ChatGPT is adding real cartoonists' signatures to fake New Yorker cartoons
TL;DR Highlight
ChatGPT image generation draws real cartoonists' signatures onto fake New Yorker cartoons.
Who Should Read
A product developer who plugs an image generation model into a service and shows its output to users as-is
Core Mechanics
- The article body was not provided
- Only the comments are available as source material
- Details of the article itself could not be confirmed
- What the title says
- The title says ChatGPT adds real cartoonists' signatures to fake New Yorker cartoons
- No figures or experimental conditions from the article were provided
Community Reactions
- Opinion that signatures are learned decoration
- A commenter says many 'New Yorker-style cartoon' examples have a signature in the corner, so it is learned as part of the picture
- They argue that avoiding it would need explicit training to treat signatures specially
- Experience that other image models do it too
- Commenters say it has been easy to reproduce since early Stable Diffusion and is not OpenAI-specific
- One example: ukiyo-e requests almost always get a hanko seal unrelated to the era
- Experience with fake signatures in own comics
- One user says Nano Banana Pro and every ChatGPT generation add false signatures that they must erase
- Another says an engineering manager used a ChatGPT comic signed 'bloper' in a sprint demo without noticing
- Split reactions on legal liability and trademarks
- Some argue that commercial sales make a right of publicity claim easy to prove
- Others advise cartoonists to register their name and signature as a trademark
How to Apply
- Check generated images for corner signatures
- Pull 20 samples of your image output today and look at the corners for signatures or seals
- Record how often they appear to size the problem
- Add a no-signature rule to prompts
- Add an instruction to your image prompt not to include signatures, watermarks, or text
- Compare results before and after on the same samples
- Add a human check before publishing
- Add a step in the UI to confirm or remove signatures before users share the result
- Show a warning when a request names a specific artist or publication style
Try It
# Prompt to paste into ChatGPT or Claude (simple checklist for reviewing image generation output)
Below is the prompt I sent to an image generation model.
{{image prompt}}
# Instruction
Write 3 sentences to append to the end of this prompt. Condition: each must stop the model from putting signatures, autographs, watermarks, seals, or readable text into the image.
# Next steps
Append each sentence, generate 10 images per version with the same prompt, and compare how many show a signature in a corner.Terminology
Related Papers
The Judge Can Debias Itself: Awareness and Self-Debiasing for Reliable LLM-as-a-Judge
Warning an LLM judge about bias or letting it self-critique improved 20 of 21 model–bias combinations.
90% of US working hours are spent on tasks not covered well by AI benchmarks
Mapping 29 benchmarks to job tasks shows 90% of US working hours lack good coverage
Postgres 19 delayed after AI reviewers find more bugs than humans can fix
LLM code review surfaced so many bugs that Postgres 19 slipped to November and 8-10 features were reverted
Show HN: Gemini 4 Argon plays StarCraft
LLMs wrote StarCraft bots in one hour; GPT-6 Astra scored highest.
A Plausible Explanation Is Enough to Mislead Multimodal Judges
Adding plausible explanations to fixed answers cut MLLM judges' wrong-answer rejection by 12.7–13.8 points.
Neural networks are clean
Looking at neurons directly shows most have one main job, challenging superposition.