{"course":{"slug":"ai-images","title":"Image Generation, in Practice","url":"https://addaly.com/learn/ai-images","level":"beginner","track":"create"},"lesson":{"slug":"one-image-or-a-set","title":"One hero or twelve pieces is a different job","url":"https://addaly.com/learn/ai-images/one-image-or-a-set","position":18,"of":69,"minutes":7,"answer":"A single image is a search problem and a set is a consistency problem, and they need different budgets, different methods and a decision made before any generating starts.","keyIdea":"A single image is a search problem and a set is a consistency problem, and they need different budgets, different methods and a decision made before any generating starts."},"body":{"format":"markdown","text":"## Two jobs that look the same on the brief\n\n\"We need an image for the campaign\" and \"we need the campaign images\" differ by one word and by a factor of about four in effort, and not in the direction you would guess.\n\n**One hero image** is a search. You are looking for a frame that is good enough to carry the whole thing, and the method is breadth: many cheap candidates, a ruthless cull, then deep work on one winner. Budget shape: 70% exploration, 30% finishing.\n\n**A set of twelve** is a consistency problem. The individual images do not need to be extraordinary; they need to look like each other. One stunning frame among eleven ordinary ones actively hurts the set, because it makes the rest look like mistakes. Budget shape: 30% establishing the system, 20% generating, 50% making them agree.\n\nThe mistake is running a set like twelve heroes. You get twelve good, unrelated pictures, and then a very long day trying to reconcile them.\n\n## What \"the system\" means\n\nFor a set, the first job is not an image: it is a set of constants you will hold across every frame:\n\n- **Light**: direction, hardness, colour temperature. One choice for all twelve.\n- **Lens**: a focal length and a camera height. Mixing a 24mm wide and an 85mm portrait in one set reads as two different shoots.\n- **Palette**: three or four colours that recur.\n- **Grain and finish**: applied identically at the end, which is the cheapest consistency you can buy.\n- **Crop convention**: how much room around the subject.\n\nFix those, write them down, generate one **anchor image** that embodies all of them, and approve it before making the other eleven. The anchor then serves technically as well as editorially: as a style reference, as a colour-match target, and as the thing you hold up when frame nine drifts.\n\n## The arithmetic that decides the quote\n\nA rough but honest shape for planning, in generated images rather than hours:\n\n- **One hero**: 60 to 120 exploration renders, 6 to 10 full-quality candidates, 20 to 40 inpainting passes on the winner.\n- **Twelve-image set**: 40 renders to find the anchor, then roughly 15 to 25 renders per frame because you are constraining harder, plus repair on each.\n\nThe second is not twelve times the first, but it is comfortably three to four times, and the shape of the work is different: it is more editing and less searching, so it benefits less from a faster GPU and more from a disciplined process.\n\n> Figure: One hero and a set of twelve are different jobs. A two by two grid, rows One hero and Twelve images, columns The problem and Where the time goes: One hero and The problem: A search; One hero and Where the time goes: 70% exploring; Twelve images and The problem: Consistency; Twelve images and Where the time goes: 50% agreeing. A set is not twelve heroes. Run it that way and you get twelve good unrelated pictures and a very long day reconciling them.\n\n## Say which one it is, in writing\n\nThe practical outcome of this lesson is a sentence in the spec: \"One hero image, plus crops\" or \"A set of N images sharing one visual system\". They price differently, they schedule differently, and a client who asked for one and receives the other is right to be unhappy.\n\nWatch for the request that starts as one and becomes the other. \"This is great. Can we get five more like it?\" is the most common scope expansion in this work, and it is a different job. Five more *like it* means building the system retrospectively from an image that was composed without one, which is harder than building it from the start. The revision lesson later in the course gives you language for that conversation; the thing to do now is notice it is coming.\n\n## The honest limit\n\nConsistency across a set has a ceiling that no current method removes. You can hold light, palette, lens and finish. You can hold a character reasonably well with the methods in the consistency block. What you cannot reliably hold is a *specific* object across many angles (the same chair, with the same scuffs, seen from four sides) because nothing in the model is tracking an object identity between generations.\n\nFor those, the honest answers are to photograph it, to model it in Blender, or to design the set so the object is only seen from one angle. Deciding that now, at the brief stage, is what keeps it from becoming a crisis at frame seven.\n\n## The order to make a set in\n\nSequence matters more than people expect, and the productive order is not first-to-last.\n\n1. **The anchor.** The single frame that best embodies the system. Finish it completely, including grade and grain, and get it approved.\n2. **The hardest frame.** Whichever one you privately suspect is impossible: the group shot, the specific object, the one with legible text. Do it second, while there is still time to redesign the concept around it.\n3. **Everything straightforward.** These go quickly once the system is fixed and the hard problem is solved.\n4. **A consistency pass over all twelve together**, on one screen, at the same size. Fix whatever does not belong.