About this app
IMPORTANT — REQUIRED DEPENDENCIES: Live Window AI requires Windows 10 or Windows 11 x64, a compatible NVIDIA CUDA GPU, an Internet connection for first-time installation of its local Python and PyTorch runtime and for each model download, and substantial free disk space. Model weights are not included in the Microsoft Store package.
Live Window AI uses live generative AI. It captures frames only from the application window you explicitly select, processes them on your PC and presents each completed transformation in a separate output window. A mandatory local safety gate checks the instruction, selected source frame and generated frame before display. A rejected result is discarded. Safety controls reduce risk but do not guarantee accuracy, suitability or legality, so every displayed image should still be reviewed before use.
Choose an open window, write an edit instruction, prepare the supported local model and start transforming. A latest-frame-wins queue keeps the interface responsive: at most one inference runs while one newer frame can replace the pending frame. This is designed for iterative visual previews, not for frame-perfect video processing.
Use fixed, custom or source-relative inference sizes. Present the result at the window size, 720p or 1080p with Nearest, Bilinear or Lanczos GPU reconstruction. Optional crossfade blends completed results. Experimental optical flow estimates motion from source frames and may introduce halos or incorrect occlusions.
The Microsoft Store release uses FLUX.2 Klein 4B for instruction-guided edits. Its reviewed model revision is pinned and downloaded separately from the provider into the local model cache. FLUX.2 Klein 4B is offered under Apache License 2.0. Installing Live Window AI does not grant rights beyond the model terms, and availability remains subject to the provider. Review the linked model card and licence before download or use.
Your selected frames, prompts and generated images are not uploaded to Vellora Studio One. During setup, the app connects to third-party distribution services to obtain Python, PyTorch packages and the model you select. Standard network metadata may be processed by those providers under their own policies.
Live Window AI does not record video, export a video file, create a virtual camera, integrate with OBS or Spout, provide native video-to-video diffusion, or guarantee a particular AI frame rate. The 15, 30 and 60 FPS controls set presentation timing; actual AI inference speed depends on the model, source, resolution, prompt and GPU.
Practical starting requirements:
• Windows 10 or Windows 11, x64.
• NVIDIA CUDA-compatible GPU. AMD and Intel GPU inference are not supported by this build.
• A high-VRAM NVIDIA GPU. FLUX.2 Klein 4B used about 15.3 GB allocated VRAM in the tested implementation; a 24 GB GPU is recommended.
• At least 35 GB free disk space for the private runtime, safety component, model and working headroom.
• A visible, non-minimized source window.
Performance figures shown in the app are local measurements, not promised results. Output quality, speed, text legibility and temporal consistency vary. The app has no account, advertising, cloud rendering or subscription.
Support and AI-output reports: support@vellorastudio.one or https://vellorastudio.one/support/index.html