Images: generating, retouching, respecting rights

Understand it in 3 minutes

Where to create images, as of the date. ChatGPT’s pricing page (24/10/2026, viewed in France, in French) shows limited image creation on the free plan, and more image generation on the paid plans. Other tools exist: evaluate them with the method from module 13, lesson 4.

What to check on a generated image, before using it:

• every word written in the image (generated text can be distorted);
• every detail that shows: the place, the objects, the numbers;
• the resemblance: an image that looks like a real person, place or work;
• the style guide: the colors and style from your sheet (module 10, lesson 4). If you haven’t done module 10, simply note two colors and a font.

The rights, according to OpenAI’s terms for Europe (24/10/2026): between you and OpenAI, and to the extent permitted by law, you keep ownership of what you provide, and you own the output.

But the output may not be unique: other people may get similar results. And you are responsible for what you produce: it must not break the law or anyone’s rights.

Owning the output does not mean it is protected by copyright. It is a rule between you and OpenAI. According to the report of the CSPLA (French Ministry of Culture, July 16, 2026), content generated without significant human intervention remains outside copyright, whereas a creation in which you keep a creative role (design, retouching, selection) can be protected if it is original enough. It is an analysis report, not a law.

What you never do:

• publish the image of a recognizable real person (photo, retouched or generated by AI) without their permission. In France, the CNIL (the French data protection authority) points out that you need a person’s permission before publishing their photo (in a public place, only the permission of people who are alone in the shot and recognizable is required: to be safe, always ask). And article 226-8 of the French penal code punishes with one year in prison and a fine of 15,000 euros the act of making public, or sharing with a third party, without the person’s consent, an edit made with their words or image, or AI-generated content that shows their image or words, unless it is obvious or expressly stated that it is an edit or generated content. The penalties rise to two years and a 45,000 euros fine when it is done through an online public communication service;
• reproduce someone else’s logo or brand (Wikimedia Commons points out that brands can restrict reuse, even of a free image).

Flagging a generated image:

• provenance: according to OpenAI, images created with ChatGPT carry provenance information in the C2PA standard and an invisible watermark (SynthID); but platforms, editing tools and file conversions can remove the C2PA information, and heavy changes can make the watermark undetectable;
• YouTube asks creators to flag, when uploading, realistic content generated or modified by AI, for example a realistic-looking scene that never happened;
• the European AI law (article 50, applicable since August 2, 2026) asks anyone who uses an AI system to create or modify an image, a sound or a video that is a deep fake to flag that the content was artificially generated or modified. It targets a person or an organization (company, association…) using the tool under their own responsibility, not a purely personal, non-professional activity. A deep fake resembles existing people, places, objects or events and would falsely appear authentic. For an obviously artistic, creative, satirical or fictional work, the disclosure is limited to indicating that content is generated or modified, in a way that does not get in the way of the work.