\n\nDoing the hard frame second is the part people resist, because it is the least enjoyable. It is also the only order in which discovering a problem is cheap: finding at frame eleven that the concept cannot be delivered means eleven finished frames may need to change.\n\nThe final pass is not optional. Images look consistent one at a time and reveal their disagreements the moment they are side by side, which is exactly how the audience will see them."},"check":{"question":"Why does a twelve-image set need a different method from a single hero image?","options":[{"text":"Because generating twelve images costs roughly twelve times as much compute as one, so the budget has to be allocated far more carefully across the frames.","correct":false,"tempting":"Compute is rarely the binding constraint at this scale, and the harder problem is that constrained frames need more attempts each, not that renders are expensive."},{"text":"Because each image in a set will be seen for a shorter time than a hero image, so individual frames can be finished to a noticeably lower standard.","correct":false,"tempting":"Viewing time varies by placement rather than by set membership, and a visibly weaker frame damages a set more than it would damage a standalone image."},{"text":"Because the frames are judged against each other, so the work shifts from searching for one good frame to enforcing constants across all of them.","correct":true},{"text":"Because a set is more likely to be re-cropped across different placements, which makes crop-safe composition the dominant concern throughout the whole process.","correct":false,"tempting":"Crop safety is a real constraint but an independent one; a single hero reused across placements faces it just as much as a set does."}]},"prev":{"slug":"brand-colour-is-a-constraint","title":"A hex value is not something a diffusion model can hit","url":"https://addaly.com/learn/ai-images/brand-colour-is-a-constraint"},"next":{"slug":"what-to-shoot-instead","title":"The frames where a phone camera wins","url":"https://addaly.com/learn/ai-images/what-to-shoot-instead"},"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) 'One hero or twelve pieces is a different job', in Image Generation, in Practice. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-images/one-image-or-a-set (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":"One hero or twelve pieces is a different job","courseTitle":"Image Generation, in Practice","module":{"index":2,"slug":"the-brief","title":"From a request to a specification you can generate against"},"answer":null,"keyIdea":"A single image is a search problem and a set is a consistency problem, and they need different budgets, different methods and a decision made before any generating starts.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure","figure":{"cols":["The problem","Where the time goes"],"rows":["One hero","Twelve images"],"type":"matrix","cells":[["A search","70% exploring"],["Consistency","50% agreeing"]],"title":"One hero and a set of twelve are different jobs","caption":"A set is not twelve heroes. Run it that way and you get twelve good unrelated pictures and a very long day reconciling them.","cellLabels":[["breadth, then one winner","30% finishing"],["they must match","20% generating"]]},"alt":"Figure: One hero and a set of twelve are different jobs. A two by two grid, rows One hero and Twelve images, columns The problem and Where the time goes: One hero and The problem: A search; One hero and Where the time goes: 70% exploring; Twelve images and The problem: Consistency; Twelve images and Where the time goes: 50% agreeing. A set is not twelve heroes. Run it that way and you get twelve good unrelated pictures and a very long day reconciling them."}},"blocks":[{"t":"h2","id":"two-jobs-that-look-the-same-on-the-brief","runs":[{"text":"Two jobs that look the same on the brief"}]},{"t":"p","runs":[{"text":"“We need an image for the campaign” and “we need the campaign images” differ by one word and by a factor of about four in effort, and not in the direction you would guess."}]},{"t":"p","runs":[{"text":"One hero image","b":true},{"text":" is a search. You are looking for a frame that is good enough to carry the whole thing, and the method is breadth: many cheap candidates, a ruthless cull, then deep work on one winner. Budget shape: 70% exploration, 30% finishing."}]},{"t":"p","runs":[{"text":"A set of twelve","b":true},{"text":" is a consistency problem. The individual images do not need to be extraordinary; they need to look like each other. One stunning frame among eleven ordinary ones actively hurts the set, because it makes the rest look like mistakes. Budget shape: 30% establishing the system, 20% generating, 50% making them agree."}]},{"t":"p","runs":[{"text":"The mistake is running a set like twelve heroes. You get twelve good, unrelated pictures, and then a very long day trying to reconcile them."}]},{"t":"h2","id":"what-the-system-means","runs":[{"text":"What “the system” means"}]},{"t":"p","runs":[{"text":"For a set, the first job is not an image: it is a set of constants you will hold across every frame:"}]},{"t":"ul","items":[[{"text":"Light","b":true},{"text":": direction, hardness, colour temperature. One choice for all twelve."}],[{"text":"Lens","b":true},{"text":": a focal length and a camera height. Mixing a 24mm wide and an 85mm portrait in one set reads as two different shoots."}],[{"text":"Palette","b":true},{"text":": three or four colours that recur."}],[{"text":"Grain and finish","b":true},{"text":": applied identically at the end, which is the cheapest consistency you can buy."}],[{"text":"Crop convention","b":true},{"text":": how much room around the subject."}]]},{"t":"p","runs":[{"text":"Fix those, write them down, generate one "},{"text":"anchor image","b":true},{"text":" that embodies all of them, and approve it before making the other eleven. The anchor then serves technically as well as editorially: as a style reference, as a colour-match target, and as the thing you hold up when frame nine drifts."