VideoProc Converter
New AI features are available
Listed in Video Super Resolution approaches have shown impressive results since their emergence. On April 18, 2024, Adobe announced VideoGigaGAN, sparking a wider discussion. Adobe claims this new generative VSR model is superior to previous video upscaling methods as it generates videos with more details but no AI artifacts.
Is VideoGigaGAN a new game changer to AI super resolution? How to use it? Let's take a closer look.
VideoGigaGAN: 8x Video Upscaling
VideoGigaGAN is a cutting-edge AI model from Adobe Research for Video Super Resolution (VSR). Built on Adobe's GigaGAN, an effective image upsampler, it aims to bring image-level detail enhancement into the video domain.
The key challenge in video upscaling is adding realistic details while keeping the video smooth and consistent over time. According to the research project, VideoGigaGAN can upscale low-resolution, blurry videos up to 8x their original resolution, producing rich new details while no flicker or distortions between frames.
Key Features and Innovations
Exceptional Detail Enhancement
Most notably, VideoGigaGAN can upsample a video up to 8x its original resolution, adding plausible high-frequency details that look natural.
The project page shows a demo clip of ants crawling on dirt in a very low-res video. VideoGigaGAN's 8x upscale transforms this clip so you can actually discern the ants' shapes and the texture of the soil. Typically, tiny objects would blur out, but here the model adds clarity. You can see the ants' legs and distinct outlines where the original was just a few pixels.
High Temporal Stability
The model strikes a great balance between detail and consistency, showing minimal flicker, ghosting, or jitter.
From the demo showcase, we saw it turn the tiny 128x128 video into a 1024x1024 HD footage. The person's skin shows pores and natural texture, and individual hairs (eyebrows, facial hair) become visible. The added details look organic and not like obvious CGI, making it hard to tell AI was involved. Importantly, as the face moves, those fine details do not shimmer or disappear. They remain consistently placed, which gives the illusion of a genuinely high-resolution video.
Fast Generative Upscaling
As mentioned, VideoGigaGAN's generator architecture is inherited from GigaGAN. Unlike diffusion-based upscalers that require slow iterative refinement, GAN architecture produces results in a single pass, making it much faster at inference.
Thanks to this, VideoGigaGAN can upscale frames in real-time or near real-time on modern hardware. The research paper notes that despite a large model size, it achieves competitive inference speeds vs. prior feed-forward models. This efficiency opens the door to practical uses like possibly real-time video enhancement or integration into editing software without unbearable render times.
Limitations
Research Preview Only
The most immediate downside is that VideoGigaGAN isn't available for public use yet. As of now (May 2025), VideoGigaGAN remains a lab project. Adobe showcased it as a research prototype. While rumors suggest potential integration with Adobe Premiere Pro or After Effects, there's no consumer release guaranteed.
Long Video Challenges
VideoGigaGAN struggles with very long videos or sequences. The paper notes a drop in performance beyond ~200 frames (about 8–10 seconds of footage). Over long durations, small errors accumulate and the optical flow guidance can become inaccurate, leading to mis-propagated details or flicker over time.
In essence, the model's temporal consistency techniques have a limited "memory" window. This makes it less suited for upscaling an entire movie in one go. It might work best on shorter clips or by processing long videos in segments. That might be one of the reasons why it has not yet been released as a marketable product.
Struggles with Tiny Details
Like many AI upscalers, VideoGigaGAN has trouble reconstructing very small or fine features that are heavily lost in the low-resolution input. If the original video frame lacks almost all information about a tiny object, the AI can only guess – often poorly. This can lead to weird artifacts.
For instance, in one demo a girl's teeth became "AI-generated horror teeth" when upscaled, appearing unnaturally distorted. So, while large textures and broad details upscale nicely, sub-pixel details can be misinterpreted, producing gibberish text or odd patterns.
Hallucinated Details and Potential Artifacts
It's worth noting that the ultra-sharp details VideoGigaGAN creates are not real. They are AI-invented based on training data. In most cases this is a pro, but it can be a con in scenarios where authenticity matters. For instance, using it on security footage or forensic video would be problematic because it might alter visual evidence. Even for creative work, there's a minor risk of uncanny artifacts, such as overly smooth skin or textures that look too perfect.
That's probably a common problem with all AI models. Although VideoGigaGAN claims that there's "no AI weirdness", we should always keep in mind that the output is an AI's best guess, not a magic restoration of actual lost pixels.
How to Use VideoGigaGAN
Actually, there's no way to use VideoGigaGAN directly for now. Adobe showed it as a tech demo, not yet a product. It's not in Photoshop's Neural Filters, not in Premiere, nor available via Adobe's beta features, and there's no guarantee it will ship in consumer software.
VideoGigaGAN Alternative
Since VideoGigaGAN isn't available for everyday users yet, those who need to upscale or enhance videos today will turn to existing tools. One highly recommended, user-friendly option is VideoProc Converter AI, an all-in-one video processing software that includes an AI-powered upscaler and enhancer. Unlike VideoGigaGAN, VideoProc is a mature product you can download and use on your PC or Mac right now.
VideoProc Converter AI has a built-in Super Resolution module. It leverages several AI models to improve video resolution, sharpness, and overall quality. This AI video upscaler can increase resolution by up to 4x (400%), for instance, converting a 480p video to 1080p or 1080p to 4K, while preserving and enhancing details. Its AI models have been trained to enhance details while minimizing artifacts, so you can see improved textures and reduced noise. In addition, the program also offers features like denoising, deblurring, colorizing, and frame interpolation as part of the package. The tool is designed with a user-friendly interface, so you don't need technical expertise to use the AI features.
While it may not yet hit the extreme detail of VideoGigaGAN, VideoProc Converter AI will absolutely make your videos look sharper, clearer, and more detailed, which is what most of us need.
