Topaz launches Starlight Precise 2.5 with 4K upscaling for GenAI footage
Topaz says Starlight Precise 2.5 improves realism, cuts plastic-looking artifacts and upscales AI video to 4K in Astra, partner apps and API. Use it as a finishing pass when generated footage needs cleanup.
TL;DR
- Topaz has launched Starlight Precise 2.5, which the company says is built to make GenAI characters look more realistic while reducing the plastic sheen and artifacting common in generated footage, according to Topaz launch post.
- The new model is positioned as a finishing pass: Topaz's launch thread says it can upscale footage from recent GenAI video models to 4K.
- Availability is broad on day one. Topaz says Starlight Precise 2.5 is live in Astra plus integrations with HeyGen, Higgsfield, and ComfyUI, as shown in availability post.
- Topaz also says the model is already open to API customers, expanding it beyond the consumer app workflow described in Astra post and API update.
What shipped
Starlight Precise 2.5 is an update to Topaz's AI video enhancement stack aimed squarely at generated footage. In its launch post, the company frames the release around three promises: better realism, fewer synthetic-looking artifacts, and 4K upscaling for footage coming out of current GenAI models.
That positioning matters for creators because Topaz is not pitching this as a generator. It is a cleanup and enhancement stage that sits after image-to-video or text-to-video output, with Astra's product page describing Precise mode as the more fidelity-preserving option. Topaz's later API update confirms the same model is available to API users, which suggests the release is meant for both manual finishing and automated post pipelines.
Where creators can use it
Distribution is the practical part of this launch. Topaz says creators can start with Starlight Precise 2.5 inside Astra, and the same post also names HeyGen, Higgsfield, and ComfyUI as supported surfaces for the model.
That makes the tool easiest to read as a last-mile polish step for AI video workflows already happening elsewhere: generate clips in a video model, run a realism-focused pass, then deliver at higher resolution. The launch thread includes multiple short demo clips demo clip and demo clip, but Topaz has not published performance benchmarks or side-by-side measurement data in the evidence here, so the quality case is still based on the company's own examples.