{"course":{"slug":"making-things-with-ai","title":"Making Things with AI","url":"https://addaly.com/learn/making-things-with-ai","level":"beginner","track":"create"},"lesson":{"slug":"editing-not-prompting","title":"Editing what you already have","url":"https://addaly.com/learn/making-things-with-ai/editing-not-prompting","position":29,"of":84,"minutes":8,"answer":"Prompting gets you a starting frame; masks, denoise strength and structural inputs get you a finished image.","keyIdea":"Prompting gets you a starting frame; masks, denoise strength and structural inputs get you a finished image."},"body":{"format":"markdown","text":"## Nobody good ships a first generation\n\nThe workflow people actually use looks like this: generate until you have a base with the right composition and light, fix regions one at a time, extend the frame if you need to, upscale last. The prompt got you sixty percent. The rest is masking.\n\n## Inpainting, and the setting that changes everything\n\nInpainting means painting a mask over a region, describing what should be there, and regenerating only that region while the model reads the surrounding pixels for context.\n\nThe setting that separates a clean result from a smear is usually called \"inpaint at full resolution\" or \"inpaint only masked area\". Without it, the entire image is squeezed down to the model's working size, so a face occupying five percent of a wide shot gets maybe forty pixels and comes back as mush. With it, the tool crops around your mask, generates that crop at the model's native resolution, and blends it back. Same model, same prompt, completely different outcome.\n\nThree practical rules:\n\n- Mask a little wider than the problem. Tight masks leave visible seams.\n- Change the prompt to describe the region, not the scene. If you are fixing a hand, prompt the hand.\n- Fix one thing per pass, and save between passes.\n\n## Denoising strength is the dial you actually turn\n\nWhether you are inpainting or running image-to-image, one number governs how far the result may travel from the input. It sets how much noise is added before denoising begins.\n\n- **0.2 to 0.35**: texture and finish only. Shapes survive intact.\n- **0.4 to 0.6**: the same objects with different details. Faces will change. Most real work happens here.\n- **0.7 and up**: the input is a loose suggestion. Composition survives, content does not.\n\nIf your edit is not taking, raise it. If your subject keeps turning into a different person, lower it and do two passes.\n\n## Keeping the structure, changing everything else\n\nStructural conditioning feeds the model a second input alongside your prompt: a depth map, an edge trace, a pose skeleton, a rough scribble. The model must respect that structure while the prompt decides everything else.\n\nThis is how you get the exact same room at night in the rain, or a pencil sketch turned into a finished illustration that keeps your composition. It is the single biggest jump in control that most people have never tried.\n\n## Instruction editing, and its quiet problem\n\nA newer class of model edits from plain instructions with no mask at all. \"Make it evening.\" \"Remove the man on the left.\" \"Put her in a green jacket.\" These are good and have become the default for casual work.\n\nTheir weakness is worth knowing. Most of them re-render the whole image, not just the part you named. One edit is invisible. After five sequential edits, the face has drifted, the sign on the wall has changed its wording, and fine texture has softened toward the model's house style. It is a photocopy of a photocopy.\n\nSo keep the original. Do the biggest change first. Where you can, run independent edits from the same original and composite them, rather than stacking edits on edits.\n\n## Upscaling is not detail recovery\n\nAn upscaler invents plausible detail. It does not recover detail that was never captured. A good one turns a soft image sharp and convincing, and it will also confidently invent the wrong lettering, the wrong eyelashes, and a fabric pattern that was not in the shot. Upscale last, look closely at anything that carries meaning, and expect one more round of inpainting afterwards.\n\n## The mindset shift\n\nTreat generation as photography, not as ordering. You take a lot of frames, choose one, then do the darkroom work. The people producing consistently good AI images are not writing better prompts than you. They are doing four rounds of masking you never see."},"check":{"question":"Someone inpaints a small, distant face in a wide landscape shot. The face comes back soft and mushy, while the same tool fixes faces beautifully in close-up portraits. What is the most likely cause?","options":[{"text":"The denoising strength was set too low, so the model preserved the blur it started from.","correct":false,"tempting":"Names a real cause of soft inpainting results, but one that would affect close-ups equally."},{"text":"The masked region is being generated at a small fraction of the model's native resolution, so the face is only a few dozen pixels wide when the model works on it.","correct":true},{"text":"The prompt described the whole scene instead of just the face, so the model's attention was divided across everything.","correct":false,"tempting":"Imagines the model splits a fixed budget of effort across the things named in the prompt."},{"text":"The model has a weak prior for distant faces, because training photos of faces are overwhelmingly close-ups.","correct":false,"tempting":"Reaches for a training-data explanation when the cause is a mechanical property of how the crop is processed."}]},"prev":{"slug":"the-tells","title":"The tells, and why they are disappearing","url":"https://addaly.com/learn/making-things-with-ai/the-tells"},"next":{"slug":"denoise-strength","title":"Denoise strength: the number that decides how much survives","url":"https://addaly.com/learn/making-things-with-ai/denoise-strength"},"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) 'Editing what you already have', in Making Things with AI. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/making-things-with-ai/editing-not-prompting (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":"Editing what you already have","courseTitle":"Making Things with AI","module":{"index":4,"slug":"control-not-luck","title":"Control: getting the picture you actually need"},"answer":null,"keyIdea":"Prompting gets you a starting frame; masks, denoise strength and structural inputs get you a finished image.