0.4Rights, ethics & commercial use
The essentials in 30 seconds
- Three questions to distinguish: the training data (disputed, ongoing lawsuits), image ownership (tool ToS + varying local law), and what the image SHOWS (the everyday filter).
- Content to rule out: identifiable real person, brand or logo, protected work or character — and caution with the style of a named living artist.
- Law varies by country and evolves fast: read the up-to-date ToS — this guide gives reference points, not legal advice.
Running mini-case — Sonia secures her first sale in Algiers
Sonia, freelance graphic designer in Algiers, is about to sell a series of visuals to a client: before invoicing, she runs her filter. She rereads the up-to-date ToS of her tool and plan (what does commercial use cover?), rules out a track that looked too much like the style of a named living artist, replaces a blurry logo with a neutral shape, and fixes a portrait series carrying a stereotype. She keeps her prompts and versions, discloses the AI visuals, and notes the point to validate with a local lawyer before the major campaign. Her sale is clean — and traceable.
Standalone case for this section: it returns in the course, the diagram, the script and the 48h challenge of section 0.4.
1. Three legal questions not to confuse
1) Training data: models were trained on huge corpora of images collected from the web, many protected by copyright. Several lawsuits are ongoing in various countries over the legality of this practice; the matter is not settled everywhere and is evolving. 2) Ownership of the generated image: who owns the result? Each tool's terms of service (ToS) settle this contractually — but whether a purely generated image can be protected by copyright gets different answers by country, with several jurisdictions requiring sufficient human input. 3) What the image shows: even with full rights to the file, you cannot show just anything. Sonia learns all three by heart: training, ownership, content.
2. The content filter and everyday ethics
The third point is the most concrete day to day. Watch out for real people (image rights, deepfake risk — never generate an identifiable person without consent), brands, logos and protected products, protected works and characters, and the style of living artists: explicitly imitating the style of a named artist is ethically questionable and legally risky depending on context. On the ethics side, four watch points: correct model biases (gender, origin or body stereotypes, in the prompt and the selection), avoid disinformation (a realistic image can deceive — traceability, section 5.5), be transparent with the audience, and respect creators' work. Sonia runs each visual through the filter before publishing: person, brand, work, named style, bias.
3. Commercial use: the careful method
To sell or commercially exploit: check the ToS of your plan (free offers are often more restrictive), avoid any protected or identifiable element, add real human work (composition, retouching, integration — section 4.5: it also strengthens your position), keep prompts and versions as proof of your process, and disclose the AI nature when expected (section 5.5). Essential warning: this guide gives common-sense reference points, not legal advice. The applicable law depends on your country and evolves fast — for engaging commercial use (major campaign, packaging, trademark filing), read the up-to-date ToS and consult a local legal professional. Sonia writes it on her quote: guide reference points + lawyer validation for major stakes.
🔎 Concrete example — 🔎 Concrete example — The content filter before publishing
Sonia runs her series of visuals through the filter, one by one, before delivering.
/imagine prompt: smiling shopkeeper in a market, neutral style, natural light, documentary photo
→ FILTER: identifiable person? brand/logo? protected work? named artist style? bias?
→ visual with a lookalike logo: RULED OUT, regenerated as neutral shape.
→ keep the prompt + version, disclose the AI nature.Reading: the filter catches what enthusiasm misses: a logo, a named style, a stereotype — before delivery, not after.
Lesson: read the up-to-date ToS and keep your prompts: traceability is half of care.
Section vocabulary
- Copyright
- Protection of original works; its application to generated images varies by country and keeps evolving.
Simple example: Sonia knows her purely generated image is not protectable everywhere.
What it is not: a global, uniform, settled right for AI.
Reflex: add substantial human work and check your local law.
- Terms of service (ToS)
- The tool's contract defining your rights over generated images; read them, they differ from one service to another.
Simple example: Sonia rereads the up-to-date ToS of her plan before selling her series.
What it is not: an identical formality at every tool.
Reflex: read the up-to-date version, for your exact plan, before any commercial use.
- Image rights
- A person's right over the representation of their face; generating an identifiable person without consent is risky.
Simple example: Sonia rules out any portrait resembling a real person.
What it is not: a right that would only concern photographers.
Reflex: no identifiable face without explicit consent.
- Deepfake
- Deceptive synthetic representation of a real person; to be ruled out outside an explicit, consented framework.
Simple example: Sonia refuses to generate a fake client testimonial.
