{"course":{"slug":"making-things-with-ai","title":"Making Things with AI","url":"https://addaly.com/learn/making-things-with-ai","level":"beginner","track":"create"},"lesson":{"slug":"video-shots-not-films","title":"Video, and what it still cannot do","url":"https://addaly.com/learn/making-things-with-ai/video-shots-not-films","position":38,"of":84,"minutes":8,"answer":"Video models produce shots, not films; plan the edit and budget for the takes you throw away.","keyIdea":"Video models produce shots, not films; plan the edit and budget for the takes you throw away."},"body":{"format":"markdown","text":"## The extra dimension costs more than you think\n\nAn image model has to make one frame plausible. A video model has to make 150 frames plausible *and* mutually consistent. It cannot make them one at a time, because frame 80 has to know what frame 12 did. So it compresses time as well as space (a video autoencoder might squash eight times in each spatial direction and four times in time) and a transformer attends across that whole compressed block at once.\n\nAttention cost grows steeply with the size of the block. That is the real reason clips are short. Five to ten seconds is where quality, price and attention length currently meet. It is a physics-of-the-method limit, not an arbitrary product decision.\n\n## What works now\n\nSingle shots up to about ten seconds, with a clear subject and a simple camera move, look good. Image-to-video (where you generate or photograph a still first, then animate it) is far more controllable than text-to-video, and it is what most working people use. Several models now produce synchronised dialogue and ambience along with the picture, which was science fiction two years earlier.\n\n## What still fails\n\n**Physics at the point of contact.** Anything touching anything else. Hands gripping, feet meeting ground, liquid pouring, cloth folding, a knife going through a tomato. The model learned what these look like, not what they are, and contact is where the difference surfaces.\n\n**Object permanence.** When something leaves frame and comes back, it comes back subtly wrong. Crowds are the worst case; background faces churn continuously.\n\n**Identity across shots.** Your character in shot three is a cousin of your character in shot one. Locked keyframes and reference features help. They do not solve it.\n\n**Timed action.** \"She looks up on the fourth beat\" is not currently something you can ask for. You get an approximation and cut around it.\n\n**Text.** Signage in video is roughly where image models were in 2022.\n\n**Length.** You do not get a film. You get shots.\n\n## The number that matters is cost per usable second\n\nList prices in 2026 run from about five cents per generated second for budget models to around seventy cents for premium models at 1080p, and some 4K tiers cost more than a dollar (checked October 2026; prices move constantly). The list price misleads, because what you actually spend is cost per second you can use. At a one-in-eight hit rate (normal once a shot has a specific requirement) an eight-second clip listed at two dollars cost sixteen, plus an hour of your attention.\n\nBudget for the takes you throw away. Teams that budget for the keeper run out of money at shot four of thirty.\n\n## A workflow that survives a deadline\n\n1. Write the shot list first, as shots, with durations. Treat it like a real shoot, because it is one.\n2. Make a still for every shot and get the still right. Image tools give you far more control than video tools do.\n3. Animate one shot at a time, several takes each.\n4. Assemble in an ordinary editor. Cuts hide an enormous number of sins. A long continuous move hides none.\n5. Do sound separately unless the model's native audio happens to land.\n\n## Where the honest line sits\n\nFor mood pieces, abstract backgrounds, product motion and B-roll nobody studies frame by frame, this is production-ready and cheap. For anything with a named character doing specific things in a specific order, you are fighting the tool, and a small live shoot or 2D animation is often faster, cheaper and better.\n\nSay that in week one rather than week four. Being the person who knows where the tool stops is more valuable than being the person who can drive it."},"check":{"question":"A team plans a three-minute explainer as one continuous camera move through a factory, generated as a single piece. What is the strongest reason to change the plan?","options":[{"text":"No model accepts a prompt long enough to describe three minutes of action in adequate detail.","correct":false,"tempting":"Locates the limit in how much instruction the model can take in, rather than in what it can hold together."},{"text":"Three minutes of generation costs far more than thirty six-second clips would.","correct":false,"tempting":"Assumes the binding constraint is price, when roughly the same number of frames gets generated either way."},{"text":"Consistency degrades over duration, because the model has no persistent representation of the objects it must keep identical from start to finish.","correct":true},{"text":"A single unbroken shot removes the cuts an editor would use to hide errors, so any mistake stays on screen.","correct":false,"tempting":"A true point about editing craft that skips the prior question of why the footage would be wrong at all."}]},"prev":{"slug":"the-free-toolchain","title":"What runs on the machine you have","url":"https://addaly.com/learn/making-things-with-ai/the-free-toolchain"},"next":{"slug":"frames-are-expensive","title":"Why a second of video costs what it does","url":"https://addaly.com/learn/making-things-with-ai/frames-are-expensive"},"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) 'Video, and what it still cannot do', in Making Things with AI. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/making-things-with-ai/video-shots-not-films (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":"Video, and what it still cannot do","courseTitle":"Making Things with AI","module":{"index":5,"slug":"time-and-space","title":"Time and space: video and three dimensions"},"answer":null,"keyIdea":"Video models produce shots, not films; plan the edit and budget for the takes you throw away.