{"course":{"slug":"image-editing","title":"Editing an Image","url":"https://addaly.com/learn/image-editing","level":"beginner","track":"create"},"lesson":{"slug":"when-not-to-use-any-of-it","title":"The decision, and the parts nobody has settled","url":"https://addaly.com/learn/image-editing/when-not-to-use-any-of-it","position":67,"of":78,"minutes":9,"answer":"Ask whether the picture is a record or a construction, because that determines whether generated content is permissible, and treat the copyright status of model training as genuinely unresolved.","keyIdea":"Ask whether the picture is a record or a construction, because that determines whether generated content is permissible, and treat the copyright status of model training as genuinely unresolved, documenting what you used rather than assuming an answer."},"body":{"format":"markdown","text":"## One question, asked before the tool is opened\n\n> **Is this picture a record of something, or a construction?**\n\nA record asserts that the world looked like this. A construction asserts nothing beyond itself. Almost every difficult case in this module resolves once you have answered that, and the answer depends on the picture's use rather than on how it was made.\n\nThe same photograph of a shop can be a record (in a news story about that shop, in an insurance claim, in a planning application) and a construction, on a poster advertising it. The file is identical. The obligations are not.\n\n## Where generated content is not acceptable\n\n**News and documentary.** The major wire services and photojournalism competitions prohibit adding or removing content. Only what a darkroom could do (exposure, contrast, colour, crop) is permitted. Photographers have lost their careers for cloning a single object out of a news picture. Generative fill is not in a different category from cloning here; it is a more capable version of the same prohibited act.\n\n**Evidence.** Insurance, legal, forensic, accident, inspection. A denoised, upscaled or filled region is not a record of what was there. If the image will be relied on, process it minimally and keep the original.\n\n**Medical and scientific imaging.** Learned denoising and upscaling produce plausible structure. In an image used for diagnosis or measurement, plausible structure is a hazard, and the relevant professional standards address this directly.\n\n**Identification.** Faces, number plates, documents, serial numbers. Face restoration models regenerate features. An upscaled number plate is a guess.\n\n**Products, where the buyer receives the item.** Covered in the retouching block, and generative tools make it easier to cross the line without noticing.\n\n**People's likeness without consent.** Generating a person doing something they did not do, placing a real person in a scene, altering someone's body or face beyond what they agreed to. This runs into personality and publicity rights, defamation, and in many places specific new offences around synthetic intimate imagery. It is also not something to do.\n\n## Where it is uncontroversial\n\nAdvertising, illustration, book covers, posters, concept work, game art, mood boards, backgrounds for design, extending a canvas for a layout, restoring a family photograph for a frame, removing a distraction from a personal picture.\n\nIn these, nobody is being told that the world looked like this, and the question does not arise.\n\n## The middle, where judgement lives\n\n**Real estate.** Widely done, increasingly complained about, disclosure conventions emerging and varying by market.\n\n**Portraits and weddings.** Retouching is expected. The line is the subject's expectations, which is why asking is the answer.\n\n**Editorial illustration**: an image accompanying an opinion piece rather than a report. Usually acceptable with a label.\n\n**Corporate and internal.** Usually fine, occasionally not, and a line in the delivery note covers it.\n\n**Competitions.** Rules vary enormously and change. Read them, for the specific category, this year.\n\nThe working habit that covers all of these: **write down what you did.** Every module of this course has come back to that, because it converts an unstated assumption into a stated one, and almost every dispute in this trade is an unstated assumption meeting a different unstated assumption.\n\n## Labelling requirements, as they actually stand\n\nSeveral jurisdictions have introduced or proposed obligations to disclose synthetically generated or substantially altered content. The scope, the definitions, the thresholds and the enforcement differ between them; some target platforms, some target creators, some target specific categories such as political advertising or intimate imagery. Some are in force, some are phasing in, and some are being challenged.\n\nThere is no single rule to state, and a course that gave you one would be describing one country and misleading everybody else. What is generally safe:\n\n- Disclose generative work in writing to your client.\n- Do not present generated content as a record of an event.