On being listed as an artist whose work was used to train Midjourney
Generative AI tools like Midjourney and OpenAI’s models are provoking a backlash from artists whose work appears to have been scraped without consent to train commercial systems. Commenters argue over whether this is protected “fair use” or large‑scale copyright infringement that undermines livelihoods, noting the asymmetry between how tech companies treat others’ IP versus their own model weights and outputs. Broader concerns include the power shift toward a few dominant platforms, the risk of hollowing out human creative professions, and what kinds of licensing, regulation, or open alternatives—if any—could make AI-assisted art ethically and economically sustainable.
Scope of AI Training on Art
- Major split over whether using online art to train models is akin to human learning or a fundamentally different, industrial‑scale copying.
- One side: training = “reading” or “looking”; models store patterns, not works; this should be fair use and licenses on input are irrelevant if the use isn’t infringing.
- Other side: models plus outputs are derivative works of the training set, especially when they can emit near‑verbatim images or text; that should trigger copyright obligations.
Fair Use, Law, and Open Questions
- Repeated emphasis that fair use in this context is legally unsettled; multiple factors (commercial intent, market harm, degree of substitution) could cut either way.
- Debate over whether company size or profits matter legally (most say no) vs morally (many say yes).
- Some argue that even if training is allowed, specific infringing outputs (near‑exact NYT text, copied images) are not.
Company Behavior and Hypocrisy
- Strong criticism of big AI firms: they invoke fair use to scrape others’ work but impose restrictive licenses on their own models and outputs.
- Argument that if scraping models is fair use, then training competitors on those models’ outputs/weights should also be fair use; otherwise it’s one‑way “have your cake and eat it too.”
Artist Livelihood, Consent, and Exploitation
- Many posters focus on loss of consent and unpaid labor: posting work online now effectively means helping train commercial models without choice or compensation.
- Concerns that gen‑AI will displace illustrators, game artists, and other visual roles, especially in cheap, high‑volume contexts (book covers, stock art, kids’ books, marketing images).
- Others counter that tools mostly augment artists, and that fears resemble earlier panics about photography, printing, and recorded music.
Midjourney and “Cat and Girl” Specifically
- Multiple users try to generate images “in the style of” the comic; most reports say Midjourney does poorly unless custom‑trained, often treating “Cat and Girl” as generic nouns.
- Some conclude the comic is describing a threat that, for this specific artist, is not yet technically real; others reply that the broader systemic issue still stands.
Proposed Remedies and Futures
- Suggestions include: paid licensing for training data, public‑domain‑only models, models trained on explicitly licensed stock, or requiring open weights for models built on public data.
- More radical views: abolish or drastically shorten copyright, fund UBI with AI profits, or accept that “art jobs” become niche/patronage work while creation becomes mostly tool‑mediated.