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"the-extra-dimension-costs-more-than-you-think","runs":[{"text":"The extra dimension costs more than you think"}]},{"t":"p","runs":[{"text":"An image model has to make one frame plausible. A video model has to make 150 frames plausible "},{"text":"and","i":true},{"text":" mutually consistent. It cannot make them one at a time, because frame 80 has to know what frame 12 did. So it compresses time as well as space (a video autoencoder might squash eight times in each spatial direction and four times in time) and a transformer attends across that whole compressed block at once."}]},{"t":"p","runs":[{"text":"Attention cost grows steeply with the size of the block. That is the real reason clips are short. Five to ten seconds is where quality, price and attention length currently meet. It is a physics-of-the-method limit, not an arbitrary product decision."}]},{"t":"h2","id":"what-works-now","runs":[{"text":"What works now"}]},{"t":"p","runs":[{"text":"Single shots up to about ten seconds, with a clear subject and a simple camera move, look good. Image-to-video (where you generate or photograph a still first, then animate it) is far more controllable than text-to-video, and it is what most working people use. Several models now produce synchronised dialogue and ambience along with the picture, which was science fiction two years earlier."}]},{"t":"h2","id":"what-still-fails","runs":[{"text":"What still fails"}]},{"t":"p","runs":[{"text":"Physics at the point of contact.","b":true},{"text":" Anything touching anything else. Hands gripping, feet meeting ground, liquid pouring, cloth folding, a knife going through a tomato. The model learned what these look like, not what they are, and contact is where the difference surfaces."}]},{"t":"p","runs":[{"text":"Object permanence.","b":true},{"text":" When something leaves frame and comes back, it comes back subtly wrong. Crowds are the worst case; background faces churn continuously."}]},{"t":"p","runs":[{"text":"Identity across shots.","b":true},{"text":" Your character in shot three is a cousin of your character in shot one. Locked keyframes and reference features help. They do not solve it."}]},{"t":"p","runs":[{"text":"Timed action.","b":true},{"text":" “She looks up on the fourth beat” is not currently something you can ask for. You get an approximation and cut around it."}]},{"t":"p","runs":[{"text":"Text.","b":true},{"text":" Signage in video is roughly where image models were in 2022."}]},{"t":"p","runs":[{"text":"Length.","b":true},{"text":" You do not get a film. You get shots."}]},{"t":"h2","id":"the-number-that-matters-is-cost-per-usable-second","runs":[{"text":"The number that matters is cost per usable second"}]},{"t":"p","runs":[{"text":"List prices in 2026 run from about five cents per generated second for budget models to around seventy cents for premium models at 1080p, and some 4K tiers cost more than a dollar (checked October 2026; prices move constantly). The list price misleads, because what you actually spend is cost per second you can use. At a one-in-eight hit rate (normal once a shot has a specific requirement) an eight-second clip listed at two dollars cost sixteen, plus an hour of your attention."}]},{"t":"p","runs":[{"text":"Budget for the takes you throw away. Teams that budget for the keeper run out of money at shot four of thirty."}]},{"t":"h2","id":"a-workflow-that-survives-a-deadline","runs":[{"text":"A workflow that survives a deadline"}]},{"t":"ol","start":1,"items":[[{"text":"Write the shot list first, as shots, with durations. Treat it like a real shoot, because it is one."}],[{"text":"Make a still for every shot and get the still right. Image tools give you far more control than video tools do."}],[{"text":"Animate one shot at a time, several takes each."}],[{"text":"Assemble in an ordinary editor. Cuts hide an enormous number of sins. A long continuous move hides none."}],[{"text":"Do sound separately unless the model’s native audio happens to land."}]]},{"t":"h2","id":"where-the-honest-line-sits","runs":[{"text":"Where the honest line sits"}]},{"t":"p","runs":[{"text":"For mood pieces, abstract backgrounds, product motion and B-roll nobody studies frame by frame, this is production-ready and cheap. For anything with a named character doing specific things in a specific order, you are fighting the tool, and a small live shoot or 2D animation is often faster, cheaper and better."}]},{"t":"p","runs":[{"text":"Say that in week one rather than week four. Being the person who knows where the tool stops is more valuable than being the person who can drive it."}]}],"check":{"question":[{"text":"A team plans a three-minute explainer as one continuous camera move through a factory, generated as a single piece. What is the strongest reason to change the plan?"}],"options":[{"text":[{"text":"No model accepts a prompt long enough to describe three minutes of action in adequate detail."}],"correct":false,"tempting":[{"text":"Locates the limit in how much instruction the model can take in, rather than in what it can hold together."}]},{"text":[{"text":"Three minutes of generation costs far more than thirty six-second clips would."}],"correct":false,"tempting":[{"text":"Assumes the binding constraint is price, when roughly the same number of frames gets generated either way."}]},{"text":[{"text":"Consistency degrades over duration, because the model has no persistent representation of the objects it must keep identical from start to finish."}],"correct":true,"tempting":null},{"text":[{"text":"A single unbroken shot removes the cuts an editor would use to hide errors, so any mistake stays on screen."}],"correct":false,"tempting":[{"text":"A true point about editing craft that skips the prior question of why the footage would be wrong at all."}]}],"order":[1,2,3,0],"why":null},"prev":{"slug":"the-free-toolchain","title":"What runs on the machine you have","minutes":8},"next":{"slug":"frames-are-expensive","title":"Why a second of video costs what it does","minutes":8}}}