\n- Check the specific rules for your country and your sector, this year, and for anything with legal exposure ask a lawyer in your jurisdiction. Nothing in this course is legal advice.\n\n## The unresolved question\n\nWhether training image models on copyrighted works without permission is lawful is **not settled**, and it is important to present the shape of the disagreement rather than a conclusion.\n\nOne position: training is a transformative analytical use that does not reproduce the works, comparable to a person learning from looking, and treating it as infringement would make machine learning on public material impossible.\n\nThe other: the works were copied to build a commercial product, the creators were neither asked nor paid, and the output competes directly with the market for the originals.\n\nBetween them sit questions courts are actively working through: whether the copying for training is fair use or fair dealing, whether an output that resembles a training work infringes, whether a model's weights are a derivative work, whether text and data mining exceptions cover commercial training, and what opt-outs mean in practice. Different jurisdictions have reached different provisional answers, litigation is ongoing in several, and legislation is moving.\n\nThere is also an **output-side** question that is separate and less discussed: whether an image produced by a model can be copyrighted at all. Several copyright offices have taken the position that purely machine-generated output lacks the human authorship their law requires, with protection possible for human-authored arrangement and modification. If a client needs to own what you deliver, that matters commercially, and it is worth raising before the work rather than after.\n\n## What to do while it is unresolved\n\n- **Read the model's licence**, which is a contract question and is settled, unlike the copyright question.\n- **Prefer models with documented training data** where you have a choice.\n- **Tell clients when generative tools were used**, so the ownership question is theirs to weigh.\n- **Do not reproduce a living artist's identifiable style on commission** and represent it as generic. Whatever the copyright position turns out to be, that is a reputational and possibly a passing-off problem now.\n- **Keep your own originals**, which is the one thing that remains valuable under every possible outcome.\n\nThe honest summary: the tools are useful, several of their legal foundations are contested, and the contest is not close to finished. Working with that uncertainty means documenting what you did and being straightforward about it, which is a smaller burden than it sounds and is the same discipline this course has asked for from the first module."},"check":{"question":"A photographer is asked to use generative fill to remove a distracting parked van from a photograph. Which consideration determines whether it is acceptable?","options":[{"text":"Whether the result is convincing enough that no viewer would detect the edit.","correct":false,"tempting":"Makes quality the test: a more convincing fabrication is more misleading rather than more acceptable, which inverts the reasoning."},{"text":"Whether a conventional tool such as content-aware fill could have achieved the same removal, since only generative tools raise the question.","correct":false,"tempting":"Draws the line at the tool rather than the act: wire services prohibit adding or removing content regardless of which technique performs it."},{"text":"What the picture is for: whether it asserts that the scene looked like this, or is a construction where no such claim is made.","correct":true},{"text":"Whether the client has authorised the removal in writing, since their instruction settles the question of acceptability.","correct":false,"tempting":"Treats client authorisation as sufficient, when a client cannot authorise fabricating a news photograph or an evidential image on anyone's behalf."}]},"prev":{"slug":"provenance-watermarks-and-c2pa","title":"What can be proved about where an image came from","url":"https://addaly.com/learn/image-editing/provenance-watermarks-and-c2pa"},"next":{"slug":"resampling-algorithms","title":"Nearest, bicubic, Lanczos: what the resize dialogue is computing","url":"https://addaly.com/learn/image-editing/resampling-algorithms"},"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) 'The decision, and the parts nobody has settled', in Editing an Image. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/image-editing/when-not-to-use-any-of-it (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":"The decision, and the parts nobody has settled","courseTitle":"Editing an Image","module":{"index":7,"slug":"ai-inside-the-image-editor","title":"AI inside the image editor"},"answer":null,"keyIdea":"Ask whether the picture is a record or a construction, because that determines whether generated content is permissible, and treat the copyright status of model training as genuinely unresolved, documenting what you used rather than assuming an answer.