NVAI-open 0.5.22 adds a separate narrator track to its local video studio

NVAI-open 0.5.22 for Windows, and 0.5.4 for macOS, is a local-first studio for assembling whole videos from scenes, plus separate generators for images, video with sound, speech, music, sound effects, and textured 3D models from an image. The project positions itself as free, private, and without an account, subscription, or cloud processing. Everything runs on the user’s own graphics card, according to the README, after installing the app and downloading a model under the Models section. The change that is actually new in this release is narrow: a Conform function places the Studio’s narrator lines on their own audio track, labeled Voice. Later lines are added there as well, and the track can be removed like any other. Voice scenes now also list Qwen3-TTS as an available model. That builds on 0.5.21, which introduced narrator voice in the Studio using Qwen3-TTS, described by the project as Apache-2.0 and supporting 10 languages. A voice can be described in one sentence or taken from a recording, with a Same voice as option to keep one speaker across lines. The same speech capability appears in a new Speech tab and, the README says, for Claude Code. What it generates, and on what hardware The Studio builds a video from scenes, cuts it, and exports it using the graphics card encoder. Outside the Studio, the app lists image models including FLUX 3, Qwen-Image-2512, Krea 2, and MOSS-Image, with reference-image editing, inpainting, and extension. LTX-2.5 is listed for video with sound in one pass, including speech, at clips up to 20 seconds and up to 1080p. Music generation is listed at up to five