A spin-off of the University of Florence, Small Pixels merges academic research with technological innovation to reshape the audiovisual sector.
By leveraging artificial intelligence and pre-trained neural networks, its solutions significantly enhance video quality, reduce operational costs, improve transmission efficiency, and minimize environmental impact.
Applications span live broadcasting, VOD, gaming, historical archive enhancement, remote production, and video surveillance, optimizing workflows without requiring modifications to existing processes and infrastructure.
In a previous feature, we explored Small Pixels’ video upscaling technology, which employs pre-trained neural networks to upscale SD to HD or HD to UHD, intelligently generating additional pixels without the artifacts typical of interpolation-based techniques.
[Link to article]
Today, we examine another key strategic application of this fast-growing company: video archive restoration and “revamping.”
Unlocking the Potential of Historical Archives
Digital archives play a crucial role in preserving cultural and informational heritage, yet technological obsolescence, media degradation, and the complexity of multiple analog and digital formats risk rendering them inaccessible.
Ensuring long-term security and accessibility requires advanced migration and data updating solutions, sustainable formats, modern quality standards, and appropriate storage systems.
Dedicated software and hardware solutions exist to restore the treasures of the past efficiently. The restoration of historical video or film archives generally involves two main phases.
The first is a thorough review of the material, focusing on targeted interventions exclusively on deteriorated, damaged, or compromised sections, often requiring frame-by-frame work.
The traditional approach to recovering film transferred to video is a time-consuming, almost entirely manual process, feasible mainly for historically or cinematically significant works.
For large-scale projects, a more generalized enhancement approach is often preferred, relying on automated or semi-automated processes. However, these methods are frequently superficial and unsuitable, particularly when dealing with extensive archives, such as long-running TV series.
Today, valuable audiovisual archives have a new opportunity.
A New Approach: Revamping Instead of Restoration
“In Small Pixels,” explains CEO Fabio Clabot, “we prefer the term ‘revamping’ over ‘restoration.’ We do not focus on artistic restoration or content remastering but rather on the renewal, modernization, and enhancement of video material. This is how we bring our clients’ historical archives back to life through our intelligent software.
The manual intervention required is minimal: one of our experts simply analyzes the material and configures one or more neural networks to adapt to the content type and its specific flaws. This approach enables archive owners to reposition their historical content strategically, updating it to the quality standards required by modern distribution networks with minimal effort.”
Nobody Does It Better
A prime example is the work of a major broadcaster revamping two well-known television series originally broadcast in SD over 20 years ago.
- The first is an Italian period drama that aired from 2003 to 2005.
- The second is a sitcom, mostly shot indoors, that aired from the late 1980s until the early 2000s.
Small Pixels’ scientists selected the optimal neural network for each series, given their different visual characteristics and the need for HD upscaling. [Link to press release]
By choosing the most effective settings for each, the goal is to achieve the best possible results. Once a neural network is determined for one episode, the same parameters can be applied across the entire series, making the revamping process both fast and cost-effective.
“Our pre-trained neural networks offer a wide range of adjustable parameters,” Clabot continues. “Moreover, we can improve footage by processing it through multiple neural networks in sequence. For instance, we might start with a noise reduction network to eliminate high-frequency video noise. Unlike traditional software that flattens the image, our AI preserves intricate details, maintaining the authenticity of the content.
Once this initial step is complete, a second intelligent network handles upscaling, further refining the material. A final processing pass can then adjust the overall look and feel to match the client’s desired aesthetic.”
The entire revamping process operates at speeds between 2x and 5x real-time playback. Training on a sample set of material is performed only once, after which processing becomes fully automated for entire seasons or similar content. This streamlined approach underscores the remarkable efficiency and scalability of Small Pixels’ revamping service, offering a cost-effective solution for large-scale video archive enhancement.
© 2025 Small Pixels – PressPool PressOffice Roberto Landini
INFO: https://www.smallpixels.ai
Tags
#AI #VideoRestoration #Revamping #NeuralNetworks #Broadcast #VOD #ArchivalPreservation #FilmRestoration #MediaInnovation #TVSeries