NVIDIA has officially announced PiD (Pixel-in-Diffusion), a new super-resolution solution that allows upscaling image resolution by 4X directly from the latent space of diffusion models.
Developments
The PiD technology is designed to operate faster than traditional upscaling methods by processing directly in the pixel space from latent models. Currently, PiD supports popular model families such as FLUX.1, FLUX.2, and Z-Image. Notably, NVIDIA also announced that support for Qwen Image will be added soon.
A live demo has been launched, allowing users to generate 4K resolution images using Z-Image, demonstrating the high-speed processing and detail-retention capabilities of this method.
Why It Matters
Increasing resolution from the latent space is a significant step forward, reducing the computational burden when generating high-quality images. For the Vietnamese AI community, PiD opens up opportunities to optimize workflows with next-generation models like FLUX, making 4K image generation more feasible and faster, even on mid-range hardware configurations.