}]},{"t":"h2","id":"the-arithmetic-that-decides-the-quote","runs":[{"text":"The arithmetic that decides the quote"}]},{"t":"p","runs":[{"text":"A rough but honest shape for planning, in generated images rather than hours:"}]},{"t":"ul","items":[[{"text":"One hero","b":true},{"text":": 60 to 120 exploration renders, 6 to 10 full-quality candidates, 20 to 40 inpainting passes on the winner."}],[{"text":"Twelve-image set","b":true},{"text":": 40 renders to find the anchor, then roughly 15 to 25 renders per frame because you are constraining harder, plus repair on each."}]]},{"t":"p","runs":[{"text":"The second is not twelve times the first, but it is comfortably three to four times, and the shape of the work is different: it is more editing and less searching, so it benefits less from a faster GPU and more from a disciplined process."}]},{"t":"h2","id":"say-which-one-it-is-in-writing","runs":[{"text":"Say which one it is, in writing"}]},{"t":"p","runs":[{"text":"The practical outcome of this lesson is a sentence in the spec: “One hero image, plus crops” or “A set of N images sharing one visual system”. They price differently, they schedule differently, and a client who asked for one and receives the other is right to be unhappy."}]},{"t":"p","runs":[{"text":"Watch for the request that starts as one and becomes the other. “This is great. Can we get five more like it?” is the most common scope expansion in this work, and it is a different job. Five more "},{"text":"like it","i":true},{"text":" means building the system retrospectively from an image that was composed without one, which is harder than building it from the start. The revision lesson later in the course gives you language for that conversation; the thing to do now is notice it is coming."}]},{"t":"h2","id":"the-honest-limit","runs":[{"text":"The honest limit"}],"role":"limit"},{"t":"p","runs":[{"text":"Consistency across a set has a ceiling that no current method removes. You can hold light, palette, lens and finish. You can hold a character reasonably well with the methods in the consistency block. What you cannot reliably hold is a "},{"text":"specific","i":true},{"text":" object across many angles (the same chair, with the same scuffs, seen from four sides) because nothing in the model is tracking an object identity between generations."}]},{"t":"p","runs":[{"text":"For those, the honest answers are to photograph it, to model it in Blender, or to design the set so the object is only seen from one angle. Deciding that now, at the brief stage, is what keeps it from becoming a crisis at frame seven."}]},{"t":"h2","id":"the-order-to-make-a-set-in","runs":[{"text":"The order to make a set in"}]},{"t":"p","runs":[{"text":"Sequence matters more than people expect, and the productive order is not first-to-last."}]},{"t":"ol","start":1,"items":[[{"text":"The anchor.","b":true},{"text":" The single frame that best embodies the system. Finish it completely, including grade and grain, and get it approved."}],[{"text":"The hardest frame.","b":true},{"text":" Whichever one you privately suspect is impossible: the group shot, the specific object, the one with legible text. Do it second, while there is still time to redesign the concept around it."}],[{"text":"Everything straightforward.","b":true},{"text":" These go quickly once the system is fixed and the hard problem is solved."}],[{"text":"A consistency pass over all twelve together","b":true},{"text":", on one screen, at the same size. Fix whatever does not belong."}]]},{"t":"p","runs":[{"text":"Doing the hard frame second is the part people resist, because it is the least enjoyable. It is also the only order in which discovering a problem is cheap: finding at frame eleven that the concept cannot be delivered means eleven finished frames may need to change."}]},{"t":"p","runs":[{"text":"The final pass is not optional. Images look consistent one at a time and reveal their disagreements the moment they are side by side, which is exactly how the audience will see them."}]}],"check":{"question":[{"text":"Why does a twelve-image set need a different method from a single hero image?"}],"options":[{"text":[{"text":"Because generating twelve images costs roughly twelve times as much compute as one, so the budget has to be allocated far more carefully across the frames."}],"correct":false,"tempting":[{"text":"Compute is rarely the binding constraint at this scale, and the harder problem is that constrained frames need more attempts each, not that renders are expensive."}]},{"text":[{"text":"Because each image in a set will be seen for a shorter time than a hero image, so individual frames can be finished to a noticeably lower standard."}],"correct":false,"tempting":[{"text":"Viewing time varies by placement rather than by set membership, and a visibly weaker frame damages a set more than it would damage a standalone image."}]},{"text":[{"text":"Because the frames are judged against each other, so the work shifts from searching for one good frame to enforcing constants across all of them."}],"correct":true,"tempting":null},{"text":[{"text":"Because a set is more likely to be re-cropped across different placements, which makes crop-safe composition the dominant concern throughout the whole process."}],"correct":false,"tempting":[{"text":"Crop safety is a real constraint but an independent one; a single hero reused across placements faces it just as much as a set does."}]}],"order":[3,1,2,0],"why":null},"prev":{"slug":"brand-colour-is-a-constraint","title":"A hex value is not something a diffusion model can hit","minutes":8},"next":{"slug":"what-to-shoot-instead","title":"The frames where a phone camera wins","minutes":7}}}