{"course":{"slug":"ai-video","title":"Video with AI","url":"https://addaly.com/learn/ai-video","level":"intermediate","track":"create"},"lesson":{"slug":"lens-and-focus-language","title":"Focal length, depth of field, and the words that carry them","url":"https://addaly.com/learn/ai-video/lens-and-focus-language","position":23,"of":66,"minutes":9,"answer":"Focal length changes the relationship between foreground and background rather than just the field of view.","keyIdea":"Focal length changes the relationship between foreground and background rather than just the field of view, so naming a lens steers the geometry of the frame in a way that no amount of describing the subject can."},"body":{"format":"markdown","text":"## What a lens actually changes\n\nMost people think a wide lens shows more and a long lens shows less. That is true and it is the least interesting part.\n\nThe real effect is on **relative size between near and far**. A wide lens used close to a subject makes the subject large and the background tiny and distant: space feels stretched. A long lens used far from the same subject makes the background loom up behind them, compressed and close.\n\nThis is why a 24mm portrait looks like a confrontation and a 135mm portrait looks intimate, even framed identically. It is not about sharpness or quality. It is about the geometry of the space.\n\n## The words that steer\n\nCaptions named lenses constantly, in two forms. Both work:\n\n- **Numerically**: \"shot on a 35mm lens\", \"85mm portrait lens\", \"wide-angle 18mm\", \"telephoto 200mm\".\n- **Descriptively**: \"wide-angle with visible distortion at the edges\", \"compressed telephoto background\", \"shallow depth of field, background falls out of focus\".\n\nUse both together for the strongest effect. The number pins the family; the description pins what you actually want from it.\n\nA rough map worth carrying:\n\n- **14-24mm.** Very wide. Edge distortion, dramatic space, everything sharp. Interiors, landscapes, an aggressive close-up.\n- **35mm.** The documentary lens. Close to how a room feels to be in.\n- **50mm.** Neutral. Roughly human perspective.\n- **85mm.** The portrait lens. Flattering, background softly separated.\n- **135mm and up.** Compressed. Background looms. Surveillance, isolation, a face picked out of a crowd.\n\n## Depth of field and why it helps more than it should\n\nShallow depth of field (a sharp subject against a soft background) does three useful things at once in generated video.\n\n1. **It directs attention**, which is its normal job.\n2. **It hides background churn.** The most common visible failure in generated clips is the background quietly reorganising itself. A background that is a soft wash of colour cannot visibly reorganise. This is an enormous practical benefit and it is free.\n3. **It hides background faces.** A crowd rendered as coloured shapes has no identities to drift.\n\nSo the honest advice is to use it more than a photographer would, for reasons a photographer would not recognise. \"Shallow depth of field, background out of focus\" is one of the highest-value phrases in generative video.\n\nThe cost is that you lose the location. If the shot has to establish where you are, it needs depth, and you will have to accept the churn or shorten the shot.\n\n## Rack focus\n\nPulling focus from foreground to background mid-shot is a real, nameable move (\"rack focus from the foreground bottle to the woman behind it\") and models do attempt it.\n\nWhat you usually get is a blur that ramps globally rather than a true focal plane travelling through the scene. It reads as a defocus and then a refocus, which is close enough for many shots and obviously wrong for others.\n\nThe better route is to do it yourself. Generate the shot with everything acceptably sharp, then apply a keyframed blur in the editor with a mask. Resolve's free tier does this cleanly, and it is deterministic, which the generated version is not.\n\n## Lens artefacts, and the honesty of asking for them\n\nFlares, chromatic aberration, vignetting, barrel distortion, anamorphic streaks. Captions named these, so they work as instructions.\n\nThey are also a well-known shortcut to making generated material read as photographed, because real lenses do these things and clean renders do not. Used lightly, that is legitimate craft: matching the imperfections of the medium you are pretending to be, exactly as a colourist adds grain.\n\nUsed heavily, it becomes the visual equivalent of shouting, and it is one of the reliable markers of amateur generative work. An anamorphic flare in every shot is not cinematic; it is a filter.\n\nIf you want the artefact, a better route is often to add it in the edit, where you control the amount and can remove it at the client's request.\n\n## Mixing lenses across a sequence\n\nA real production picks a small lens set and stays inside it. Doing the same in generated work pays twice: it looks deliberate, and it reduces one more axis on which two shots can fail to match.\n\nPick two focal lengths for a project (say 35mm for wides and 85mm for everything closer) and put them in the shot bible. Then the lens is no longer a decision you make thirty times.