A $7 million bet that people will shop from AI hallucinations instead of real product photos — months before the product even existed.

The Summary

  • Dalet Labs raised $7M pre-seed from Lightspeed and SV Angel for Scene, a shopping app where users scroll AI-generated images linked to real products
  • Founded by ex-Runway AI researcher Omer Bar-Tal, who raised capital before finalizing the product roadmap
  • The pitch: Scene replaces Pinterest/TikTok browsing with AI imagery designed specifically for shopping intent, not entertainment

The Signal

Lightspeed just wrote a seven-figure check for vaporware in the most literal sense possible. Dalet Labs raised its entire pre-seed round in late 2025 without a finished product, banking on founder credibility and a single contrarian bet: that shoppers prefer AI-generated lifestyle imagery over actual product photography.

Scene launched Wednesday on iOS as an invite-only feed of synthetic images showing people and environments that don't exist, embedded with links to real products from actual brands. You scroll through AI-rendered living rooms, AI people wearing real sneakers, AI cafés with real coffee makers. The images are fake. The checkout is real.

"You come to Scene to shop, not be entertained."

Bar-Tal's thesis splits shopping discovery into two modes: entertainment (TikTok, Pinterest) versus transaction intent. He's betting there's white space for a visual discovery tool where every scroll assumes purchase intent. No influencers. No user-generated content. Just synthetic imagery optimized for conversion.

The economics here get interesting fast:

  • Traditional e-commerce creative costs brands $500-5000 per lifestyle shot
  • AI generation drops marginal cost per image to near-zero
  • Scene can personalize imagery at scale without photographer scheduling or model contracts

But the real signal isn't cost efficiency. It's that Lightspeed funded this before the product existed. Lisa Han's investment memo praised Bar-Tal's "video AI research expertise" and "deep understanding of what consumers want" — which translates to: we're betting on the founder's technical chops to build generative models that feel real enough to convert browsers into buyers.

Pull quote from the model training challenge: The hard part isn't generating pretty images. It's generating images that make you want to buy the embedded product without triggering the uncanny valley response that screams "this is fake and I don't trust it."

The Implication

If Scene works, every DTC brand's creative team becomes an AI prompt engineer. If it doesn't, Lightspeed just paid $7M to learn that shoppers still need to see real products on real people in real contexts before they'll pull out their credit cards.

Watch the conversion metrics. If Scene hits double-digit conversion rates in its first year, expect every commerce platform to spin up an AI imagery layer. If it plateaus at "interesting but doesn't convert," we'll learn that shopping still requires some tether to physical reality — at least until the next generation grows up never expecting product photos to be real in the first place.

Sources

Business Insider Tech