{"course":{"slug":"ai-images","title":"Image Generation, in Practice","url":"https://addaly.com/learn/ai-images","level":"beginner","track":"create"},"lesson":{"slug":"a-visual-system-across-a-set","title":"Five constants that do more than the face","url":"https://addaly.com/learn/ai-images/a-visual-system-across-a-set","position":58,"of":69,"minutes":8,"answer":"Coherence in a set is produced by holding light, lens, palette, crop and finish constant, and these are cheap to enforce and far more visible to a viewer than small drift in a character's features.","keyIdea":"Coherence in a set is produced by holding light, lens, palette, crop and finish constant, and these are cheap to enforce and far more visible to a viewer than small drift in a character's features."},"body":{"format":"markdown","text":"## What a viewer actually perceives\n\nShow somebody twelve images and ask whether they belong together. They are not examining the nose. They are responding to whether the images look like they were made by one person, in one session, with one intent.\n\nThat perception is built from five things, all of which you control completely.\n\n## The five constants\n\n**1. Light.** One direction, one quality, one colour temperature across every frame. Hard sun from the left in all twelve, or soft window light in all twelve. Mixing them is the single strongest signal that a set is not a set.\n\n**2. Lens.** One focal length and one camera height. A 35mm waist-level look throughout. Mixing a 24mm wide with an 85mm portrait reads as two different shoots regardless of anything else.\n\n**3. Palette.** Three or four colours that recur, and one that does not appear anywhere. A restriction is more visible than an inclusion: if nothing in the set is pure saturated red, the set acquires a discipline viewers feel without identifying.\n\n**4. Crop convention.** How much room around the subject, where the horizon sits, whether subjects are centred or offset. Consistency here is what makes a grid of images feel designed.\n\n**5. Finish.** Grain, contrast curve, colour grade, vignette. Applied identically to every frame as the last step. This is the cheapest of the five and possibly the most effective.\n\n> Figure: What a viewer perceives as a set, and what absorbs the effort. What actually holds a set together: One light direction, quality and temperature; One focal length and one camera height; Three or four recurring colours, and one absent; One crop convention, held across every frame; One grade and grain, applied last and identically. What people spend the time on instead: Identity, because it is the interesting problem; A face that holds to the pixel; Twelve individually beautiful frames; Judging each image on its own; Regrading by eye, frame by frame, at the end. A set with a slightly drifting face and a rigorous look reads as consistent. A set with a perfect face and twelve lighting setups does not. If time is short, spend it on the left.\n\n## Enforcing them\n\nWrite them down as text, in the project folder, and reuse the text:\n\n```text\n# system — spring campaign\nLIGHT    hard sun from camera left, late afternoon, long shadows\nLENS     35mm, waist level, slight wide\nPALETTE  terracotta, sage, bone white, deep brown.  No blue.\nCROP     subject in left third, horizon at 2/3, generous headroom\nFINISH   colour negative grain, lifted blacks, warm mids, -10 saturation\n```\n\nSlots five, six and seven of every prompt come straight from this file. The finish is applied as one saved adjustment set (a GIMP script, an exported curve, or a LUT) over every flattened frame.\n\n## The anchor image\n\nPick or build one frame that embodies the system completely, finish it fully, and get it approved. It then does three jobs:\n\n- **Editorially**, it is what the client signed off on.\n- **Technically**, it is a style reference for an adapter and a colour-match target for grading.\n- **Practically**, it is the comparison for every later frame. Does frame nine belong next to the anchor? If not, frame nine changes.\n\n## The final pass that people skip\n\nWhen all twelve exist, put them on one screen at the same size, in the order they will be seen. Do not skip this because they each looked right on their own.\n\nYou will find: one frame warmer than the rest; one with the light from the wrong side; one cropped tighter; one with more contrast. Every set has them, and they are invisible one at a time and obvious together.\n\nFix them globally, not individually. A warm frame is one curves adjustment. A frame with different contrast is one curve. These are two-minute corrections at this stage and would have been twenty-minute regenerations earlier.