What it is not: a simple fun filter with no consequence.
Reflex: say no to non-consented real faces, always.
- Model bias
- Stereotypes reproduced by the model (gender, origin, body); to be actively corrected in the prompt and selection.
Simple example: Sonia explicitly diversifies her portrait prompts and sorts her selection.
What it is not: a personal opinion of the designer.
Reflex: check each series: who is shown, who is missing?
In practice — In practice — Securing a commercial use
- Read the UP-TO-DATE terms of service of your tool, like Sonia: what exactly does it allow you (commercial use, under which plan)?
- Run the image through the content filter: real person? brand/logo? protected work or character? named-artist style?
- Correct biases: check your visuals carry no unintended stereotypes; diversify explicitly if needed.
- Document and disclose: keep your prompts, and disclose the AI nature when expected (see section 5.5).
Ready-to-say script — setting the rights framework before selling (30s)
'Before selling these visuals, I did like Sonia: up-to-date ToS of our plan reread, content filter passed — no identifiable person, no brand, no protected work, no named-artist style — biases corrected, prompts kept, AI visuals disclosed. And I recall the limit: this counts as reference points, not legal advice — for the major campaign, we have it validated by a local lawyer.'
Say with the filter checklist in front of you; have major stakes validated by a lawyer.
Copy-paste template — my responsible delivery checklist
Checklist before delivery (tick or rule out):
[ ] Up-to-date ToS read for MY plan (commercial use covered?)
[ ] Identifiable real person? → consent or regeneration.
[ ] Brand / logo / character / work? → neutralize.
[ ] Living artist style named? → describe by components (light, palette, medium).
[ ] Biases checked? prompts + versions kept? AI disclosure planned?
→ Major stake (campaign, packaging, brand): validation by a local lawyer.Freelance sale
Sonia attaches the ticked checklist to her invoice: the client sees the process, and the kept prompts prove the human work.
Social posting
Visual for social media: AI-nature disclosure planned from the caption, filter rerun (treacherous background logos first).
Team work
As a pair in Algiers: one generates, the other runs the filter with no indulgence — person, brand, work, named style, bias — before any release.
Big stake
Major campaign or packaging: Sonia does not settle for the guide — up-to-date ToS printed in the file + written validation by a local lawyer.
✓ Best practices
- Read the up-to-date ToS of your plan before any commercial use — never from memory or via a forum.
- Rule out real people, brands, protected works and characters; describe styles by components.
- Correct biases, keep your prompts and disclose AI when expected.
✗ Common pitfalls
- Confusing 'rights to the file' (ToS) with 'right to show just anything' (third-party rights).
- Imitating the style of a named living artist: questionable, risky, and counterproductive.
- Taking this guide for legal advice: law varies by country and evolves fast.
Key takeaways
- Distinguish 3 questions: training data (disputed), image ownership (ToS + local law), and what the image SHOWS.
- Do not generate identifiable real people, brands/logos, protected works or characters; caution with living artists' styles.
- Ethics: correct biases, avoid disinformation, be transparent, respect creators' work.
- Law VARIES BY COUNTRY and evolves fast: read the up-to-date ToS; this guide is not legal advice.
48h challenge — apply tomorrow
Run three of your visuals through the content filter like Sonia (person, brand, work, named style, bias), tick the template checklist and note the point to have validated by a lawyer.
- I can name the three legal questions without confusing them.
- I ran my visuals through the content filter and ruled out risk zones.
- I know when to have validation by a local lawyer.
Frequently asked questions
Can I sell or commercially use the images I generate?
This depends on three levels to validate together: your CONTRACT with the tool (ToS of your plan, to read up to date — free offers are often more restrictive), the PROTECTION of the image (a purely generated image with no human input may not be protectable depending on the country — substantial human work strengthens your position), and the represented CONTENT (no ToS authorizes violating a third party's rights: person, brand, work). In practice: verified ToS, protected elements ruled out, human work added, prompts kept — and validation by a lawyer in your country for any major stake.
Is it acceptable to imitate the style of a specific artist?
Technically possible, but AVOID it: ethically questionable (the commercial identity of a living artist, sometimes via a model trained without consent), legally risky depending on context (identifiable elements, confusion over origin, unfair competition — varying by country), and practically counterproductive (generic visuals, dependency). The best approach (level 3): describe style by its components — light, palette, medium, texture, framing, era — or prefer movements and eras over named living artists.
Sonia wants to sell a series of visuals. Which reflex best sums up care?