🚨 Core Copyright Issues in AI‑Generated Molecular Visualizations

Before the cases, it helps to understand what the legal questions look like in this technical field:

1) What is being copyrighted?

In scientific visualization (e.g., molecules), you’re creating images or animations that represent data. Copyright law protects original artistic expression, not raw facts, formulas, or underlying data. This matters when AI generates visuals from scientific data. (Copyright protects expression, not ideas or facts.)

2) Who qualifies as the author?

Traditional copyright only attaches to human authorship. If an AI “just does it all,” many jurisdictions say there’s no copyright. But if a human guides, edits, or curates the result meaningfully, there might be copyright in the human part of the work.

3) AI training data and infringement risks

Many lawsuits involve claims that AI was trained on copyrighted material without permission. Even if the output is new, if it’s generated from copyrighted data, there may be claims of infringement.

📌 Key Cases and Rulings (Detailed)

1. Zarya of the Dawn — US Copyright Office Limits AI Image Protection

Jurisdiction: United States (Copyright Office ruling)
Summary:
An 18‑page graphic novel with AI‑generated images was submitted for copyright registration. The US Copyright Office initially granted protection, then withdrew it for the AI images when it learned they were produced by AI with minimal human creativity. Only the text and arrangement of pages were protected.

Why this matters for molecular visuals:
If you generate a molecular image via an AI tool without significant human creativity, that visual is unlikely to be copyrightable on its own in the U.S. — even if embedded in a larger scientific work.

2. A Recent Entrance to Paradise (Théâtre D’opéra Spatial)

Jurisdiction: United States (USCO / Review Board)
Summary:
The US Copyright Office refused copyright for a Midjourney image titled Théâtre D’opéra Spatial (and upheld refusal at review). The decision focused on lack of significant human authorship and treated the AI output as non‑copyrightable.

Why this matters:
Models created entirely by AI without meaningful human input (e.g., no real editing, refinement, or creative control) generally can’t be copyrighted in the U.S. Whether this principle applies to complex scientific graphics depends on how much human intervention there was.

3. Li v Liu — Beijing Internet Court Grants Copyright to AI‑Assisted Work

Jurisdiction: China (Beijing Internet Court)
Summary:
A Chinese court ruled that an AI‑generated image did have copyright because the human user provided “significant intellectual input,” including prompt refinement and creative choices in generating the image. The court awarded copyright to the human, not the AI.

Why this matters:
Some jurisdictions (like China) are willing to grant copyright for AI‑assisted works if the human’s creative choices substantially influence the final image. This has direct parallels for molecular visualization: if a scientist iteratively refines prompts and edits images, that might count as creative input under certain laws.

*4. TDC Circuit ruling on Human Authorship (Stephen Thaler / DABUS)

Jurisdiction: U.S. Court of Appeals (District of Columbia Circuit)
Summary:
A federal appeals court affirmed that a work created entirely by AI with no human involvement cannot be copyrighted in the U.S. Under long‑standing U.S. law, works must be authored by humans to qualify.

Why this matters:
If your molecular visualization workflow leaves AI in complete control (no editorial input, no creative shaping), the output probably has no U.S. copyright protection. This influences institutional IP strategies.

**5. Andersen v. Stability AI / Midjourney Artist Lawsuit

Jurisdiction: U.S. District Court (Northern District of California)
Summary:
Artists sued Stability AI, Midjourney, and others claiming the companies infringed their copyrights by training AI on millions of copyrighted images. Plaintiffs allege the AI output contains derivative elements.

Why this matters:
While not a resolution yet, these cases underscore that AI training datasets — even for scientific visuals — can create copyright risk if they include copyrighted images without permission. This is separate from copyrightability of output and goes to infringement liability.

**6. U.S. Copyright Office Report — Human Authorship Required

Jurisdiction: United States (Copyright Office)
Summary:
In a 2025 report, the U.S. Copyright Office reaffirmed that AI outputs alone aren’t copyrightable. However, human use of AI creatively (selection, arrangement, modification) may create protectable elements. It stressed that simple prompting isn’t enough; material contributions must reflect human originality.

Why this matters:
The report is not a case per se, but it guides courts and lawyers. For molecular visualization, this means if you don’t exert enough human creative control, the visuals won’t be protected — but if you, say, design custom rendering pipelines or artistic enhancements, those parts might be.

🧪 Applying These Rulings to Molecular Modeling Visuals

When AI is used to generate scientific images or animations of molecules, courts will typically ask:

🔹 Is the output “original expression,” or just data visualization?

Factual graphs, charts, and molecular coordinates are not copyrightable — only artistic interpretation is.

🔹 How much did the human contribute?

If all you did was run a model and save the output, protection is weak. If you curated outputs, edited images, and made creative choices, you may have copyright in the human creative contribution.

🔹 Is your training process infringing on others’ copyrighted content?

If you train your AI tools on proprietary visual datasets without permission, plaintiffs might argue derivative infringement even if the output looks “new.”

🔹 Jurisdictional differences matter.

– In the U.S., purely AI output = no copyright; human creativity is key. 
– In China, significant human prompt refinement can suffice. 
– Other countries (UK, EU, Japan) have intermediate or evolving standards.

📌 Practical Takeaways

✅ Human involvement is decisive. Simply generating a molecular graphic using AI with minimal input likely won’t qualify for copyright in the U.S.
✅ Document your creative process. Detailed records of human edits, choice of parameters, and artistic decisions strengthen copyright claims.
✅ Training data matters too. Even if outputs are original, lawsuits over unauthorized training could impact your rights or liability.
✅ International variation complicates global publishing. Visuals might be copyrighted in one country and not in another.

🌍 Final Thought

AI‑generated content is not outside copyright law — it just challenges traditional concepts like authorship and originality. For scientific visuals, it’s crucial to distinguish (a) copyrightability of the output, and (b) infringement based on how the AI was trained or what data it uses. These lines are still evolving, often guided by cases like those above.