Google Research Unveils BlazeEdit for On-Device Mobile Image Editing

GoogleGoogle

Google Research announced BlazeEdit, an efficient, generalist image-to-image diffusion model designed for on-device mobile deployment. This model enables interactive image editing tasks like outpainting and relighting directly on mobile devices, addressing the computational and privacy challenges of server-side AI.

Google Research introduced BlazeEdit, a highly efficient, generalist image-to-image diffusion model tailored for on-device mobile deployment. This model consolidates multiple image editing tasks, including object removal, outpainting, tone correction, relighting, and sticker generation, into a single, compact architecture. Unlike larger models that often require server-side inference, BlazeEdit operates with only 195 million parameters.
Parameter Count
195M
Inference Time (Pixel 10)
290ms
Deployment
On-device mobile
Consolidated Tasks
Object removal, outpainting, tone correction, relighting, sticker generation
Workshop Acceptance
CVPR 2026 EDGE Workshop

The development of BlazeEdit addresses the high computational costs and potential privacy risks associated with massive parameter counts in modern diffusion models. By eliminating text-conditioning components for common editing tasks, BlazeEdit achieves a substantial reduction in download size and memory overhead. This design allows for a full inference pass in just 290 milliseconds on a Pixel 10, delivering fast and private image editing.

BlazeEdit provides a seamless, privacy-preserving, and lightning-fast experience for generalist image editing directly on mobile devices. This aligns with Google's broader strategy to enable efficient, local AI capabilities, as seen with On-Device Agent Skills for Offline Gemma 4 Workflows.

Google booth showcasing computer vision and depth prediction capabilities to conference attendees at a trade show.
View the full update on arxiv.org
Google Research
Google Research
@GoogleResearch
X

Check out BlazeEdit at our kiosk (booth #557) at 12 PM! A highly efficient, generalist image-to-image diffusion model tailored for on-device deployment. See interactive outpainting, relighting & more on mobile! #CVPR2026 https://t.co/nrdlJpU0pH https://t.co/OR0e1Slb4Q

6retweets45likes
View on X

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Share this update