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"nobody-good-ships-a-first-generation","runs":[{"text":"Nobody good ships a first generation"}]},{"t":"p","runs":[{"text":"The workflow people actually use looks like this: generate until you have a base with the right composition and light, fix regions one at a time, extend the frame if you need to, upscale last. The prompt got you sixty percent. The rest is masking."}]},{"t":"h2","id":"inpainting-and-the-setting-that-changes-everything","runs":[{"text":"Inpainting, and the setting that changes everything"}]},{"t":"p","runs":[{"text":"Inpainting means painting a mask over a region, describing what should be there, and regenerating only that region while the model reads the surrounding pixels for context."}]},{"t":"p","runs":[{"text":"The setting that separates a clean result from a smear is usually called “inpaint at full resolution” or “inpaint only masked area”. Without it, the entire image is squeezed down to the model’s working size, so a face occupying five percent of a wide shot gets maybe forty pixels and comes back as mush. With it, the tool crops around your mask, generates that crop at the model’s native resolution, and blends it back. Same model, same prompt, completely different outcome."}]},{"t":"p","runs":[{"text":"Three practical rules:"}]},{"t":"ul","items":[[{"text":"Mask a little wider than the problem. Tight masks leave visible seams."}],[{"text":"Change the prompt to describe the region, not the scene. If you are fixing a hand, prompt the hand."}],[{"text":"Fix one thing per pass, and save between passes."}]]},{"t":"h2","id":"denoising-strength-is-the-dial-you-actually-turn","runs":[{"text":"Denoising strength is the dial you actually turn"}]},{"t":"p","runs":[{"text":"Whether you are inpainting or running image-to-image, one number governs how far the result may travel from the input. It sets how much noise is added before denoising begins."}]},{"t":"ul","items":[[{"text":"0.2 to 0.35","b":true},{"text":": texture and finish only. Shapes survive intact."}],[{"text":"0.4 to 0.6","b":true},{"text":": the same objects with different details. Faces will change. Most real work happens here."}],[{"text":"0.7 and up","b":true},{"text":": the input is a loose suggestion. Composition survives, content does not."}]]},{"t":"p","runs":[{"text":"If your edit is not taking, raise it. If your subject keeps turning into a different person, lower it and do two passes."}]},{"t":"h2","id":"keeping-the-structure-changing-everything-else","runs":[{"text":"Keeping the structure, changing everything else"}]},{"t":"p","runs":[{"text":"Structural conditioning feeds the model a second input alongside your prompt: a depth map, an edge trace, a pose skeleton, a rough scribble. The model must respect that structure while the prompt decides everything else."}]},{"t":"p","runs":[{"text":"This is how you get the exact same room at night in the rain, or a pencil sketch turned into a finished illustration that keeps your composition. It is the single biggest jump in control that most people have never tried."}]},{"t":"h2","id":"instruction-editing-and-its-quiet-problem","runs":[{"text":"Instruction editing, and its quiet problem"}]},{"t":"p","runs":[{"text":"A newer class of model edits from plain instructions with no mask at all. “Make it evening.” “Remove the man on the left.” “Put her in a green jacket.” These are good and have become the default for casual work."}]},{"t":"p","runs":[{"text":"Their weakness is worth knowing. Most of them re-render the whole image, not just the part you named. One edit is invisible. After five sequential edits, the face has drifted, the sign on the wall has changed its wording, and fine texture has softened toward the model’s house style. It is a photocopy of a photocopy."}]},{"t":"p","runs":[{"text":"So keep the original. Do the biggest change first. Where you can, run independent edits from the same original and composite them, rather than stacking edits on edits."}]},{"t":"h2","id":"upscaling-is-not-detail-recovery","runs":[{"text":"Upscaling is not detail recovery"}]},{"t":"p","runs":[{"text":"An upscaler invents plausible detail. It does not recover detail that was never captured. A good one turns a soft image sharp and convincing, and it will also confidently invent the wrong lettering, the wrong eyelashes, and a fabric pattern that was not in the shot. Upscale last, look closely at anything that carries meaning, and expect one more round of inpainting afterwards."}]},{"t":"h2","id":"the-mindset-shift","runs":[{"text":"The mindset shift"}]},{"t":"p","runs":[{"text":"Treat generation as photography, not as ordering. You take a lot of frames, choose one, then do the darkroom work. The people producing consistently good AI images are not writing better prompts than you. They are doing four rounds of masking you never see."}]}],"check":{"question":[{"text":"Someone inpaints a small, distant face in a wide landscape shot. The face comes back soft and mushy, while the same tool fixes faces beautifully in close-up portraits. What is the most likely cause?"}],"options":[{"text":[{"text":"The denoising strength was set too low, so the model preserved the blur it started from."}],"correct":false,"tempting":[{"text":"Names a real cause of soft inpainting results, but one that would affect close-ups equally."}]},{"text":[{"text":"The masked region is being generated at a small fraction of the model’s native resolution, so the face is only a few dozen pixels wide when the model works on it."}],"correct":true,"tempting":null},{"text":[{"text":"The prompt described the whole scene instead of just the face, so the model’s attention was divided across everything."}],"correct":false,"tempting":[{"text":"Imagines the model splits a fixed budget of effort across the things named in the prompt."}]},{"text":[{"text":"The model has a weak prior for distant faces, because training photos of faces are overwhelmingly close-ups."}],"correct":false,"tempting":[{"text":"Reaches for a training-data explanation when the cause is a mechanical property of how the crop is processed."}]}],"order":[2,1,3,0],"why":null},"prev":{"slug":"the-tells","title":"The tells, and why they are disappearing","minutes":8},"next":{"slug":"denoise-strength","title":"Denoise strength: the number that decides how much survives","minutes":8}}}