{"course":{"slug":"ai-video","title":"Video with AI","url":"https://addaly.com/learn/ai-video","level":"intermediate","track":"create"},"lesson":{"slug":"what-it-cannot-do","title":"Six things it cannot do, and the reason for each","url":"https://addaly.com/learn/ai-video/what-it-cannot-do","position":39,"of":66,"minutes":10,"answer":"Each failure has a mechanism (no simulation, no symbols, no persistent objects, no reference for your specific thing) and the mechanism tells you whether to wait for a better model or change the shot.","keyIdea":"Each failure has a mechanism (no simulation, no symbols, no persistent objects, no reference for your specific thing) and the mechanism tells you whether to wait for a better model or change the shot."},"body":{"format":"markdown","text":"## Hands doing precise tasks\n\nThreading a needle, tying a lace, typing, playing a guitar, counting notes. The mechanism is contact and occlusion. The model learned the statistics of what hands look like; it did not learn the constraint that a finger cannot pass through a string. Contact events are exactly where a plausible-looking frame becomes an impossible one, and hands make contact constantly, at speed, while occluding each other and the object.\n\nThis is not a resolution problem and more sampling steps will not fix it.\n\nWork around it: cut away at the contact. Hand approaches, cut, result. A century of film grammar exists for precisely this reason, because real crews could not always film the contact either.\n\n## Readable text\n\nSigns, packaging, screens, anything with letters in the frame. Text is a symbolic system with a small alphabet and strict rules; the model has pixel statistics. Image models improved on this enormously. Video models are behind, and there is an extra reason: a letterform must stay identical across a hundred and fifty frames, and one that shifts by two pixels a frame reads as melting even when each individual frame is fine.\n\nWork around it: composite the text in afterwards, in the editor, as an actual title with an actual font. Or put it in the first frame and keep the camera nearly still, which sometimes survives and usually does not.\n\n## Physical cause and effect\n\nA glass that falls and shatters into the right number of pieces. A ball that bounces lower each time. Cloth that settles and stays settled. Fire that reacts when the wind changes. The model interpolates plausible-looking frames; it does not simulate. There is no state, no mass, no conservation of anything.\n\nIt gets short, common, heavily represented physics roughly right (a person walking, water running downhill, hair moving in a breeze) because those patterns are everywhere in the training data. Anything unusual it gets wrong, confidently.\n\nWork around it: single events, short, and never in slow motion. Slow motion is where fake physics is most visible, because you have given the viewer time to check.\n\n## A specific real product\n\nYour client's bottle, with their label, their cap, their exact proportions. The model has no reference for that object. It produces a member of the category with a label that is almost right, and almost right is worse than obviously wrong, because it reads as a counterfeit rather than as a stylisation.\n\nWork around it: this is a compositing job, not a generation job. Generate the environment and the camera move; mask and track the real product in. Or shoot the product on a phone against a plain wall, which is one of the places where thirty seconds of real filming beats an afternoon of generating and everyone can see it did.\n\n## Continuity over minutes\n\nCovered earlier, but the mechanism is worth stating once more because it explains four of the other five items. There is no persistent object memory and no scene model. Every frame is inferred, the only anchor is the conditioning image, and error grows with distance from that anchor. Anything that requires the same object to be the same object several shots later is fighting the architecture, not the settings.\n\n## Action timed to a beat\n\n\"She looks up on the fourth beat\" is not currently something you can ask for. Duration maps loosely onto the request and the timing of an action within a clip is not controllable.\n\nWork around it: generate longer than you need and find the frame in the edit, then slip the clip until the action lands. That is what an editor does with real footage anyway.\n\n## Two things people over-claim in both directions\n\n**Not all of this gets fixed by the next model.** Hands and short-range physics have improved measurably from one generation of models to the next, and will keep improving. A specific real product is different in kind: it is not a capability gap, the information is not available to the model, and no scale of training changes that.\n\n**Not all of it is permanent either.