\n\n## The deliberate exception\n\nA set does not have to be uniform, and a set where every frame is identical in treatment can be monotonous. What it needs is a *system*, which can include variation that is itself systematic: three close-ups among nine wides, one dark frame among eleven light ones, a rhythm rather than a drone.\n\nThe distinction is between variation you chose and variation that happened. A dark frame placed deliberately at the midpoint of a sequence reads as intent. A dark frame because that generation came out dark reads as a mistake.\n\nSo when the final pass finds an outlier, the question is not only \"is this different\" but \"is this different on purpose\". Occasionally the answer is yes and you keep it. Usually the answer is no, and one curves adjustment fixes it.\n\n## Making the finish reusable\n\nThe fifth constant is the one you can automate, and automating it is worth twenty minutes once.\n\nBuild the grade on the anchor image, then save it in a form you can reapply:\n\n- **GIMP**: export the curve from the Curves dialog as a settings file, and save the grain layer as a reusable XCF you paste in.\n- **Krita**: save the adjustment stack as a layer style, or keep a template document.\n- **Any tool**: export a LUT from the anchor and apply it everywhere. Free LUT generators exist, and a LUT is the most portable form: it will still apply in five years in software that does not exist yet.\n\nThen the finish step on every frame is: flatten, apply LUT, add grain layer, done. Two minutes rather than fifteen, and identical by construction rather than by care.\n\n## What to do when the client changes a constant\n\nThey will, usually after three frames are finished. \"Can it be cooler?\"\n\nBecause the constants live in one file and the finish lives in one preset, this is a small change: edit the system file, rebuild the preset, reapply to the finished frames. Twenty minutes for the whole set.\n\nWithout that structure, it is a per-frame regrade done by eye, which takes an afternoon and produces twelve frames that are each individually cooler and no longer identical to each other. The system file is not administration; it is what makes a late change survivable."},"check":{"question":"Why does holding five visual constants matter more to a set's coherence than perfect character consistency?","options":[{"text":"Because character drift is only visible at large sizes, whereas lighting and palette inconsistencies remain visible even in thumbnails.","correct":false,"tempting":"Display size is not the distinction; a mismatched frame stands out at any size and so does a substantially different face."},{"text":"Because viewers judge whether images were made together from light, lens, palette, crop and finish rather than from small facial differences.","correct":true},{"text":"Because the five constants can be applied after generation, which makes them the only aspects of a set that can still be corrected once the frames exist.","correct":false,"tempting":"Three of the five are generation-time decisions, and character drift can also be corrected afterwards by compositing."},{"text":"Because a set is usually viewed as a sequence rather than side by side, and sequential viewing hides facial variation while exposing tonal shifts.","correct":false,"tempting":"Sets are frequently seen side by side in grids, and the argument would reverse if they were not."}]},"prev":{"slug":"product-consistency","title":"A product is harder than a face","url":"https://addaly.com/learn/ai-images/product-consistency"},"next":{"slug":"naming-and-versioning","title":"A file scheme that survives the revision round","url":"https://addaly.com/learn/ai-images/naming-and-versioning"},"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) 'Five constants that do more than the face', in Image Generation, in Practice. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-images/a-visual-system-across-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":"Five constants that do more than the face","courseTitle":"Image Generation, in Practice","module":{"index":6,"slug":"consistency","title":"The same thing, twelve times"},"answer":null,"keyIdea":"Coherence in a set is produced by holding light, lens, palette, crop and finish constant, and these are cheap to enforce and far more visible to a viewer than small drift in a character’s features.