\n\n## The free path\n\nAll of this is words, plus a blur node in a free editor. Nothing here requires a subscription. The one thing worth spending a free daily allowance on is a lens test: one still, generated with 24mm, 50mm and 135mm in the instruction, to see whether your model distinguishes them. Some do it well; some ignore focal length entirely and respond only to the depth-of-field description. Knowing which you have saves you writing words that do nothing.\n\n---\n\nToday: add \"shallow depth of field, background softly out of focus\" to one shot that has been churning in the background, and compare it against the original."},"check":{"question":"Adding \"shallow depth of field, background out of focus\" to a shot reliably improves how finished it looks, beyond the usual aesthetic reasons. What is the mechanical benefit?","options":[{"text":"Defocused regions compress to fewer latent tokens, so more of the model's capacity is available for the sharp subject.","correct":false,"tempting":"Invents an adaptive token budget; the latent grid is uniform regardless of what is in focus."},{"text":"A background rendered as a soft wash has no visible structure to reorganise, so the most common generated-video failure becomes invisible.","correct":true},{"text":"Blur suppresses high-frequency detail that the autoencoder would otherwise discard, so less information is lost at the encode stage.","correct":false,"tempting":"Reverses the causality: the encoder discards that detail anyway, and nothing is preserved by blurring it first."},{"text":"Shallow depth of field appears more often in the aesthetically filtered training corpus, so prompts requesting it land in a denser, better-learned region of the model's distribution.","correct":false,"tempting":"Probably true and not the point; the benefit is visible on any model regardless of how its corpus was filtered."}]},"prev":{"slug":"shot-size-and-angle","title":"The two words that decide most of the frame","url":"https://addaly.com/learn/ai-video/shot-size-and-angle"},"next":{"slug":"blocking-and-screen-direction","title":"Where people stand, and which way they face","url":"https://addaly.com/learn/ai-video/blocking-and-screen-direction"},"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) 'Focal length, depth of field, and the words that carry them', in Video with AI. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-video/lens-and-focus-language (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":"Focal length, depth of field, and the words that carry them","courseTitle":"Video with AI","module":{"index":3,"slug":"directing-the-shot","title":"Directing, not describing"},"answer":null,"keyIdea":"Focal length changes the relationship between foreground and background rather than just the field of view, so naming a lens steers the geometry of the frame in a way that no amount of describing the subject can.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"what-a-lens-actually-changes","runs":[{"text":"What a lens actually changes"}]},{"t":"p","runs":[{"text":"Most people think a wide lens shows more and a long lens shows less. That is true and it is the least interesting part."}]},{"t":"p","runs":[{"text":"The real effect is on "},{"text":"relative size between near and far","b":true},{"text":". A wide lens used close to a subject makes the subject large and the background tiny and distant: space feels stretched. A long lens used far from the same subject makes the background loom up behind them, compressed and close."}]},{"t":"p","runs":[{"text":"This is why a 24mm portrait looks like a confrontation and a 135mm portrait looks intimate, even framed identically. It is not about sharpness or quality. It is about the geometry of the space."}]},{"t":"h2","id":"the-words-that-steer","runs":[{"text":"The words that steer"}]},{"t":"p","runs":[{"text":"Captions named lenses constantly, in two forms. Both work:"}]},{"t":"ul","items":[[{"text":"Numerically","b":true},{"text":": “shot on a 35mm lens”, “85mm portrait lens”, “wide-angle 18mm”, “telephoto 200mm”."}],[{"text":"Descriptively","b":true},{"text":": “wide-angle with visible distortion at the edges”, “compressed telephoto background”, “shallow depth of field, background falls out of focus”."}]]},{"t":"p","runs":[{"text":"Use both together for the strongest effect. The number pins the family; the description pins what you actually want from it."}]},{"t":"p","runs":[{"text":"A rough map worth carrying:"}]},{"t":"ul","items":[[{"text":"14-24mm.","b":true},{"text":" Very wide. Edge distortion, dramatic space, everything sharp. Interiors, landscapes, an aggressive close-up."}],[{"text":"35mm.","b":true},{"text":" The documentary lens. Close to how a room feels to be in."}],[{"text":"50mm.","b":true},{"text":" Neutral. Roughly human perspective."}],[{"text":"85mm.","b":true},{"text":" The portrait lens. Flattering, background softly separated."}],[{"text":"135mm and up.","b":true},{"text":" Compressed. Background looms. Surveillance, isolation, a face picked out of a crowd."}]]},{"t":"h2","id":"depth-of-field-and-why-it-helps-more-than-it-should","runs":[{"text":"Depth of field and why it helps more than it should"}]},{"t":"p","runs":[{"text":"Shallow depth of field (a sharp subject against a soft background) does three useful things at once in generated video."}]},{"t":"ol","start":1,"items":[[{"text":"It directs attention","b":true},{"text":", which is its normal job."