{"course":{"slug":"ai-video","title":"Video with AI","url":"https://addaly.com/learn/ai-video","level":"intermediate","track":"create"},"lesson":{"slug":"compositing-the-real-thing","title":"Put the real object in before you animate it","url":"https://addaly.com/learn/ai-video/compositing-the-real-thing","position":15,"of":66,"minutes":10,"answer":"A model has no reference for your client's specific product or person, so the professional split is to generate the environment and the camera move and composite the real thing into the still.","keyIdea":"A model has no reference for your client's specific product or person, so the professional split is to generate the environment and the camera move and composite the real thing into the still, because a nearly-right label reads as a counterfeit rather than a stylisation."},"body":{"format":"markdown","text":"## The problem, stated precisely\n\nAsk a model for \"a bottle of Himalaya face wash on a bathroom shelf\" and you get a bottle. It will be the right shape family, the right colour family, and the label will be almost right, which is worse than obviously wrong, because almost right reads as a fake.\n\nThe reason is not capability, it is information. The model has no representation of that specific bottle. No amount of prompting, steps or guidance retrieves data that is not there. This is the one limit in the whole subject that does not get better with scale.\n\nThe answer is not to try harder. It is to change the division of labour: **the model makes the world, you supply the thing.**\n\n## Two places to composite, and how to choose\n\n**Into the still, before generating.** The object becomes part of the first frame and gets animated with everything else. Cheap, simple, and the object moves naturally with the scene.\n\nUse this when the camera move is small: a slow push, a slight drift, a locked-off frame with ambient motion. The object will soften and may distort slightly as the clip runs, which a short shot and a light move will hide.\n\n**Into the finished clip, in the editor.** Generate the environment and camera move with a stand-in or an empty surface, then track the real object in afterwards with motion tracking.\n\nUse this when the object must stay pin-sharp and exactly itself, when the camera move is large, or when the client will look at it frame by frame. This is more work and it is the only version that survives scrutiny.\n\nThe honest rule: anything a client is paying for a close-up of goes in during the edit, not in the still.\n\n## Making a composite hold up in the still\n\nFive things, in order. Get the first three right and most people never notice the rest.\n\n1. **Perspective.** The object's viewing angle must match the scene's. A bottle photographed straight on cannot sit on a table seen from above. Reshoot or regenerate the background. Do not try to fix this by distorting, because distortion is exactly what the eye catches.\n2. **Light direction.** The key light on the object must come from where the key light in the scene comes from. You decided that direction in the previous lesson; now you shoot the object under it. A window and a piece of white card is enough.\n3. **Scale and contact.** The object needs a shadow where it touches the surface, and that shadow needs the same softness as other shadows in the frame. A composite with no contact shadow floats, and floating is the single most common tell.\n4. **Colour and contrast.** Match the object's black point and colour cast to the plate. Nine times out of ten this is one curves adjustment.\n5. **Grain and edge.** Generated frames are smooth; a real photograph has noise. Add matching grain to the composited object, and soften its cut-out edge by a pixel or two so it does not look laser-cut.\n\n## Shooting the object without a studio\n\nThis is where thirty seconds of real filming beats an afternoon of generating, and everybody can see that it did.\n\nA phone, a plain wall or a sheet of paper, and daylight from a window. Turn the phone's flash off. Put the object a metre from the wall so its own shadow does not fall on the background. Shoot from the angle the composite needs, not the angle that is comfortable. Take twenty frames and pick one.\n\nCut it out in Photopea or GIMP with the pen tool for hard edges and a mask for soft ones. Background removal tools handle most objects in one click now and are worth trying first; they fail on hair, glass and thin structures, which is where you take over by hand.\n\n## Tracking it into a moving clip\n\nWhen the object has to go into the finished footage:\n\n- **DaVinci Resolve's Fusion page**, in the free tier, does planar and point tracking, and it is capable. The workflow is: track a feature in the plate, attach the object to the track, match grade, add grain.\n- **Blender** does the same with its motion tracker and is free and open, with a steeper climb.\n- **Kdenlive** has basic tracking; enough to stick a logo on a slow move, not enough for a hero product shot.\n\nGenerate the plate with a simple, high-contrast trackable feature in it: a table corner, a doorframe, a mark on the wall. A featureless white surface is untrackable, and you will discover this after you have paid for the clip.