Real-ESRGAN in Warlock Studio: upscaling settings

This Russian-language tutorial uses Real-ESRGAN family upscaling models through the Windows application Warlock Studio. This write-up is based on saved English translated captions. The source description identifies Real-ESRGAN; the translated model names are imperfect, so exact checkpoint spellings are not inferred.

The presenter downloads a GitHub release, selects image models and explains Input and Output Resolution with X4 scaling. He reports usable performance on a GeForce 1080 with 8 GB of VRAM; his RX580 crashes, which he suspects may reflect a faulty card. On a difficult low-resolution photo, restoration introduces artificial-looking facial detail. A brief fourfold-upscale comparison in Upscayl produces similar problems. These are his examples, not general performance rankings.

Other tests cover denoising and face restoration with limited visible improvement. Process Chaining combines actions and saves presets; the presenter says a separate output folder is needed. For a damaged VHS clip, he switches to BSRGAN. His 50/100 input/output settings produce the original resolution despite the preview, while 50/200 produces a twofold increase. That mismatch is reported from this test rather than presented as a universal setting rule.

The Keep Frames option retains PNG frames alongside an MP4. He suggests processing frames as images, renaming them with PowerShell and importing the sequence into a video editor. The full restoration batch is not completed. A final short RIFE demonstration explains frame interpolation and slow motion.