A new class of AI startups is betting that the fastest way to defuse the artist backlash against generative video is to open a checkbook. According to The Verge AI, a company called Pippa now pays illustrators directly every time a subscriber generates an image or clip in their style. It’s a real attempt to answer a question the whole industry has been dodging: can you build a gen AI business without training on people’s work without consent?
The short answer, based on The Verge AI’s reporting, is not yet. And that gap between intent and reality is the story worth watching.
What Pippa is actually doing
Pippa’s cofounders, Hogan Shrum and Sean Wright, frame their model as a break from what Wright calls “the bloody history that the AI industry has been built on.” Their pitch borrows straight from music history: think Napster giving way to the 99-cent iTunes song. Artists opt in through a vetting process, get models trained on their approved work, and earn per generation.
The numbers are modest. Artists get $0.005 per image and $0.003 per second of video, plus a slice of a 5 percent royalty pool funded by subscriptions that run $14.99 to $99.99 a month. Pippa compares its structure to Spotify, which is a bold analogy given how many musicians say streaming shortchanged them.
The catch nobody has solved
Here’s what stands out. Pippa’s models are still built on open foundations that were, in the company’s own words, trained on “the broader set of content out there.” Translated: scraped internet art, no consent. The artist payments sit on top of a base layer that has the exact problem Pippa claims to fix.
That’s not a Pippa-specific flaw. It’s the structural trap for every “ethical” AI startup. The Verge AI notes the alternative, building a fully proprietary dataset from scratch the way Ben Affleck’s InterPositive does, takes capital most startups don’t have and produces narrow tools, not consumer products.
The traction is early too. Pippa launched in May, has around 800 paying subscribers, and has signed just four artists with four more in talks. Some partners want to publish under pseudonyms to avoid being seen as “crossing the picket line.” That detail tells you how raw the trust deficit still is.
Why this matters now
The legal ground is shifting under every gen AI company. Copyright suits over training data are working through the courts, and licensing is moving from a nice-to-have to a liability shield. Paying artists isn’t only an ethics play. It’s insurance against the day scraping gets ruled offside.
That’s why the “consent-washing” risk is worth naming. A thin royalty layer over a scraped base model can look ethical in a press release while changing little underneath. Buyers, artists, and regulators are going to get better at spotting the difference.
The Future Cast: where this goes
Over the next one to three years, expect a split in the market:
- Licensing becomes table stakes. As court rulings land, “trained with permission” moves from marketing edge to baseline requirement, the way privacy policies became mandatory.
- Provenance gets audited. Vague claims like “initial training on broader content” won’t survive scrutiny. Expect demand for verifiable dataset lineage, not vibes.
- Payment models get tested in public. The Spotify comparison is a warning. Artists have seen streaming economics and will push for better splits before they sign.
Practical takeaways
- If you build AI products: treat data provenance as a core feature, not a footnote. Document what your base model was trained on. Ambiguity is now a legal and reputational cost.
- If you’re an artist weighing an offer: run the math on per-generation rates against expected volume, and read the royalty-pool terms closely. Ask what the underlying model was trained on before you sign.
- If you’re a business buying gen AI tools: ask vendors to show their licensing chain. “Ethical” is a claim you can verify.
Pippa’s intentions read as genuine, but good intent doesn’t erase a scraped foundation. The company worth backing will be the one that closes that gap, not just papers over it. You can find the full report at the original source.