}],[{"text":"It hides background churn.","b":true},{"text":" The most common visible failure in generated clips is the background quietly reorganising itself. A background that is a soft wash of colour cannot visibly reorganise. This is an enormous practical benefit and it is free."}],[{"text":"It hides background faces.","b":true},{"text":" A crowd rendered as coloured shapes has no identities to drift."}]]},{"t":"p","runs":[{"text":"So the honest advice is to use it more than a photographer would, for reasons a photographer would not recognise. “Shallow depth of field, background out of focus” is one of the highest-value phrases in generative video."}]},{"t":"p","runs":[{"text":"The cost is that you lose the location. If the shot has to establish where you are, it needs depth, and you will have to accept the churn or shorten the shot."}]},{"t":"h2","id":"rack-focus","runs":[{"text":"Rack focus"}]},{"t":"p","runs":[{"text":"Pulling focus from foreground to background mid-shot is a real, nameable move (“rack focus from the foreground bottle to the woman behind it”) and models do attempt it."}]},{"t":"p","runs":[{"text":"What you usually get is a blur that ramps globally rather than a true focal plane travelling through the scene. It reads as a defocus and then a refocus, which is close enough for many shots and obviously wrong for others."}]},{"t":"p","runs":[{"text":"The better route is to do it yourself. Generate the shot with everything acceptably sharp, then apply a keyframed blur in the editor with a mask. Resolve’s free tier does this cleanly, and it is deterministic, which the generated version is not."}]},{"t":"h2","id":"lens-artefacts-and-the-honesty-of-asking-for-them","runs":[{"text":"Lens artefacts, and the honesty of asking for them"}]},{"t":"p","runs":[{"text":"Flares, chromatic aberration, vignetting, barrel distortion, anamorphic streaks. Captions named these, so they work as instructions."}]},{"t":"p","runs":[{"text":"They are also a well-known shortcut to making generated material read as photographed, because real lenses do these things and clean renders do not. Used lightly, that is legitimate craft: matching the imperfections of the medium you are pretending to be, exactly as a colourist adds grain."}]},{"t":"p","runs":[{"text":"Used heavily, it becomes the visual equivalent of shouting, and it is one of the reliable markers of amateur generative work. An anamorphic flare in every shot is not cinematic; it is a filter."}]},{"t":"p","runs":[{"text":"If you want the artefact, a better route is often to add it in the edit, where you control the amount and can remove it at the client’s request."}]},{"t":"h2","id":"mixing-lenses-across-a-sequence","runs":[{"text":"Mixing lenses across a sequence"}]},{"t":"p","runs":[{"text":"A real production picks a small lens set and stays inside it. Doing the same in generated work pays twice: it looks deliberate, and it reduces one more axis on which two shots can fail to match."}]},{"t":"p","runs":[{"text":"Pick two focal lengths for a project (say 35mm for wides and 85mm for everything closer) and put them in the shot bible. Then the lens is no longer a decision you make thirty times."}]},{"t":"h2","id":"the-free-path","runs":[{"text":"The free path"}],"role":"free"},{"t":"p","runs":[{"text":"All of this is words, plus a blur node in a free editor. Nothing here requires a subscription. The one thing worth spending a free daily allowance on is a lens test: one still, generated with 24mm, 50mm and 135mm in the instruction, to see whether your model distinguishes them. Some do it well; some ignore focal length entirely and respond only to the depth-of-field description. Knowing which you have saves you writing words that do nothing."}]},{"t":"rule"},{"t":"p","runs":[{"text":"Today: add “shallow depth of field, background softly out of focus” to one shot that has been churning in the background, and compare it against the original."}]}],"check":{"question":[{"text":"Adding “shallow depth of field, background out of focus” to a shot reliably improves how finished it looks, beyond the usual aesthetic reasons. What is the mechanical benefit?"}],"options":[{"text":[{"text":"Defocused regions compress to fewer latent tokens, so more of the model’s capacity is available for the sharp subject."}],"correct":false,"tempting":[{"text":"Invents an adaptive token budget; the latent grid is uniform regardless of what is in focus."}]},{"text":[{"text":"A background rendered as a soft wash has no visible structure to reorganise, so the most common generated-video failure becomes invisible."}],"correct":true,"tempting":null},{"text":[{"text":"Blur suppresses high-frequency detail that the autoencoder would otherwise discard, so less information is lost at the encode stage."}],"correct":false,"tempting":[{"text":"Reverses the causality: the encoder discards that detail anyway, and nothing is preserved by blurring it first."}]},{"text":[{"text":"Shallow depth of field appears more often in the aesthetically filtered training corpus, so prompts requesting it land in a denser, better-learned region of the model’s distribution."}],"correct":false,"tempting":[{"text":"Probably true and not the point; the benefit is visible on any model regardless of how its corpus was filtered."}]}],"order":[2,0,3,1],"why":null},"prev":{"slug":"shot-size-and-angle","title":"The two words that decide most of the frame","minutes":9},"next":{"slug":"blocking-and-screen-direction","title":"Where people stand, and which way they face","minutes":9}}}