A spin-off from the University of Florence, Small Pixels merges academic research with technological innovation to transform the audiovisual sector.
Leveraging artificial intelligence and pre-trained neural networks, its solutions enhance video quality, reduce operational costs, increase transmission efficiency, and minimise environmental impact.
Applications range from live broadcasting to gaming, from the enhancement of historical archives to remote production, optimising workflows without requiring modifications to existing processes.
The advanced integration of neural networks extends beyond broadcasting, enabling new applications such as video conferencing, aerospace image processing, action camera and drone footage management, and much more.
Below, we explore one of the key strategic applications of this rapidly growing company: upscaling to a higher video standard.
Other use cases will be examined later.
SPAIQ Scaleup: AI-Powered Video Upscaler
The transition from SD to HD or from HD to 4K involves an increase in resolution, enhancing sharpness and detail.
By employing multiple neural networks, Small Pixels’ intelligent upscaling generates the additional pixels required for SD content to be converted to HD or for HD content to be upscaled to 4K.
This process optimises quality without introducing the artefacts commonly associated with traditional upscaling technologies, which rely on “simple” interpolation mechanisms.
Unlike conventional methods, Small Pixels’ upscaling technology is based on pre-trained neural networks.
This means the neural network has analysed a vast dataset of videos in SD, HD, and 4K formats, learning over time how to intelligently and effectively generate the missing pixels to achieve higher resolution.
This is the key difference between Small Pixels’ upscalers and traditional ones.
The neural networks employed here are designed and trained to generate and add new pixels, enhancing resolution in a smart way rather than through the standard mechanical approach used by most conventional upscalers on the market.
Real-Time Processing and Compression Resilience
The software operates in real time with ultra-low latency.
Additionally, the upscaled signal is more resistant and resilient to the subsequent compression typically applied by the encoder in the final processing stage.
In the example diagram provided, the television signal is extracted from the playout in SD (in any format: SDI, 2110, SRT, RTPM, Mezzanine) and processed through Small Pixels’ software, which increases its resolution by doubling or tripling the number of pixels.
From there, the signal is output in an enhanced format, in real time, at the desired resolution and higher standard before entering the compression stage, where it undergoes encoding for transmission.
SCHEMATIC EXAMPLE – UPSCALING PROCESS
In an upcoming analysis, we will explore how increasing video compression by 30, 40, or even 50% does not degrade perceived quality but can actually improve it.
This enables the transmission of a signal that occupies significantly less bandwidth than the original—without compromises.
INFO: https://www.smallpixels.ai
© 2025 Small Pixels - PressPool PressOffice Roberto Landini
Synopsis
A spin-off from the University of Florence, Small Pixels merges academic research and AI innovation to optimise video production and transmission. Using advanced neural networks, its upscaling technology enhances resolution without interpolation artefacts, improving efficiency across broadcasting, gaming, archives, and remote production. Operating in real time with low latency, it also enhances compression resilience, enabling high-quality transmission with reduced bandwidth.
Tags:
#SmallPixels #AI #upscaling #broadcast #videoquality #compression #streaming #HD #4K #SD
