{"course":{"slug":"ai-video","title":"Video with AI","url":"https://addaly.com/learn/ai-video","level":"intermediate","track":"create"},"lesson":{"slug":"the-edit-is-the-work","title":"The generator makes footage; the editor makes the video","url":"https://addaly.com/learn/ai-video/the-edit-is-the-work","position":49,"of":66,"minutes":10,"answer":"Trim the settle and the drift, unify everything with one grade and one grain pass, and put real sound underneath: that is most of the difference between footage and output.","keyIdea":"Trim the settle and the drift, unify everything with one grade and one grain pass, and put real sound underneath: that is most of the difference between footage and output."},"body":{"format":"markdown","text":"## Nobody ships a raw generation\n\nWhat separates work that looks made from work that looks generated is almost entirely the edit: pace, cut points, sound, colour continuity, and a ruthless discard rate.\n\nExpect shooting ratios. A documentary crew shoots thirty minutes for one that airs. You will generate somewhere between four and ten clips for every one that ships. That is not failure, it is the normal shape of the job, and the sooner you budget for it the sooner you stop taking each bad generation personally.\n\n> Figure: The same thirty generations, before and after an afternoon. Straight out of the generator: Every clip carries its quarter-second settle and its drifting last second; Thirty independent decisions about white balance and contrast; A smooth, denoised surface that reads as synthetic before anything else does; No audio at all, so every cut lands in true digital silence. After trim, grade, grain and sound: Each shot ends one moment before it stopped working; One shot-match pass, then a single grade over the whole timeline; One grain layer, so every clip shares a surface it did not have alone; A continuous ambience bed, unbroken across every cut. None of the right-hand column costs money, and most of it is subtraction. Expect to generate four to ten clips for every one that ships: that is the normal shape of the job rather than a sign that you are doing it badly.\n\n## Conform the frame rate before you import anything\n\nModels output at different frame rates (commonly 24, 25 or 30) often at 720p or 1080p, and sometimes at odd durations like 5.1 seconds. Drop mixed rates onto one timeline and the mismatched clips judder, because frames are being duplicated or dropped to fit.\n\nPick the project frame rate first: 24 for a film feel, 30 for social platforms. Set it before importing. In DaVinci Resolve, changing the project frame rate once media is on the timeline is awkward or blocked depending on version, and re-conforming later is a genuine nuisance.\n\nUpscale at the end, once, on the finished cut, not per clip. Upscaling a 720p generation to 4K adds no information. It adds a sharpening signature, and doing it thirty times gives you thirty slightly different signatures.\n\n## The three moves that make generated clips read as footage\n\n1. **Trim the head and the tail by default.** Drop roughly the first quarter-second and the last second of every clip. You are cutting off the settle and the drift, which is where a clip announces itself.\n2. **Unify colour and texture.** Different generations drift warm or cool independently. One shot-match pass, then a single grade on an adjustment layer across the whole timeline, pulls everything into one world. Then put a light film grain over the entire sequence. Grain is astonishingly effective here, because it gives every clip the same surface texture and it hides the smooth, plastic, denoised quality that reads as synthetic more than any other single property.\n3. **Sound carries the picture.** Real room tone, real footsteps, real traffic under a generated street and the brain accepts what it is seeing. Free sources: Freesound, the BBC sound effects archive under its own terms, and your own phone held out of a window. Clean it in Audacity. This is the cheapest quality upgrade in the entire pipeline and it is the one most people skip.\n\n## Cut on motion, and cut before the failure\n\nA cut placed during movement is far less visible than a cut on a static frame: the eye is already tracking change. And you decide where the failure lands. If the hand goes wrong at six seconds, the shot is five seconds long and nobody will ever know there was a sixth.\n\n## The free toolchain, end to end\n\n- **DaVinci Resolve**, free tier: a professional editor, colourist and audio suite in one application, free permanently, no watermark on output. Honest limits: the free version withholds some effects and noise reduction, has restrictions around certain high-resolution and codec outputs, and it is heavy software that wants a reasonable GPU and 16GB of RAM.\n- **Kdenlive** and **Shotcut**: open source, much lighter, and they run happily on older Windows and Linux machines that Resolve will refuse. For cuts, titles and audio they are not worse. For colour work they are clearly behind.\n- **CapCut** on a phone: the whole assembly is achievable on a phone, and for vertical short-form it is often faster than a desktop. Its free tier and watermarking on certain templates have moved around over time, so check before you deliver something commercial.\n- **Audacity** for audio cleanup on any machine at all.\n\nMore depth on all of this in [video-editing](/learn/video-editing), and on sound design and motion graphics in [motion-and-audio](/learn/motion-and-audio).