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure","figure":{"left":{"title":"What actually holds a set together","items":["One light direction, quality and temperature","One focal length and one camera height","Three or four recurring colours, and one absent","One crop convention, held across every frame","One grade and grain, applied last and identically"]},"type":"compare","right":{"title":"What people spend the time on instead","items":["Identity, because it is the interesting problem","A face that holds to the pixel","Twelve individually beautiful frames","Judging each image on its own","Regrading by eye, frame by frame, at the end"]},"title":"What a viewer perceives as a set, and what absorbs the effort","caption":"A set with a slightly drifting face and a rigorous look reads as consistent. A set with a perfect face and twelve lighting setups does not. If time is short, spend it on the left."},"alt":"Figure: What a viewer perceives as a set, and what absorbs the effort. What actually holds a set together: One light direction, quality and temperature; One focal length and one camera height; Three or four recurring colours, and one absent; One crop convention, held across every frame; One grade and grain, applied last and identically. What people spend the time on instead: Identity, because it is the interesting problem; A face that holds to the pixel; Twelve individually beautiful frames; Judging each image on its own; Regrading by eye, frame by frame, at the end. A set with a slightly drifting face and a rigorous look reads as consistent. A set with a perfect face and twelve lighting setups does not. If time is short, spend it on the left."}},"blocks":[{"t":"h2","id":"what-a-viewer-actually-perceives","runs":[{"text":"What a viewer actually perceives"}]},{"t":"p","runs":[{"text":"Show somebody twelve images and ask whether they belong together. They are not examining the nose. They are responding to whether the images look like they were made by one person, in one session, with one intent."}]},{"t":"p","runs":[{"text":"That perception is built from five things, all of which you control completely."}]},{"t":"h2","id":"the-five-constants","runs":[{"text":"The five constants"}]},{"t":"p","runs":[{"text":"1. Light.","b":true},{"text":" One direction, one quality, one colour temperature across every frame. Hard sun from the left in all twelve, or soft window light in all twelve. Mixing them is the single strongest signal that a set is not a set."}]},{"t":"p","runs":[{"text":"2. Lens.","b":true},{"text":" One focal length and one camera height. A 35mm waist-level look throughout. Mixing a 24mm wide with an 85mm portrait reads as two different shoots regardless of anything else."}]},{"t":"p","runs":[{"text":"3. Palette.","b":true},{"text":" Three or four colours that recur, and one that does not appear anywhere. A restriction is more visible than an inclusion: if nothing in the set is pure saturated red, the set acquires a discipline viewers feel without identifying."}]},{"t":"p","runs":[{"text":"4. Crop convention.","b":true},{"text":" How much room around the subject, where the horizon sits, whether subjects are centred or offset. Consistency here is what makes a grid of images feel designed."}]},{"t":"p","runs":[{"text":"5. Finish.","b":true},{"text":" Grain, contrast curve, colour grade, vignette. Applied identically to every frame as the last step. This is the cheapest of the five and possibly the most effective."}]},{"t":"h2","id":"enforcing-them","runs":[{"text":"Enforcing them"}]},{"t":"p","runs":[{"text":"Write them down as text, in the project folder, and reuse the text:"}]},{"t":"code","lang":"text","label":"Text","text":"# system — spring campaign\nLIGHT    hard sun from camera left, late afternoon, long shadows\nLENS     35mm, waist level, slight wide\nPALETTE  terracotta, sage, bone white, deep brown.  No blue.\nCROP     subject in left third, horizon at 2/3, generous headroom\nFINISH   colour negative grain, lifted blacks, warm mids, -10 saturation"},{"t":"p","runs":[{"text":"Slots five, six and seven of every prompt come straight from this file. The finish is applied as one saved adjustment set (a GIMP script, an exported curve, or a LUT) over every flattened frame."}]},{"t":"h2","id":"the-anchor-image","runs":[{"text":"The anchor image"}]},{"t":"p","runs":[{"text":"Pick or build one frame that embodies the system completely, finish it fully, and get it approved. It then does three jobs:"}]},{"t":"ul","items":[[{"text":"Editorially","b":true},{"text":", it is what the client signed off on."}],[{"text":"Technically","b":true},{"text":", it is a style reference for an adapter and a colour-match target for grading."}],[{"text":"Practically","b":true},{"text":", it is the comparison for every later frame. Does frame nine belong next to the anchor? If not, frame nine changes."}]]},{"t":"h2","id":"the-final-pass-that-people-skip","runs":[{"text":"The final pass that people skip"}]},{"t":"p","runs":[{"text":"When all twelve exist, put them on one screen at the same size, in the order they will be seen. Do not skip this because they each looked right on their own."