Topaz Labs adds Starlight Precise 2.5, Gaia 2, and five API models

• Topaz’s March “Precision” update adds Starlight Precise 2.5 for making GenAI video look more realistic, Gaia 2 for animated-video upscaling, and local Wonder 2 support for AMD GPU users, according to Precision update.
• The bigger creator-facing change is in Topaz’s API: API launch says all Starlight, Astra, and Bloom models are now available there, alongside five newly added models.
• Those five API additions are Starlight Precise 2.5, Gaia 2, Starlight Fast 2, Starlight HQ, and image Background Removal, as shown in the model card.
• Topaz’s release details also says premium models are no longer enterprise-gated, and the company has reorganized pricing and documentation to make cost forecasting easier for production use.

Topaz framed this release around “precision,” but the practical split is simple: Studio gets new video-enhancement models, and local image users on AMD get broader model access. The headline additions are Starlight Precise 2.5, pitched for improving realism in GenAI video, and Gaia 2, described as a new frame-by-frame approach for upscaling animated footage launch thread.

The update also adds local Wonder 2 support on AMD GPUs, which matters for creators who want image enhancement without a cloud round-trip launch thread. Topaz also pointed to an early downstream use case: its repost of HeyGen’s announcement says Starlight Precise 2.5 is already being integrated into avatar-video workflows HeyGen repost.

For developers and creative-tool builders, the API update is the bigger shift. Topaz says its full Starlight, Astra, and Bloom catalog is now exposed through the API, rather than reserving some higher-end models for enterprise access API launch. The five newly added API models cover two different needs: quality-first enhancement with Starlight Precise 2.5, Gaia 2, and Starlight HQ, plus a speed tier in Starlight Fast 2 and a utility tool in Background Removal model list.

The linked release notes say pricing has been regrouped by model family, image costs are standardized per megapixel, common video jobs now use fixed pricing tables, and a credit calculator has been updated for forecasting. Topaz also published new developer docs, and its own example pairing — “Seedance 2.0 + Starlight Precise 2.5” — suggests the intended workflow is not just restoration, but polishing AI-native video after generation workflow example.