\n\n## The same logic applies to people\n\nA real person's face works the same way. Film them against a plain wall on a phone, key them or mask them, and composite into a generated environment. This is a decades-old technique and it is the honest route to putting a specific real person into a generated scene, with their written permission, which is a separate matter dealt with later in the course.\n\n---\n\nToday: photograph one real object against a wall in window light, cut it out in Photopea, and drop it into a generated still. Judge only the contact shadow."},"check":{"question":"A composited product sits in a generated kitchen. The perspective, colour and scale all match, yet it still reads as pasted on. What is the most likely cause?","options":[{"text":"The object was shot at a higher resolution than the generated plate, so its detail level is inconsistent with the surrounding pixels.","correct":false,"tempting":"Names a real mismatch that is usually fixed by downscaling, and which rarely produces the floating quality described."},{"text":"There is no contact shadow where the object meets the surface, so nothing anchors it to the plane it is supposed to be resting on.","correct":true},{"text":"The cut-out edge is too soft, and a softer edge than the surrounding focus makes an object read as a separate layer.","correct":false,"tempting":"Inverts the usual error: composites more often fail from edges that are too hard and clean, not too soft."},{"text":"The generated plate has no grain, while the photograph does, and that texture difference is what the eye registers as a separate element.","correct":false,"tempting":"Identifies a genuine tell, but the stated scenario has everything matched except the anchoring, and grain alone rarely makes an object float."}]},"prev":{"slug":"repairing-a-still-by-hand","title":"Fix the frame yourself instead of re-rolling it","url":"https://addaly.com/learn/ai-video/repairing-a-still-by-hand"},"next":{"slug":"start-and-end-frames","title":"Deciding where the shot lands","url":"https://addaly.com/learn/ai-video/start-and-end-frames"},"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) 'Put the real object in before you animate it', in Video with AI. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-video/compositing-the-real-thing (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":"Put the real object in before you animate it","courseTitle":"Video with AI","module":{"index":2,"slug":"the-first-frame","title":"The first frame, and everything decided in it"},"answer":null,"keyIdea":"A model has no reference for your client’s specific product or person, so the professional split is to generate the environment and the camera move and composite the real thing into the still, because a nearly-right label reads as a counterfeit rather than a stylisation.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"the-problem-stated-precisely","runs":[{"text":"The problem, stated precisely"}]},{"t":"p","runs":[{"text":"Ask a model for “a bottle of Himalaya face wash on a bathroom shelf” and you get a bottle. It will be the right shape family, the right colour family, and the label will be almost right, which is worse than obviously wrong, because almost right reads as a fake."}]},{"t":"p","runs":[{"text":"The reason is not capability, it is information. The model has no representation of that specific bottle. No amount of prompting, steps or guidance retrieves data that is not there. This is the one limit in the whole subject that does not get better with scale."}]},{"t":"p","runs":[{"text":"The answer is not to try harder. It is to change the division of labour: "},{"text":"the model makes the world, you supply the thing.","b":true}]},{"t":"h2","id":"two-places-to-composite-and-how-to-choose","runs":[{"text":"Two places to composite, and how to choose"}]},{"t":"p","runs":[{"text":"Into the still, before generating.","b":true},{"text":" The object becomes part of the first frame and gets animated with everything else. Cheap, simple, and the object moves naturally with the scene."}]},{"t":"p","runs":[{"text":"Use this when the camera move is small: a slow push, a slight drift, a locked-off frame with ambient motion. The object will soften and may distort slightly as the clip runs, which a short shot and a light move will hide."}]},{"t":"p","runs":[{"text":"Into the finished clip, in the editor.","b":true},{"text":" Generate the environment and camera move with a stand-in or an empty surface, then track the real object in afterwards with motion tracking."}]},{"t":"p","runs":[{"text":"Use this when the object must stay pin-sharp and exactly itself, when the camera move is large, or when the client will look at it frame by frame. This is more work and it is the only version that survives scrutiny."}]},{"t":"p","runs":[{"text":"The honest rule: anything a client is paying for a close-up of goes in during the edit, not in the still."}]},{"t":"h2","id":"making-a-composite-hold-up-in-the-still","runs":[{"text":"Making a composite hold up in the still"}]},{"t":"p","runs":[{"text":"Five things, in order. Get the first three right and most people never notice the rest."