The report
OpenAI's November 4 Images 2.6 release introduces Flare, optimized for faster generation, and Sunburst, optimized for finer detail and greater consistency across editing rounds. OpenAI reports about 50% lower generation latency relative to its prior model generation; some low-quality Flare API requests can finish in six seconds. These are provider performance statements, not independently matched speed measurements.
The new options include GIFs assembled from generated spritesheet frames, localized editing markup through the API and SVG export for design workflows. Those functions should not be conflated. A GIF loop can preserve the same drawn character across a short sequence while lacking the motion continuity of general generated video. An SVG output can be useful downstream without guaranteeing independently editable layers for every element. Neither function proves identity verification or the survival of provenance markers after export and transformation.
The AI-security question concerns the cost and consistency of generating convincing identity representations, not an already documented scam attributable to this release. A reproducible study would use consented reference identities, repeat multiple edit rounds and export formats, then check both similarity and whether provenance survives realistic conversions. The Images 2.6 entity and deployment card describe the product and vendor safeguards. The deepfake account-recovery case is a separate observed impersonation route, not proof that it used this image model.