Topaz adds Starlight Precise 2.5 and five API models in March Precision update

• Topaz's March "Precision Update" centers on Starlight Precise 2.5, a realism-focused video upscaler that the company says improves faces, texture detail, and text clarity in AI-generated footage Topaz March update Precision update link.
• The release also adds Gaia 2 for frame-by-frame animation upscaling, local Wonder 2 support on AMD GPUs, and a broader Studio refresh built around new video model families Topaz March update Precision update page.
• On the API side, Topaz opened all Starlight, Astra, and Bloom models to developers, then added five more models on top: Starlight Precise 2.5, Gaia 2, Starlight Fast 2, Starlight HQ, and image Background Removal Topaz API release API docs and release links.
• The API release also reorganizes pricing into model families, updates the credit calculator, and revamps the docs and playground so costs map more directly to output size and common resolutions API docs and release links Release notes.

You can jump from the official announcement to the Precision update page, then into the new API docs and the full release thread. Topaz also slipped in two distribution clues: a HeyGen partnership for Starlight Precise 2.5 HeyGen partnership RT, and a Discord invite for developers who want to steer the API roadmap Topaz API release.

Topaz framed the whole March drop around precision, and the flagship model is Starlight Precise 2.5. In the API release notes, the company describes it as an upgrade for AI video realism with better faces, texture detail, and text clarity, while the main update page positions it as part of a new video model family push across local and cloud tools.

That sits alongside Gaia 2, which Topaz describes as next-generation frame-by-frame 2x upscaling for animation, aimed at a very different creative workflow than the realism push behind Starlight Precise 2.5 Release notes.

The API launch is the denser part of the story. Topaz says developers now get access to all Starlight, Astra, and Bloom models, plus five newly added models.

The five additions are:

1. Starlight Precise 2.5, upscale to 4K with improved realism.
2. Gaia 2, frame-by-frame enhancement for animated video.
3. Starlight Fast 2, 2x faster diffusion upscaling with higher fidelity.
4. Starlight HQ, maximum-quality enhancement for modern high-resolution video.
5. Background Removal, still-image subject cutout with transparent output.

The release notes add one important access change: Bloom and Starlight models were previously enterprise-only, and are now available to all API customers at the same price Release notes.

Topaz also reworked how the API is packaged. Instead of model-by-model credit costs that vary by job, the company says image models now bill by output megapixels within a model family, while video models use published pricing tables for common resolutions and durations Release notes.

The new developer docs now organize the catalog into Precision, Generative, and Creative families, and the credit calculator and playground were updated to match the expanded catalog. For creators and tool builders, that is the practical part of the launch: more models, fewer gated tiers, and a pricing sheet you can actually estimate from.

One useful detail in the March thread has nothing to do with the API. Topaz says AMD GPU users now get local support for the Wonder 2 image model, a small but concrete expansion for creators who want to stay off Nvidia hardware Topaz March update.

Topaz also boosted a HeyGen integration for Starlight Precise 2.5, suggesting the model is already moving into downstream creation tools rather than living only inside Topaz's own apps.

That combination, broader local hardware support on one side and embedded distribution on the other, makes this update feel less like a single model drop and more like Topaz widening the lanes into finished creative workflows.