Sub-second speed and ultra-low cost without sacrificing quality.
Performance
State-of-the-art AI image in 1s for $0.010.
Adherence
Faithful image generation from controllable prompt instructions.
Text Rendering
Precise, styled text rendering without mispelling.
Multi-Image Editing
High control with multiple reference images.
Fundamental techniques for prompting P-image Edit.
P-Image-Edit delivers quality on par with SoTA models, but it is 3x cheaper and 30x faster.
Acknowledgments
The launch of P-Image-Edit is a milestone for real-time, high-quality image generation. But it is not a journey we took alone.
We want to acknowledge the teams whose work defined the state-of-the-art, introduced the proper techniques, and set the high bar for quality that we aimed to match.
Black Forest Labs (BFL): Their work on the Flux series.
Hugging Face for their work on Diffusers and Transformers
Alibaba: Their work on Qwen Image, Qwen Image Edit, Z-Image and Wan Video
The Research Community: For the fantastic work on countless open-source initiatives and optimisation algorithms.
Pruna is a member of the Coalition for Sustainable AI.
Pruna co-authored "Key challenges for the environmental performance of AI".
Code Carbon is integrated into Pruna.