}]},{"t":"ol","start":1,"items":[[{"text":"Perspective.","b":true},{"text":" The object’s viewing angle must match the scene’s. A bottle photographed straight on cannot sit on a table seen from above. Reshoot or regenerate the background. Do not try to fix this by distorting, because distortion is exactly what the eye catches."}],[{"text":"Light direction.","b":true},{"text":" The key light on the object must come from where the key light in the scene comes from. You decided that direction in the previous lesson; now you shoot the object under it. A window and a piece of white card is enough."}],[{"text":"Scale and contact.","b":true},{"text":" The object needs a shadow where it touches the surface, and that shadow needs the same softness as other shadows in the frame. A composite with no contact shadow floats, and floating is the single most common tell."}],[{"text":"Colour and contrast.","b":true},{"text":" Match the object’s black point and colour cast to the plate. Nine times out of ten this is one curves adjustment."}],[{"text":"Grain and edge.","b":true},{"text":" Generated frames are smooth; a real photograph has noise. Add matching grain to the composited object, and soften its cut-out edge by a pixel or two so it does not look laser-cut."}]]},{"t":"h2","id":"shooting-the-object-without-a-studio","runs":[{"text":"Shooting the object without a studio"}]},{"t":"p","runs":[{"text":"This is where thirty seconds of real filming beats an afternoon of generating, and everybody can see that it did."}]},{"t":"p","runs":[{"text":"A phone, a plain wall or a sheet of paper, and daylight from a window. Turn the phone’s flash off. Put the object a metre from the wall so its own shadow does not fall on the background. Shoot from the angle the composite needs, not the angle that is comfortable. Take twenty frames and pick one."}]},{"t":"p","runs":[{"text":"Cut it out in Photopea or GIMP with the pen tool for hard edges and a mask for soft ones. Background removal tools handle most objects in one click now and are worth trying first; they fail on hair, glass and thin structures, which is where you take over by hand."}]},{"t":"h2","id":"tracking-it-into-a-moving-clip","runs":[{"text":"Tracking it into a moving clip"}]},{"t":"p","runs":[{"text":"When the object has to go into the finished footage:"}]},{"t":"ul","items":[[{"text":"DaVinci Resolve’s Fusion page","b":true},{"text":", in the free tier, does planar and point tracking, and it is capable. The workflow is: track a feature in the plate, attach the object to the track, match grade, add grain."}],[{"text":"Blender","b":true},{"text":" does the same with its motion tracker and is free and open, with a steeper climb."}],[{"text":"Kdenlive","b":true},{"text":" has basic tracking; enough to stick a logo on a slow move, not enough for a hero product shot."}]]},{"t":"p","runs":[{"text":"Generate the plate with a simple, high-contrast trackable feature in it: a table corner, a doorframe, a mark on the wall. A featureless white surface is untrackable, and you will discover this after you have paid for the clip."}]},{"t":"h2","id":"the-same-logic-applies-to-people","runs":[{"text":"The same logic applies to people"}]},{"t":"p","runs":[{"text":"A real person’s face works the same way. Film them against a plain wall on a phone, key them or mask them, and composite into a generated environment. This is a decades-old technique and it is the honest route to putting a specific real person into a generated scene, with their written permission, which is a separate matter dealt with later in the course."}]},{"t":"rule"},{"t":"p","runs":[{"text":"Today: photograph one real object against a wall in window light, cut it out in Photopea, and drop it into a generated still. Judge only the contact shadow."}]}],"check":{"question":[{"text":"A composited product sits in a generated kitchen. The perspective, colour and scale all match, yet it still reads as pasted on. What is the most likely cause?"}],"options":[{"text":[{"text":"The object was shot at a higher resolution than the generated plate, so its detail level is inconsistent with the surrounding pixels."}],"correct":false,"tempting":[{"text":"Names a real mismatch that is usually fixed by downscaling, and which rarely produces the floating quality described."}]},{"text":[{"text":"There is no contact shadow where the object meets the surface, so nothing anchors it to the plane it is supposed to be resting on."}],"correct":true,"tempting":null},{"text":[{"text":"The cut-out edge is too soft, and a softer edge than the surrounding focus makes an object read as a separate layer."}],"correct":false,"tempting":[{"text":"Inverts the usual error: composites more often fail from edges that are too hard and clean, not too soft."}]},{"text":[{"text":"The generated plate has no grain, while the photograph does, and that texture difference is what the eye registers as a separate element."}],"correct":false,"tempting":[{"text":"Identifies a genuine tell, but the stated scenario has everything matched except the anchoring, and grain alone rarely makes an object float."}]}],"order":[2,1,3,0],"why":null},"prev":{"slug":"repairing-a-still-by-hand","title":"Fix the frame yourself instead of re-rolling it","minutes":10},"next":{"slug":"start-and-end-frames","title":"Deciding where the shot lands","minutes":9}}}