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"one-question-asked-before-the-tool-is-opened","runs":[{"text":"One question, asked before the tool is opened"}]},{"t":"quote","runs":[{"text":"Is this picture a record of something, or a construction?","b":true}]},{"t":"p","runs":[{"text":"A record asserts that the world looked like this. A construction asserts nothing beyond itself. Almost every difficult case in this module resolves once you have answered that, and the answer depends on the picture’s use rather than on how it was made."}]},{"t":"p","runs":[{"text":"The same photograph of a shop can be a record (in a news story about that shop, in an insurance claim, in a planning application) and a construction, on a poster advertising it. The file is identical. The obligations are not."}]},{"t":"h2","id":"where-generated-content-is-not-acceptable","runs":[{"text":"Where generated content is not acceptable"}]},{"t":"p","runs":[{"text":"News and documentary.","b":true},{"text":" The major wire services and photojournalism competitions prohibit adding or removing content. Only what a darkroom could do (exposure, contrast, colour, crop) is permitted. Photographers have lost their careers for cloning a single object out of a news picture. Generative fill is not in a different category from cloning here; it is a more capable version of the same prohibited act."}]},{"t":"p","runs":[{"text":"Evidence.","b":true},{"text":" Insurance, legal, forensic, accident, inspection. A denoised, upscaled or filled region is not a record of what was there. If the image will be relied on, process it minimally and keep the original."}]},{"t":"p","runs":[{"text":"Medical and scientific imaging.","b":true},{"text":" Learned denoising and upscaling produce plausible structure. In an image used for diagnosis or measurement, plausible structure is a hazard, and the relevant professional standards address this directly."}]},{"t":"p","runs":[{"text":"Identification.","b":true},{"text":" Faces, number plates, documents, serial numbers. Face restoration models regenerate features. An upscaled number plate is a guess."}]},{"t":"p","runs":[{"text":"Products, where the buyer receives the item.","b":true},{"text":" Covered in the retouching block, and generative tools make it easier to cross the line without noticing."}]},{"t":"p","runs":[{"text":"People’s likeness without consent.","b":true},{"text":" Generating a person doing something they did not do, placing a real person in a scene, altering someone’s body or face beyond what they agreed to. This runs into personality and publicity rights, defamation, and in many places specific new offences around synthetic intimate imagery. It is also not something to do."}]},{"t":"h2","id":"where-it-is-uncontroversial","runs":[{"text":"Where it is uncontroversial"}]},{"t":"p","runs":[{"text":"Advertising, illustration, book covers, posters, concept work, game art, mood boards, backgrounds for design, extending a canvas for a layout, restoring a family photograph for a frame, removing a distraction from a personal picture."}]},{"t":"p","runs":[{"text":"In these, nobody is being told that the world looked like this, and the question does not arise."}]},{"t":"h2","id":"the-middle-where-judgement-lives","runs":[{"text":"The middle, where judgement lives"}]},{"t":"p","runs":[{"text":"Real estate.","b":true},{"text":" Widely done, increasingly complained about, disclosure conventions emerging and varying by market."}]},{"t":"p","runs":[{"text":"Portraits and weddings.","b":true},{"text":" Retouching is expected. The line is the subject’s expectations, which is why asking is the answer."}]},{"t":"p","runs":[{"text":"Editorial illustration","b":true},{"text":": an image accompanying an opinion piece rather than a report. Usually acceptable with a label."}]},{"t":"p","runs":[{"text":"Corporate and internal.","b":true},{"text":" Usually fine, occasionally not, and a line in the delivery note covers it."}]},{"t":"p","runs":[{"text":"Competitions.","b":true},{"text":" Rules vary enormously and change. Read them, for the specific category, this year."}]},{"t":"p","runs":[{"text":"The working habit that covers all of these: "},{"text":"write down what you did.","b":true},{"text":" Every module of this course has come back to that, because it converts an unstated assumption into a stated one, and almost every dispute in this trade is an unstated assumption meeting a different unstated assumption."}]},{"t":"h2","id":"labelling-requirements-as-they-actually-stand","runs":[{"text":"Labelling requirements, as they actually stand"}]},{"t":"p","runs":[{"text":"Several jurisdictions have introduced or proposed obligations to disclose synthetically generated or substantially altered content. The scope, the definitions, the thresholds and the enforcement differ between them; some target platforms, some target creators, some target specific categories such as political advertising or intimate imagery. Some are in force, some are phasing in, and some are being challenged."