SeedVR2 in Wan2GP: tutorial and Topaz comparison

Learn SeedVR2 upscaling and batch processing in Wan2GP, with 12 or 16 GB VRAM guidance and comparisons against Topaz Starlight Mini and Precise.

Player not loading? Watch on YouTube

David Clark demonstrates SeedVR2 video upscaling in Wan2GP through Pinokio, then compares the results with Topaz Starlight Mini and Precise. He describes SeedVR2 as a free local AI option. Installation is covered in a separate video; this tutorial starts with updating and launching an existing setup.

Clark recommends an Nvidia GPU with 12 or 16 GB of VRAM, based on his experience with a 3060 and a 5060 Ti. He says his 8 GB 2080 works less well. Initial processing requires model downloads, which he notes can reach 20 or 30 GB for some models. Processing time depends on the computer and output resolution, and the progress estimate fluctuates as chunks complete.

The walkthrough covers source paths, scaling presets and output settings. SeedVR2 uses scale factors rather than a specified output size. Batch mode accepts a folder pattern, but its settings do not carry over from single-file mode. MOV output needs an explicit destination. Clark chooses ProRes 422 and reports audio compatibility problems with some editors.

His comparisons use old Hi8, VHS-C and low-resolution camera footage. He generally prefers Starlight Precise for detail, but favors SeedVR2 over Mini on some countryside shots. SeedVR2 leaves ghosting in the VHS-C example. Clark also says its developer has stopped development; the video does not independently establish that status.