Shadows.** Every one pointing consistently. Every object that should cast one having one.\n\n**8. Reflections.** Mirrors, windows, glass, polished surfaces, water. Reflecting something that is actually in the scene.\n\n**9. Repeating patterns.** Tiles, brickwork, fabric, foliage, crowds. Looking for the repeat, the discontinuity, and the face that has been duplicated in a crowd.\n\n**10. Edges and seams.** Around every composited element, every inpainted region, every outpainted extension.\n\n**11. Anatomy and structure.** Limbs attached correctly, furniture with the right number of legs, walls meeting floors, stairs that go somewhere.\n\n**12. Banding and compression.** Gradients and skies, viewed at 100% on a good screen.\n\n## Then the three views\n\n**At delivery size.** How the audience sees it. Faults invisible here are not faults.\n\n**As a thumbnail, 200 pixels.** Does it still read? Is the subject clear?\n\n**On a phone.** Different gamma, different brightness, often different colour handling. Send it to yourself and look.\n\n## The second pair of eyes\n\nThe single most effective step, and the cheapest. Show it to somebody who has not been working on it, for ten seconds, and ask what they notice.\n\nThey will find the thing you cannot see. Every time. Ten seconds is enough: you want a first impression, not a considered review, because a first impression is what your audience will have.\n\nIf there is nobody to ask, the next best thing is time. Look at it tomorrow morning. Failing that, flip it horizontally, which resets your eye's adaptation and makes compositional and structural faults jump out. This is an old illustrator's trick and it works.\n\n## Recording the pass\n\nFor client work, keep the record:\n\n```text\nspring26-07-final-v03   QA 18 Mar\n  hands ok / eyes ok / teeth n-a / jewellery: bangle fixed v03\n  text: none in frame / logos: none — checked background signage\n  shadows ok / reflections: window ok / patterns: tiles ok\n  edges: jar composite re-feathered / anatomy ok / banding: sky ok\n  delivery size ok / thumbnail ok / phone ok\n  second pair of eyes: R, no notes\n```\n\nTwo minutes. It means that if something does surface later you know whether it was checked, and it makes the pass an actual event rather than an intention.\n\n## What the list cannot do\n\nIt catches faults. It does not tell you whether the image is any good, whether it answers the brief, or whether it will work in its placement. Those are judgements made much earlier, and a technically flawless image that misses the brief has failed completely.\n\nThe list also has a real failure mode worth naming: applied mechanically, it produces images that are correct and lifeless, because everything unusual has been checked out of them. Some asymmetry, some imperfection and some unresolved detail is what makes a photograph look photographed. The list exists to catch things that are *wrong*, not things that are merely irregular, and telling those apart is judgement that no checklist supplies."},"check":{"question":"Why does flipping an image horizontally help at the end of a long session?","options":[{"text":"Because it reveals composition faults that only appear when the visual weight of the frame is reversed against the reading direction.","correct":false,"tempting":"Reading direction does affect how a composition is scanned, and the flip is useful for faults that were always present rather than ones introduced by reversal."},{"text":"Because your eye has adapted to the familiar version, and reversing it breaks that adaptation so structural faults become visible again.","correct":true},{"text":"Because generated images frequently contain a directional bias inherited from training data, which a flip makes apparent.","correct":false,"tempting":"A directional bias in the data would show as a tendency across many images, not as a fault detectable by flipping a single one."},{"text":"Because inpainted and composited regions were worked on in one orientation, so their seams are easier to detect when the image is viewed the other way round.","correct":false,"tempting":"Seams are orientation-independent; what changes is your familiarity with the image, not any property of the joins."}]},"prev":{"slug":"type-over-an-image","title":"Making words readable on a photograph","url":"https://addaly.com/learn/ai-images/type-over-an-image"},"next":{"slug":"export-and-compression","title":"Exporting without wrecking it at the last step","url":"https://addaly.com/learn/ai-images/export-and-compression"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Twelve things to check before you send', in Image Generation, in Practice. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-images/the-qa-pass (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Twelve things to check before you send","courseTitle":"Image Generation, in Practice","module":{"index":7,"slug":"finishing","title":"Finishing and handing over"},"answer":null,"keyIdea":"A written check performed at 100% on the final file catches the specific faults that generated images produce and that the person who made the image has stopped being able to see.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"why-a-list-rather-than-looking","runs":[{"text":"Why a list rather than looking"}]},{"t":"p","runs":[{"text":"You have been looking at this image for three hours. You cannot see it any more: your eye skips the areas you have already resolved, and the fault you missed at hour one is now invisible to you permanently."}]},{"t":"p","runs":[{"text":"A written list defeats this, because it directs attention rather than relying on it. It is the same reason surgeons and pilots use checklists, and the same reason it feels unnecessary right up until it catches something."}]},{"t":"h2","id":"the-twelve","runs":[{"text":"The twelve"}]},{"t":"p","runs":[{"text":"Run these at 100% on the final flattened file, not on the layered working document."}]},{"t":"p","runs":[{"text":"1. Hands and fingers.","b":true},{"text":" Count them. Every visible hand."}]},{"t":"p","runs":[{"text":"2. Eyes.","b":true},{"text":" Both looking the same way, matched in size, catchlights consistent with the light source."