Adobe is buying videos for $3 per minute to build AI model
Adobe is offering around $3 per minute for contributor-shot video of everyday human actions to train its AI models, positioning this as a legally safe alternative to scraping web content. Commenters debate whether this rate fairly compensates creators given production costs and the risk of enabling AI that might compete with their work, while noting that many tech firms still train on unlicensed data. The move is seen as both a reputational play aimed at risk‑averse customers and a potential step toward regulatory capture that could favor large incumbents over open-source and smaller players.
Adobe’s Strategy and Legal Positioning
- Many see Adobe’s paid, licensed-data approach as a deliberate contrast to “scrape everything” tactics from other AI companies.
- Supporters say this reassures risk‑averse enterprise customers who fear copyright liability and want a clean provenance story.
- Critics argue Adobe is betting that courts will eventually penalize large‑scale unlicensed scraping, thus giving Adobe and similar rights‑holders a defensible moat.
- Some question whether long‑term licenses over persona/likeness should even be signable, raising possible future constraints (e.g., time-limited or revocable rights), though others see that as an overreach.
Copyright, Fair Use, and Training Data
- Debate over whether using content for training is more like humans learning from existing works (often seen as fair use) or like copying without permission.
- Comparisons made to: streaming vs downloading, ripping DRM’d content, libraries, classroom use, memes, and derivative characters in comics.
- One side argues law and policy should prevent big content conglomerates from monopolizing training data.
- Others think some form of licensing/compensation is inevitable, but worry it still may leave most creators underpaid, as with music streaming.
Compensation: Is $3/Minute Fair?
- Some view $3/min as attractive “found money,” especially for low‑effort footage or unused material.
- Others point out that high‑quality or animated content can take hours per minute to produce, reducing the effective rate to near exploitative levels.
- Because the deal is reportedly non‑exclusive, some see it as incremental revenue; others still object to selling training rights so cheaply.
- Concern that Adobe is selectively attracting creators willing to sell cheap, which might bias model output quality.
Creator Impact and Future of Work
- Worries that AI trained on creators’ work will undercut their own future income, effectively paying once to be replaced.
- Counterargument: technology has always displaced some jobs while creating others, and society still runs at low unemployment, albeit with many “bullshit jobs.”
- Existential concerns raised about AI/robotic automation, UBI, and who ultimately captures the value.
Open vs Corporate AI and Regulatory Capture
- Some argue open models trained on unlicensed data are crucial to prevent AI being locked behind corporate paywalls.
- Others note current open models are mostly downstream of big-company spending and may be threatened if licensing becomes mandatory.
- Fears expressed about “data laundering” (using synthetic data to obscure original unlicensed sources) and future lobbying to restrict open-source or small‑scale model training.