Overview
Jasna is a GPU-based application and command line for selectively restoring mosaics in videos and images. This series follows the chain from detection and tracking through clips, crops, masks, model choice, resource limits, selective ranges, special media, encoding, and diagnosis. Its central lesson is controlled change: use temporal or geometric context to generate a bounded patch, preserve surrounding source pixels where the workflow allows, and match the pipeline to hardware and output needs. The result may be coherent or more detailed, but it remains a model-generated reconstruction, not proof that hidden pixels were recovered exactly.
Learning outcomes
- Jasna's quality depends on selection, temporal tracking, spatial masking, resource fit, and output handling, not the restoration model alone.
- Local compositing limits model changes outside the selected mask, but it cannot make inferred detail equivalent to recovered source pixels.
- BasicVSR++ is the broad, fast path; LTX is a slower, supporter-enabled NVIDIA option whose value depends on compatible media, hardware, and time.
- Clip size, overlap, secondary restoration, and compilation trade continuity, spatial scale, VRAM, startup time, and throughput under stated conditions.
- Selected ranges, auto-split, VR180, still images, and streaming use different evidence and compatibility boundaries rather than interchangeable workflows.
Keywords
Preview every chapter · Unlock full text, audio, and video from chapter 4
PlansFAQ
How does Jasna restore only a selected mosaic region?
Jasna detects a region, expands a context crop, generates a regional patch, and blends it through a mask while leaving pixels outside the mask as source pixels. This is localized reconstruction, not proof of hidden detail.
Why can a Jasna result have seams or missed mosaics?
Missed mosaics usually begin with detection or thresholding, while seams can arise from tracking gaps, scene cuts, clip splits, overlap, masks, or encoding. A short scan or representative preview helps separate these causes.
Should I use BasicVSR++ or LTX in Jasna?
BasicVSR++ is the broad, fast, free path; LTX is a slower supporter-enabled NVIDIA path that may offer more detail when its hardware, media, time, and access requirements fit. Trial mode measures LTX speed but does not restore video.
Can Jasna recover the original pixels hidden by a mosaic?
No. Jasna generates a model-based reconstruction from visible crop and temporal or geometric evidence, so a convincing patch does not prove exact recovery of the hidden original.