Bringing product accuracy to generative AI

Omi has developed new technology that combines generative AI with 3D Digital Twins to keep products accurate in AI-generated visuals. Accelerated by NVIDIA, this technology will power an upcoming feature, with more details to be shared at NVIDIA GTC Berlin.

“The product is at the heart of the brand. Its appearance cannot be approximate. We’ve brought AI and 3D together so brands can create entirely new worlds while keeping their products faithful to reality. NVIDIA’s accelerated computing makes this technology faster, helping teams bring that creative freedom to production at scale. We look forward to sharing more at GTC Berlin on October 21.”

Hugo Borensztein, CEO & Co-Founder, Omi

Why is product accuracy important at scale?

Generative AI makes it possible to imagine new visual worlds. But it can also alter the product it depicts: the shape of a bottle, the color of a lipstick, or the text on a label. The image may look realistic, yet represent a product different from the one sold in stores.

For brands managing hundreds or thousands of SKUs, maintaining this accuracy becomes a major challenge. A campaign may cover dozens of shades, multiple formats, and different versions depending on the market.

Products also evolve over time: packaging changes, labeling is updated, and new variants are introduced. Teams need an approved, up-to-date reference that keeps their content aligned with the products they sell.

Omi’s Digital Twins provide this product source of truth: accurate 3D representations that teams can reuse across campaigns, markets, and creative tools.

Combining AI creativity with 3D product accuracy

Combining AI-generated scenes with realistic 3D Digital Twins requires substantial computing power. NVIDIA technology supports the scene analysis, image generation and rendering behind ProductDrop, helping teams move from creative exploration to production across large product ranges.

Omi's upcoming feature is built on NVIDIA OptiX and NVIDIA CUDA, combined with open-source image understanding and generation models including the FLUX family and Qwen-Image, running on NVIDIA RTX PRO 6000 NVIDIA Blackwell Server Edition GPUs. Omi's pipeline also integrates open-source models such as MoGe for depth estimation and Microsoft TRELLIS for 3D asset generation.

Optimized on NVIDIA GPUs and served through a custom inference endpoint on Verda, a European full-stack AI cloud, TRELLIS inference runs 10x faster, cutting overall generation pipeline latency by 4x.

“Our AI Lab's work with Omi shows what's possible when inference serving is optimized specifically for a workload, in this case cutting TRELLIS latency by 10x. Speed shouldn't come at the cost of quality, and that's exactly the kind of problem our applied research is focused on solving,"said Paul Chang, AI Infrastructure Lead at Verda.

What’s next?

Omi will share more about the new feature and the technology behind it at NVIDIA GTC Berlin on October 21 at booth #7059.

“The product is at the heart of the brand. Its appearance cannot be approximate. We’ve brought AI and 3D together so brands can create entirely new worlds while keeping their products faithful to reality. NVIDIA’s accelerated computing makes this technology faster, helping teams bring that creative freedom to production at scale. We look forward to sharing more at GTC Berlin on October 21.”

Hugo Borensztein

CEO & Co-Founder, Omi

Press enquiries

For an interview with Omi’s founders, a ProductDrop demonstration or high-resolution images showing the workflow, contact Agence Ballou at omi@balloupr.com.

About Omi

Omi is the Product Source of Truth for consumer brands. Its platform generates product photos, videos and AI campaign visuals from approved 3D Digital Twins. Teams can scale content across entire catalogs and regenerate affected visuals when packaging, variants or market requirements change.