}]},{"t":"p","runs":[{"text":"3. Teeth.","b":true},{"text":" Even but not uniform, correct in number, no extra row."}]},{"t":"p","runs":[{"text":"4. Ears, glasses, jewellery.","b":true},{"text":" Glasses arms that reach the ears, earrings matching in pairs, watches with plausible faces, necklaces that pass behind rather than through."}]},{"t":"p","runs":[{"text":"5. Text.","b":true},{"text":" Anything legible in frame. Signage, labels, book spines, screens. Either correct or deliberately illegible, and never accidentally spelling something."}]},{"t":"p","runs":[{"text":"6. Logos and brand marks.","b":true},{"text":" Against the forbidden list. Including background signage, clothing and packaging."}]},{"t":"p","runs":[{"text":"7. Shadows.","b":true},{"text":" Every one pointing consistently. Every object that should cast one having one."}]},{"t":"p","runs":[{"text":"8. Reflections.","b":true},{"text":" Mirrors, windows, glass, polished surfaces, water. Reflecting something that is actually in the scene."}]},{"t":"p","runs":[{"text":"9. Repeating patterns.","b":true},{"text":" Tiles, brickwork, fabric, foliage, crowds. Looking for the repeat, the discontinuity, and the face that has been duplicated in a crowd."}]},{"t":"p","runs":[{"text":"10. Edges and seams.","b":true},{"text":" Around every composited element, every inpainted region, every outpainted extension."}]},{"t":"p","runs":[{"text":"11. Anatomy and structure.","b":true},{"text":" Limbs attached correctly, furniture with the right number of legs, walls meeting floors, stairs that go somewhere."}]},{"t":"p","runs":[{"text":"12. Banding and compression.","b":true},{"text":" Gradients and skies, viewed at 100% on a good screen."}]},{"t":"h2","id":"then-the-three-views","runs":[{"text":"Then the three views"}]},{"t":"p","runs":[{"text":"At delivery size.","b":true},{"text":" How the audience sees it. Faults invisible here are not faults."}]},{"t":"p","runs":[{"text":"As a thumbnail, 200 pixels.","b":true},{"text":" Does it still read? Is the subject clear?"}]},{"t":"p","runs":[{"text":"On a phone.","b":true},{"text":" Different gamma, different brightness, often different colour handling. Send it to yourself and look."}]},{"t":"h2","id":"the-second-pair-of-eyes","runs":[{"text":"The second pair of eyes"}]},{"t":"p","runs":[{"text":"The single most effective step, and the cheapest. Show it to somebody who has not been working on it, for ten seconds, and ask what they notice."}]},{"t":"p","runs":[{"text":"They will find the thing you cannot see. Every time. Ten seconds is enough: you want a first impression, not a considered review, because a first impression is what your audience will have."}]},{"t":"p","runs":[{"text":"If there is nobody to ask, the next best thing is time. Look at it tomorrow morning. Failing that, flip it horizontally, which resets your eye’s adaptation and makes compositional and structural faults jump out. This is an old illustrator’s trick and it works."}]},{"t":"h2","id":"recording-the-pass","runs":[{"text":"Recording the pass"}]},{"t":"p","runs":[{"text":"For client work, keep the record:"}]},{"t":"code","lang":"text","label":"Text","text":"spring26-07-final-v03   QA 18 Mar\n  hands ok / eyes ok / teeth n-a / jewellery: bangle fixed v03\n  text: none in frame / logos: none — checked background signage\n  shadows ok / reflections: window ok / patterns: tiles ok\n  edges: jar composite re-feathered / anatomy ok / banding: sky ok\n  delivery size ok / thumbnail ok / phone ok\n  second pair of eyes: R, no notes"},{"t":"p","runs":[{"text":"Two minutes. It means that if something does surface later you know whether it was checked, and it makes the pass an actual event rather than an intention."}]},{"t":"h2","id":"what-the-list-cannot-do","runs":[{"text":"What the list cannot do"}]},{"t":"p","runs":[{"text":"It catches faults. It does not tell you whether the image is any good, whether it answers the brief, or whether it will work in its placement. Those are judgements made much earlier, and a technically flawless image that misses the brief has failed completely."}]},{"t":"p","runs":[{"text":"The list also has a real failure mode worth naming: applied mechanically, it produces images that are correct and lifeless, because everything unusual has been checked out of them. Some asymmetry, some imperfection and some unresolved detail is what makes a photograph look photographed. The list exists to catch things that are "},{"text":"wrong","i":true},{"text":", not things that are merely irregular, and telling those apart is judgement that no checklist supplies."}]}],"check":{"question":[{"text":"Why does flipping an image horizontally help at the end of a long session?"}],"options":[{"text":[{"text":"Because it reveals composition faults that only appear when the visual weight of the frame is reversed against the reading direction."}],"correct":false,"tempting":[{"text":"Reading direction does affect how a composition is scanned, and the flip is useful for faults that were always present rather than ones introduced by reversal."}]},{"text":[{"text":"Because your eye has adapted to the familiar version, and reversing it breaks that adaptation so structural faults become visible again."}],"correct":true,"tempting":null},{"text":[{"text":"Because generated images frequently contain a directional bias inherited from training data, which a flip makes apparent."}],"correct":false,"tempting":[{"text":"A directional bias in the data would show as a tendency across many images, not as a fault detectable by flipping a single one."}]},{"text":[{"text":"Because inpainted and composited regions were worked on in one orientation, so their seams are easier to detect when the image is viewed the other way round."}],"correct":false,"tempting":[{"text":"Seams are orientation-independent; what changes is your familiarity with the image, not any property of the joins."}]}],"order":[2,0,3,1],"why":null},"prev":{"slug":"type-over-an-image","title":"Making words readable on a photograph","minutes":8},"next":{"slug":"export-and-compression","title":"Exporting without wrecking it at the last step","minutes":8}}}