\n\n## Keep four things per shot\n\nFor every clip that ships, save the prompt, the seed, the first frame and the model name in a plain text file next to the project. Six weeks later a client will ask for one small change to shot nine, and without those four things you are not editing, you are starting over.\n\n---\n\nToday: take three clips you already generated. Trim the head and tail off each, put a single grade and a light grain across all three, and lay any real ambient sound underneath. Compare that against the raw exports."},"check":{"question":"An editor assembles twelve individually good generated clips and the result still reads as a slideshow of AI clips rather than a piece of film. Which single change does the most?","options":[{"text":"Upscale every clip to 4K before assembly so the sequence carries more perceived production value.","correct":false,"tempting":"Assumes resolution is what reads as finish, when upscaling adds a sharpening signature and no information."},{"text":"Put one grade and a light grain across the whole timeline and lay real sound under every shot, so the clips share a texture and a world.","correct":true},{"text":"Regenerate the four weakest clips until each one matches the quality of the best.","correct":false,"tempting":"Treats the problem as per-clip quality when the failure is the mismatch between clips."},{"text":"Generate longer clips so the sequence needs fewer cuts and the joins become less noticeable.","correct":false,"tempting":"Blames the cuts for a discontinuity that lives in colour and texture, and buys more drift in exchange."}]},"prev":{"slug":"models-that-generate-their-own-audio","title":"When the model makes the sound too","url":"https://addaly.com/learn/ai-video/models-that-generate-their-own-audio"},"next":{"slug":"conforming-and-project-setup","title":"Decide the frame rate before you import anything","url":"https://addaly.com/learn/ai-video/conforming-and-project-setup"},"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) 'The generator makes footage; the editor makes the video', in Video with AI. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-video/the-edit-is-the-work (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":"The generator makes footage; the editor makes the video","courseTitle":"Video with AI","module":{"index":6,"slug":"the-edit","title":"From generations to a finished piece"},"answer":null,"keyIdea":"Trim the settle and the drift, unify everything with one grade and one grain pass, and put real sound underneath: that is most of the difference between footage and output.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure 6.1","figure":{"left":{"title":"Straight out of the generator","items":["Every clip carries its quarter-second settle and its drifting last second","Thirty independent decisions about white balance and contrast","A smooth, denoised surface that reads as synthetic before anything else does","No audio at all, so every cut lands in true digital silence"]},"type":"compare","right":{"title":"After trim, grade, grain and sound","items":["Each shot ends one moment before it stopped working","One shot-match pass, then a single grade over the whole timeline","One grain layer, so every clip shares a surface it did not have alone","A continuous ambience bed, unbroken across every cut"]},"title":"The same thirty generations, before and after an afternoon","caption":"None of the right-hand column costs money, and most of it is subtraction. Expect to generate four to ten clips for every one that ships: that is the normal shape of the job rather than a sign that you are doing it badly."},"alt":"Figure: The same thirty generations, before and after an afternoon. Straight out of the generator: Every clip carries its quarter-second settle and its drifting last second; Thirty independent decisions about white balance and contrast; A smooth, denoised surface that reads as synthetic before anything else does; No audio at all, so every cut lands in true digital silence. After trim, grade, grain and sound: Each shot ends one moment before it stopped working; One shot-match pass, then a single grade over the whole timeline; One grain layer, so every clip shares a surface it did not have alone; A continuous ambience bed, unbroken across every cut. None of the right-hand column costs money, and most of it is subtraction. Expect to generate four to ten clips for every one that ships: that is the normal shape of the job rather than a sign that you are doing it badly."}},"blocks":[{"t":"h2","id":"nobody-ships-a-raw-generation","runs":[{"text":"Nobody ships a raw generation"}]},{"t":"p","runs":[{"text":"What separates work that looks made from work that looks generated is almost entirely the edit: pace, cut points, sound, colour continuity, and a ruthless discard rate."}]},{"t":"p","runs":[{"text":"Expect shooting ratios. A documentary crew shoots thirty minutes for one that airs. You will generate somewhere between four and ten clips for every one that ships. That is not failure, it is the normal shape of the job, and the sooner you budget for it the sooner you stop taking each bad generation personally."}]},{"t":"h2","id":"conform-the-frame-rate-before-you-import-anything","runs":[{"text":"Conform the frame rate before you import anything"}]},{"t":"p","runs":[{"text":"Models output at different frame rates (commonly 24, 25 or 30) often at 720p or 1080p, and sometimes at odd durations like 5.1 seconds. Drop mixed rates onto one timeline and the mismatched clips judder, because frames are being duplicated or dropped to fit."}]},{"t":"p","runs":[{"text":"Pick the project frame rate first: 24 for a film feel, 30 for social platforms. Set it before importing. In DaVinci Resolve, changing the project frame rate once media is on the timeline is awkward or blocked depending on version, and re-conforming later is a genuine nuisance."