Topaz brings image and video enhancement together in the browser

Topaz Labs has launched Topaz for Web, a cloud service that restores and enlarges images, enhances video up to 4K, creates slow motion, and converts SDR footage to HDR.

A video can now be enlarged, slowed down, and converted to HDR without installing Topaz Video or relying on the computer’s graphics card. With Topaz for Web, Topaz Labs is bringing its main image and video processing tools together in a browser-based application.

All processing takes place on the company’s servers. Users upload their files, choose a workflow, and start the render. The browser must remain open during the upload, but it can be closed once processing has begun. The result then appears in an online library shared by images, videos, and previous projects created with Astra.

This setup makes the company’s heaviest models accessible from a computer without a powerful graphics card. The tradeoff is that files must be transferred to the cloud, upload times depend on the internet connection, and video processing is billed through credits. The service is therefore not a direct replacement for Topaz’s local applications in every situation.

Topaz for Web divides its features into two main groups. The image section brings together tools previously spread across Topaz Image Web, Gigapixel, Photo, and several specialized services. The video section absorbs Astra’s features. Its previous interface will remain available until September 30, 2026, after which video enhancement will move to the new service.

Video processing covers three categories that can be used separately or combined within a single render. The first improves quality and increases resolution, such as taking 720p footage to 4K. The second generates intermediate frames to raise the frame rate or create slow motion. The third converts an SDR source into an HDR file intended for delivery, editing, or visual effects work.

Video enhancement is divided into Precise and Creative modes. The distinction determines whether the system should preserve the original content as closely as possible or generate details that are no longer present in the source.

Starlight Precise 2.6 is the main fidelity-focused model promoted by Topaz. It attempts to recover existing detail, reduce compression artifacts, and clarify faces, hair, fabrics, products, text, and logos. A five-level sharpness setting can soften the treatment when reconstructed textures begin to look artificial.

Content preservation remains central to this mode. A sign, recognizable face, or product label should not be replaced by a different interpretation. Topaz therefore recommends Starlight Precise for archival footage, close-ups, advertising, and sequences where the subject’s identity must remain stable.

Starlight Fast 2 targets larger workloads and shorter deadlines. Proteus offers a less expensive option for enlargement, noise reduction, and sharpening of older or heavily compressed footage. It exposes controls for compression recovery, noise, edges, halos, and blur.

Creative mode uses Astra 2. Instead of limiting itself to information still visible in the footage, the model generates new textures and structures. It is aimed at AI-generated video, animation, wide shots, crowds, landscapes, and material too degraded for faithful restoration.

A text field lets the user describe the expected content, while sliders control creativity and sharpness. This freedom can improve surfaces that have become indistinct, but it also changes the nature of the process. The result is no longer necessarily a faithful reconstruction of the original scene. Topaz acknowledges that faces, expressions, and small text may shift as the creativity setting increases.

That distinction matters for archives, documentary evidence, and scientific imagery. A detail generated by Astra 2 may look plausible without ever having existed in the original recording. The service’s before-and-after comparison tools make visual inspection easier, but they cannot establish whether a reconstructed element is historically accurate.

Frame interpolation and slow motion both use Apollo 8. For interpolation, the model inserts new frames between existing ones to reach 30, 60, 90, or 120 frames per second. For slow motion, those same intermediate frames extend the sequence without simply repeating existing views. Available slowdown factors range from 2x to 16x.

The two features listed separately in the announcement are therefore different applications of the same process. A higher frame rate preserves the length of the clip while making motion smoother. Slow motion increases its duration by distributing the newly created frames across a longer period.

SDR-to-HDR conversion is handled by Hyperion 2.5. The model takes 8-bit SDR footage and produces HDR output as 10-bit ProRes, 10-bit H.265, or a 16-bit EXR sequence. Topaz says it creates information in shadows, highlights, and colors instead of merely stretching the values already present.

Hyperion processes footage internally at 16-bit and uses the PQ color space for its video output. ProRes is intended for editing and color grading, H.265 for playback and delivery, and EXR for visual effects and compositing pipelines.

The model does not change the resolution or frame rate. It runs at the end of the processing chain and retains the dimensions of the result it receives. Producing a 4K, 60-frame-per-second HDR video therefore requires enlargement and interpolation before conversion. Hyperion works with Starlight and Astra, but not with Proteus in the same processing chain.

Topaz can run these operations in a single job. A 720p, 24-frame-per-second SDR source can be enlarged to 4K, converted to 30 frames per second, and exported as HDR ProRes. Each operation consumes its own credits, however, and the final cost is the sum of every model used.

The service currently accepts MP4 video uploads. Its documentation sets no fixed duration limit and cites a two-hour file as a possible example. A long video is still subject to upload time, processing queues, and, above all, the number of available credits. Several files can be submitted as a batch, but they are processed one after another.

The image side of