}]},{"t":"p","runs":[{"text":"There is no single rule to state, and a course that gave you one would be describing one country and misleading everybody else. What is generally safe:"}]},{"t":"ul","items":[[{"text":"Disclose generative work in writing to your client."}],[{"text":"Do not present generated content as a record of an event."}],[{"text":"Check the specific rules for your country and your sector, this year, and for anything with legal exposure ask a lawyer in your jurisdiction. Nothing in this course is legal advice."}]]},{"t":"h2","id":"the-unresolved-question","runs":[{"text":"The unresolved question"}]},{"t":"p","runs":[{"text":"Whether training image models on copyrighted works without permission is lawful is "},{"text":"not settled","b":true},{"text":", and it is important to present the shape of the disagreement rather than a conclusion."}]},{"t":"p","runs":[{"text":"One position: training is a transformative analytical use that does not reproduce the works, comparable to a person learning from looking, and treating it as infringement would make machine learning on public material impossible."}]},{"t":"p","runs":[{"text":"The other: the works were copied to build a commercial product, the creators were neither asked nor paid, and the output competes directly with the market for the originals."}]},{"t":"p","runs":[{"text":"Between them sit questions courts are actively working through: whether the copying for training is fair use or fair dealing, whether an output that resembles a training work infringes, whether a model’s weights are a derivative work, whether text and data mining exceptions cover commercial training, and what opt-outs mean in practice. Different jurisdictions have reached different provisional answers, litigation is ongoing in several, and legislation is moving."}]},{"t":"p","runs":[{"text":"There is also an "},{"text":"output-side","b":true},{"text":" question that is separate and less discussed: whether an image produced by a model can be copyrighted at all. Several copyright offices have taken the position that purely machine-generated output lacks the human authorship their law requires, with protection possible for human-authored arrangement and modification. If a client needs to own what you deliver, that matters commercially, and it is worth raising before the work rather than after."}]},{"t":"h2","id":"what-to-do-while-it-is-unresolved","runs":[{"text":"What to do while it is unresolved"}]},{"t":"ul","items":[[{"text":"Read the model’s licence","b":true},{"text":", which is a contract question and is settled, unlike the copyright question."}],[{"text":"Prefer models with documented training data","b":true},{"text":" where you have a choice."}],[{"text":"Tell clients when generative tools were used","b":true},{"text":", so the ownership question is theirs to weigh."}],[{"text":"Do not reproduce a living artist’s identifiable style on commission","b":true},{"text":" and represent it as generic. Whatever the copyright position turns out to be, that is a reputational and possibly a passing-off problem now."}],[{"text":"Keep your own originals","b":true},{"text":", which is the one thing that remains valuable under every possible outcome."}]]},{"t":"p","runs":[{"text":"The honest summary: the tools are useful, several of their legal foundations are contested, and the contest is not close to finished. Working with that uncertainty means documenting what you did and being straightforward about it, which is a smaller burden than it sounds and is the same discipline this course has asked for from the first module."}]}],"check":{"question":[{"text":"A photographer is asked to use generative fill to remove a distracting parked van from a photograph. Which consideration determines whether it is acceptable?"}],"options":[{"text":[{"text":"Whether the result is convincing enough that no viewer would detect the edit."}],"correct":false,"tempting":[{"text":"Makes quality the test: a more convincing fabrication is more misleading rather than more acceptable, which inverts the reasoning."}]},{"text":[{"text":"Whether a conventional tool such as content-aware fill could have achieved the same removal, since only generative tools raise the question."}],"correct":false,"tempting":[{"text":"Draws the line at the tool rather than the act: wire services prohibit adding or removing content regardless of which technique performs it."}]},{"text":[{"text":"What the picture is for: whether it asserts that the scene looked like this, or is a construction where no such claim is made."}],"correct":true,"tempting":null},{"text":[{"text":"Whether the client has authorised the removal in writing, since their instruction settles the question of acceptability."}],"correct":false,"tempting":[{"text":"Treats client authorisation as sufficient, when a client cannot authorise fabricating a news photograph or an evidential image on anyone’s behalf."}]}],"order":[1,0,3,2],"why":null},"prev":{"slug":"provenance-watermarks-and-c2pa","title":"What can be proved about where an image came from","minutes":8},"next":{"slug":"resampling-algorithms","title":"Nearest, bicubic, Lanczos: what the resize dialogue is computing","minutes":10}}}