}]},{"t":"p","runs":[{"text":"Upscale at the end, once, on the finished cut, not per clip. Upscaling a 720p generation to 4K adds no information. It adds a sharpening signature, and doing it thirty times gives you thirty slightly different signatures."}]},{"t":"h2","id":"the-three-moves-that-make-generated-clips-read-as-footage","runs":[{"text":"The three moves that make generated clips read as footage"}]},{"t":"ol","start":1,"items":[[{"text":"Trim the head and the tail by default.","b":true},{"text":" Drop roughly the first quarter-second and the last second of every clip. You are cutting off the settle and the drift, which is where a clip announces itself."}],[{"text":"Unify colour and texture.","b":true},{"text":" Different generations drift warm or cool independently. One shot-match pass, then a single grade on an adjustment layer across the whole timeline, pulls everything into one world. Then put a light film grain over the entire sequence. Grain is astonishingly effective here, because it gives every clip the same surface texture and it hides the smooth, plastic, denoised quality that reads as synthetic more than any other single property."}],[{"text":"Sound carries the picture.","b":true},{"text":" Real room tone, real footsteps, real traffic under a generated street and the brain accepts what it is seeing. Free sources: Freesound, the BBC sound effects archive under its own terms, and your own phone held out of a window. Clean it in Audacity. This is the cheapest quality upgrade in the entire pipeline and it is the one most people skip."}]]},{"t":"h2","id":"cut-on-motion-and-cut-before-the-failure","runs":[{"text":"Cut on motion, and cut before the failure"}]},{"t":"p","runs":[{"text":"A cut placed during movement is far less visible than a cut on a static frame: the eye is already tracking change. And you decide where the failure lands. If the hand goes wrong at six seconds, the shot is five seconds long and nobody will ever know there was a sixth."}]},{"t":"h2","id":"the-free-toolchain-end-to-end","runs":[{"text":"The free toolchain, end to end"}]},{"t":"ul","items":[[{"text":"DaVinci Resolve","b":true},{"text":", free tier: a professional editor, colourist and audio suite in one application, free permanently, no watermark on output. Honest limits: the free version withholds some effects and noise reduction, has restrictions around certain high-resolution and codec outputs, and it is heavy software that wants a reasonable GPU and 16GB of RAM."}],[{"text":"Kdenlive","b":true},{"text":" and "},{"text":"Shotcut","b":true},{"text":": open source, much lighter, and they run happily on older Windows and Linux machines that Resolve will refuse. For cuts, titles and audio they are not worse. For colour work they are clearly behind."}],[{"text":"CapCut","b":true},{"text":" on a phone: the whole assembly is achievable on a phone, and for vertical short-form it is often faster than a desktop. Its free tier and watermarking on certain templates have moved around over time, so check before you deliver something commercial."}],[{"text":"Audacity","b":true},{"text":" for audio cleanup on any machine at all."}]]},{"t":"p","runs":[{"text":"More depth on all of this in "},{"text":"video-editing","href":"https://addaly.com/learn/video-editing"},{"text":", and on sound design and motion graphics in "},{"text":"motion-and-audio","href":"https://addaly.com/learn/motion-and-audio"},{"text":"."}]},{"t":"h2","id":"keep-four-things-per-shot","runs":[{"text":"Keep four things per shot"}]},{"t":"p","runs":[{"text":"For every clip that ships, save the prompt, the seed, the first frame and the model name in a plain text file next to the project. Six weeks later a client will ask for one small change to shot nine, and without those four things you are not editing, you are starting over."}]},{"t":"rule"},{"t":"p","runs":[{"text":"Today: take three clips you already generated. Trim the head and tail off each, put a single grade and a light grain across all three, and lay any real ambient sound underneath. Compare that against the raw exports."}]}],"check":{"question":[{"text":"An editor assembles twelve individually good generated clips and the result still reads as a slideshow of AI clips rather than a piece of film. Which single change does the most?"}],"options":[{"text":[{"text":"Upscale every clip to 4K before assembly so the sequence carries more perceived production value."}],"correct":false,"tempting":[{"text":"Assumes resolution is what reads as finish, when upscaling adds a sharpening signature and no information."}]},{"text":[{"text":"Put one grade and a light grain across the whole timeline and lay real sound under every shot, so the clips share a texture and a world."}],"correct":true,"tempting":null},{"text":[{"text":"Regenerate the four weakest clips until each one matches the quality of the best."}],"correct":false,"tempting":[{"text":"Treats the problem as per-clip quality when the failure is the mismatch between clips."}]},{"text":[{"text":"Generate longer clips so the sequence needs fewer cuts and the joins become less noticeable."}],"correct":false,"tempting":[{"text":"Blames the cuts for a discontinuity that lives in colour and texture, and buys more drift in exchange."}]}],"order":[3,2,0,1],"why":null},"prev":{"slug":"models-that-generate-their-own-audio","title":"When the model makes the sound too","minutes":8},"next":{"slug":"conforming-and-project-setup","title":"Decide the frame rate before you import anything","minutes":9}}}