Luma Adds OpenAI's New Image Models to Its Agents

Luma says two unreleased-sounding OpenAI image models, Flare and Sunburst, are now built into Luma Agents, letting users edit images and push results straight into video generation.

Luma Labs says it has plugged a pair of new OpenAI image models directly into its Luma Agents platform, giving users an image-editing step that feeds straight into video creation. The announcement came from Luma's own X account, which is currently the only place describing these models — there's no independent confirmation yet from OpenAI or elsewhere. According to the post, the two models are called GPT-Image-2.5, split into a fast option named Flare meant for high-volume, quick-turnaround generation, and a more precise option named Sunburst aimed at edits that need to be exact rather than approximate. Luma frames the pitch around control: users can bring in a reference image, correct the specific part that's wrong, and preserve everything else in the frame. The stated workflow goes beyond a single edited image. Luma says the corrected result can then be carried directly into video generation within Agents, positioning the tool as a bridge between still-image touch-ups and motion output rather than a standalone image editor. Because the only available source is Luma's own promotional post, details on pricing, availability limits, or how these models differ technically from other OpenAI image tools haven't been independently verified. For now, the claim stands as Luma's own characterization of what shipped in Agents today.