OpenAI Image Gen 2 Turns Billboards Into Labs

OpenAI Image Gen 2 was demoed reimagining large-format US billboards from ordinary street-view scouting photos. No vulnerability, affected system or attacker requirement was described. It is a product workflow demo according to OpenAI Youtube.

StartupHub data

Companies working on this

Profiles of the companies named in this story, with funding and a one-liner from our database.

OpenAI

An artificial intelligence research organization developing and promoting friendly AI for the benefit of humanity.

OpenAI

$13.0B

Artificial intelligence research and deployment company focused on developing advanced AI models like GPT-5.6 and GPT-Live, offering products such as ChatGPT and an API platform.

OpenAI

Private / $100B+ est

OpenAI is an AI research and deployment company dedicated to ensuring that artificial general intelligence benefits all of humanity.

OpenAI

$190.6B

An AI research and deployment company building safe and beneficial artificial general intelligence.

The team called the model best of breed for fonts and multilingual rendering, able to visualize anything in the mind's eye. They used it to take ordinary ad spots and make them surreal.

The full discussion can be found on OpenAI Youtube's YouTube channel.

The first test was perspective correction. A location team shot a billboard at an off angle and prompted the model for a straight-on view of the building. The result kept the full billboard space legible.

Next came concept iteration on ChatGPT Images. The blank billboard became matching brick, then bricks falling away to reveal a portal. The same pipeline later rendered Chicago's iconic L train emerging mid-air from a facade, shown as a final composite on site.

That last example mirrors the Tennis :30 video generation test in our directory, where prompt-driven variants replaced manual selects. The pattern is consistent: models are shifting from single asset generators to controllable scene editors.

How the billboard actually gets built

It reprojects a crooked scouting photo into a clean frontal facade, then inpaints over the ad with geometry-matched brick and peels it open.

Think of a scout, retoucher and matte painter compressed into one prompt loop. You ask for four high-resolution variants and keep iterating instead of reshooting.

What this gets right, and what it doesn't

For creative teams, iteration cost collapses so a whiteboard of ideas can be tested rather than narrowed to three early bets. For builders, that means integrating human art direction with prompt versioning and approval gates.

What the demo doesn't show is reliability at billboard scale for font fidelity, multilingual copy, rights for locations, or brand safety checks. Ship with human review and file provenance before production.

The value here isn't a new ad. It's the speed to reject bad ones.