AI Model Enhances Low-Quality Videos Into High Resolution in Real time

Topaz Labs, known for its advanced photo and video enhancement software, has unveiled a new AI model that automatically restores old footage, whether from personal home videos or aging archival content stored on traditional media.

In multiple examples, the AI significantly improves video quality by enhancing details and reducing noise and artifacts. This marks the first diffusion model designed specifically for video restoration, requiring no manual adjustments to refine the footage.

A Powerful AI Model

The company states that Project Starlight, the newly developed model, was built from the ground up with a unique architecture, featuring over 6 billion parameters and optimized for the latest NVIDIA hardware. For comparison, OpenAI’s GPT-4o—an advanced multimodal model capable of processing text, audio, images, and video, released in May 2024—operates with 8 billion parameters.

Unmatched Detail and Temporal Consistency

Topaz Labs claims that this model will “precisely restore details” while offering “exceptional detail recovery with unmatched temporal consistency.” According to the company, the key feature of this new model is its ability to enhance multiple frames simultaneously, ensuring high-quality restoration without motion artifacts or inconsistencies across frames and objects.

Diffusion models function by analyzing high-quality images, progressively adding noise to understand how they degrade. They also reverse this process, starting with a noisy image and predicting its original appearance before degradation.

Project Starlight will automatically denoise, deblur, upscale, and anti-alias videos on demand, making high-quality restoration accessible even to non-experts.

With over two decades in the imaging software industry, Topaz Labs has developed popular tools for photo and video enhancement. In 2020, a YouTuber used the company’s Gigapixel AI and other software to restore a mid-1890s silent film, presenting it in stunning 4K at 60 fps.

Restoring old video requires multiple steps, including upscaling, color correction and grading, frame interpolation, damage repair, and audio enhancement. While AI-powered tools exist for each task, human oversight is still essential for optimal results.

Competition in AI Video Restoration

Notably, Topaz Labs’ competitor Tensorpix also provides a machine learning-based video restoration tool aimed at improving legacy footage. However, its documentation does not indicate the use of diffusion models, which belong to a distinct category of machine learning techniques.

In the examples provided, Project Starlight performs impressively overall. The clips featuring an astronaut and a red parrot stand out as particularly well-restored. However, the boxing match footage falls short in fidelity, with some frames appearing smudged, resembling the flawed results seen in earlier AI-generated videos.

Topaz Labs states that users can restore videos up to 10 seconds long for free, while clips up to 5 minutes will be limited to 1080p resolution and require credits. An enterprise version will offer support for longer videos and higher-resolution outputs.

It’s unclear whether Project Starlight will run locally or be integrated into Topaz Labs’ existing apps.

For early access, interested users can engage with the company’s posts on X, Threads, or YouTube.

Read the original article on: New Atlas

Read more: Why Did Technology Fail to Prevent the Fatal Air Collision Over Washington?