}]},{"t":"p","runs":[{"text":"You will find: one frame warmer than the rest; one with the light from the wrong side; one cropped tighter; one with more contrast. Every set has them, and they are invisible one at a time and obvious together."}]},{"t":"p","runs":[{"text":"Fix them globally, not individually. A warm frame is one curves adjustment. A frame with different contrast is one curve. These are two-minute corrections at this stage and would have been twenty-minute regenerations earlier."}]},{"t":"h2","id":"the-deliberate-exception","runs":[{"text":"The deliberate exception"}]},{"t":"p","runs":[{"text":"A set does not have to be uniform, and a set where every frame is identical in treatment can be monotonous. What it needs is a "},{"text":"system","i":true},{"text":", which can include variation that is itself systematic: three close-ups among nine wides, one dark frame among eleven light ones, a rhythm rather than a drone."}]},{"t":"p","runs":[{"text":"The distinction is between variation you chose and variation that happened. A dark frame placed deliberately at the midpoint of a sequence reads as intent. A dark frame because that generation came out dark reads as a mistake."}]},{"t":"p","runs":[{"text":"So when the final pass finds an outlier, the question is not only “is this different” but “is this different on purpose”. Occasionally the answer is yes and you keep it. Usually the answer is no, and one curves adjustment fixes it."}]},{"t":"h2","id":"making-the-finish-reusable","runs":[{"text":"Making the finish reusable"}]},{"t":"p","runs":[{"text":"The fifth constant is the one you can automate, and automating it is worth twenty minutes once."}]},{"t":"p","runs":[{"text":"Build the grade on the anchor image, then save it in a form you can reapply:"}]},{"t":"ul","items":[[{"text":"GIMP","b":true},{"text":": export the curve from the Curves dialog as a settings file, and save the grain layer as a reusable XCF you paste in."}],[{"text":"Krita","b":true},{"text":": save the adjustment stack as a layer style, or keep a template document."}],[{"text":"Any tool","b":true},{"text":": export a LUT from the anchor and apply it everywhere. Free LUT generators exist, and a LUT is the most portable form: it will still apply in five years in software that does not exist yet."}]]},{"t":"p","runs":[{"text":"Then the finish step on every frame is: flatten, apply LUT, add grain layer, done. Two minutes rather than fifteen, and identical by construction rather than by care."}]},{"t":"h2","id":"what-to-do-when-the-client-changes-a-constant","runs":[{"text":"What to do when the client changes a constant"}]},{"t":"p","runs":[{"text":"They will, usually after three frames are finished. “Can it be cooler?”"}]},{"t":"p","runs":[{"text":"Because the constants live in one file and the finish lives in one preset, this is a small change: edit the system file, rebuild the preset, reapply to the finished frames. Twenty minutes for the whole set."}]},{"t":"p","runs":[{"text":"Without that structure, it is a per-frame regrade done by eye, which takes an afternoon and produces twelve frames that are each individually cooler and no longer identical to each other. The system file is not administration; it is what makes a late change survivable."}]}],"check":{"question":[{"text":"Why does holding five visual constants matter more to a set’s coherence than perfect character consistency?"}],"options":[{"text":[{"text":"Because character drift is only visible at large sizes, whereas lighting and palette inconsistencies remain visible even in thumbnails."}],"correct":false,"tempting":[{"text":"Display size is not the distinction; a mismatched frame stands out at any size and so does a substantially different face."}]},{"text":[{"text":"Because viewers judge whether images were made together from light, lens, palette, crop and finish rather than from small facial differences."}],"correct":true,"tempting":null},{"text":[{"text":"Because the five constants can be applied after generation, which makes them the only aspects of a set that can still be corrected once the frames exist."}],"correct":false,"tempting":[{"text":"Three of the five are generation-time decisions, and character drift can also be corrected afterwards by compositing."}]},{"text":[{"text":"Because a set is usually viewed as a sequence rather than side by side, and sequential viewing hides facial variation while exposing tonal shifts."}],"correct":false,"tempting":[{"text":"Sets are frequently seen side by side in grids, and the argument would reverse if they were not."}]}],"order":[3,2,0,1],"why":null},"prev":{"slug":"product-consistency","title":"A product is harder than a face","minutes":8},"next":{"slug":"naming-and-versioning","title":"A file scheme that survives the revision round","minutes":7}}}