** Text in image models was a standing joke in 2022 and mostly works now. So write down the mechanism rather than the verdict, and re-test this list yourself every few months on your own shot rather than believing anyone, including this lesson.\n\n---\n\nThe most valuable thing you can tell a client or a team is where the tool stops. Being the person who says on day one that a shot is not achievable is worth considerably more than being the person who can drive the tool and discovers it on day fourteen with the deadline in sight."},"check":{"question":"A brand asks for a hero shot of their bottle rotating slowly on a plinth. The generated bottle looks convincing at a glance, but the label is subtly wrong and the cap is the wrong shape. What is the right diagnosis and fix?","options":[{"text":"The model has no reference for that specific object and can only produce a member of the category, so the real bottle should be shot or photographed and composited into the generated environment.","correct":true},{"text":"The prompt did not describe the label and cap precisely enough; a detailed written specification of both will resolve it.","correct":false,"tempting":"Believes a description can substitute for a reference when the target is one specific physical object."},{"text":"The generation resolution is too low for the label to resolve, so a higher-resolution tier will render it correctly.","correct":false,"tempting":"Treats a missing-reference problem as a detail problem that more pixels can recover."},{"text":"Feeding the model a folder of brand photographs will teach it the bottle, after which it will render it accurately.","correct":false,"tempting":"Confuses conditioning on reference images with training, and assumes either would hold an exact geometry through a full rotation."}]},"prev":{"slug":"the-compounding-hit-rate","title":"The arithmetic nobody puts in the demo reel","url":"https://addaly.com/learn/ai-video/the-compounding-hit-rate"},"next":{"slug":"faces-voices-consent","title":"Lip-sync is easy; the permission is not","url":"https://addaly.com/learn/ai-video/faces-voices-consent"},"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) 'Six things it cannot do, and the reason for each', in Video with AI. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-video/what-it-cannot-do (Accessed: 31 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Six things it cannot do, and the reason for each","courseTitle":"Video with AI","module":{"index":4,"slug":"continuity-and-character","title":"Making separate shots belong to each other"},"answer":null,"keyIdea":"Each failure has a mechanism (no simulation, no symbols, no persistent objects, no reference for your specific thing) and the mechanism tells you whether to wait for a better model or change the shot.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"hands-doing-precise-tasks","runs":[{"text":"Hands doing precise tasks"}]},{"t":"p","runs":[{"text":"Threading a needle, tying a lace, typing, playing a guitar, counting notes. The mechanism is contact and occlusion. The model learned the statistics of what hands look like; it did not learn the constraint that a finger cannot pass through a string. Contact events are exactly where a plausible-looking frame becomes an impossible one, and hands make contact constantly, at speed, while occluding each other and the object."}]},{"t":"p","runs":[{"text":"This is not a resolution problem and more sampling steps will not fix it."}]},{"t":"p","runs":[{"text":"Work around it: cut away at the contact. Hand approaches, cut, result. A century of film grammar exists for precisely this reason, because real crews could not always film the contact either."}]},{"t":"h2","id":"readable-text","runs":[{"text":"Readable text"}]},{"t":"p","runs":[{"text":"Signs, packaging, screens, anything with letters in the frame. Text is a symbolic system with a small alphabet and strict rules; the model has pixel statistics. Image models improved on this enormously. Video models are behind, and there is an extra reason: a letterform must stay identical across a hundred and fifty frames, and one that shifts by two pixels a frame reads as melting even when each individual frame is fine."}]},{"t":"p","runs":[{"text":"Work around it: composite the text in afterwards, in the editor, as an actual title with an actual font. Or put it in the first frame and keep the camera nearly still, which sometimes survives and usually does not."}]},{"t":"h2","id":"physical-cause-and-effect","runs":[{"text":"Physical cause and effect"}]},{"t":"p","runs":[{"text":"A glass that falls and shatters into the right number of pieces. A ball that bounces lower each time. Cloth that settles and stays settled. Fire that reacts when the wind changes. The model interpolates plausible-looking frames; it does not simulate. There is no state, no mass, no conservation of anything."}]},{"t":"p","runs":[{"text":"It gets short, common, heavily represented physics roughly right (a person walking, water running downhill, hair moving in a breeze) because those patterns are everywhere in the training data. Anything unusual it gets wrong, confidently."}]},{"t":"p","runs":[{"text":"Work around it: single events, short, and never in slow motion. Slow motion is where fake physics is most visible, because you have given the viewer time to check."}]},{"t":"h2","id":"a-specific-real-product","runs":[{"text":"A specific real product"}]},{"t":"p","runs":[{"text":"Your client’s bottle, with their label, their cap, their exact proportions. The model has no reference for that object. It produces a member of the category with a label that is almost right, and almost right is worse than obviously wrong, because it reads as a counterfeit rather than as a stylisation."}]},{"t":"p","runs":[{"text":"Work around it: this is a compositing job, not a generation job. Generate the environment and the camera move; mask and track the real product in. Or shoot the product on a phone against a plain wall, which is one of the places where thirty seconds of real filming beats an afternoon of generating and everyone can see it did."}]},{"t":"h2","id":"continuity-over-minutes","runs":[{"text":"Continuity over minutes"}]},{"t":"p","runs":[{"text":"Covered earlier, but the mechanism is worth stating once more because it explains four of the other five items. There is no persistent object memory and no scene model. Every frame is inferred, the only anchor is the conditioning image, and error grows with distance from that anchor. Anything that requires the same object to be the same object several shots later is fighting the architecture, not the settings."}]},{"t":"h2","id":"action-timed-to-a-beat","runs":[{"text":"Action timed to a beat"}]},{"t":"p","runs":[{"text":"“She looks up on the fourth beat” is not currently something you can ask for. Duration maps loosely onto the request and the timing of an action within a clip is not controllable."}]},{"t":"p","runs":[{"text":"Work around it: generate longer than you need and find the frame in the edit, then slip the clip until the action lands. That is what an editor does with real footage anyway."}]},{"t":"h2","id":"two-things-people-over-claim-in-both-directions","runs":[{"text":"Two things people over-claim in both directions"}]},{"t":"p","runs":[{"text":"Not all of this gets fixed by the next model.","b":true},{"text":" Hands and short-range physics have improved measurably from one generation of models to the next, and will keep improving. A specific real product is different in kind: it is not a capability gap, the information is not available to the model, and no scale of training changes that."}]},{"t":"p","runs":[{"text":"Not all of it is permanent either.","b":true},{"text":" Text in image models was a standing joke in 2022 and mostly works now. So write down the mechanism rather than the verdict, and re-test this list yourself every few months on your own shot rather than believing anyone, including this lesson."}]},{"t":"rule"},{"t":"p","runs":[{"text":"The most valuable thing you can tell a client or a team is where the tool stops. Being the person who says on day one that a shot is not achievable is worth considerably more than being the person who can drive the tool and discovers it on day fourteen with the deadline in sight."}]}],"check":{"question":[{"text":"A brand asks for a hero shot of their bottle rotating slowly on a plinth. The generated bottle looks convincing at a glance, but the label is subtly wrong and the cap is the wrong shape. What is the right diagnosis and fix?"}],"options":[{"text":[{"text":"The model has no reference for that specific object and can only produce a member of the category, so the real bottle should be shot or photographed and composited into the generated environment."}],"correct":true,"tempting":null},{"text":[{"text":"The prompt did not describe the label and cap precisely enough; a detailed written specification of both will resolve it."}],"correct":false,"tempting":[{"text":"Believes a description can substitute for a reference when the target is one specific physical object."}]},{"text":[{"text":"The generation resolution is too low for the label to resolve, so a higher-resolution tier will render it correctly."}],"correct":false,"tempting":[{"text":"Treats a missing-reference problem as a detail problem that more pixels can recover."}]},{"text":[{"text":"Feeding the model a folder of brand photographs will teach it the bottle, after which it will render it accurately."}],"correct":false,"tempting":[{"text":"Confuses conditioning on reference images with training, and assumes either would hold an exact geometry through a full rotation."}]}],"order":[3,2,1,0],"why":null},"prev":{"slug":"the-compounding-hit-rate","title":"The arithmetic nobody puts in the demo reel","minutes":9},"next":{"slug":"faces-voices-consent","title":"Lip-sync is easy; the permission is not","minutes":10}}}