Repository navigation
Releases
tapewake 0.2.1 (Windows): upscaling on AMD and Intel graphics
There's no installer to download yet: build it from source with BUILDING.md (git checkout windows-v0.2.1). Full history in CHANGELOG.md.
Released for: Windows (windows-v0.2.1). This is the first Windows release since 0.1.5, so it also brings
everything in 0.2.0 to Windows. Mac: not released, the latest Mac release is still 0.2.0. Nothing in this version
changes what happens on a Mac.
Upscaling on AMD and Intel graphics
Why: until now the Windows version needed an NVIDIA card. Many PCs have an AMD or Intel card instead.
What changed
- On a PC without an NVIDIA card, the upscaler and the flicker smoothing (RAFT) run on any DirectX 12 graphics card
through ONNX Runtime with DirectML (ort_models.py). The first upscale exports both models to ONNX once
(saved inmodels\onnx). The rest of the upscale step stays on the processor: it's light next to the models. - torch-directml was tried first and isn't used: it crashes on the upscaler's PReLU layers and inside RAFT.
- Measured on a 180-frame tape clip, running through DirectML on the test PC's card: 4.0 frames a second, against
6.7 with PyTorch on CUDA and 0.4 on the processor alone. The picture matches the CUDA one (66.8 dB PSNR). - Without
onnxruntime-directmlinstalled, or if it fails to load, the upscale step falls back to PyTorch on the
processor exactly as before (slow, but it works). - The setup checks and Settings → Performance list the graphics card whatever its make. An AMD or Intel card
passes the check when DirectML is ready, and the check says what to install when it isn't. - BUILDING.md has the setup for AMD and Intel cards: the processor build of
PyTorch,onnxruntime-directmlandonnx.THIRD_PARTY.mdlists both.
Not changed yet: on AMD and Intel cards the face models still run on the processor, which is very slow (turn
Face restoration off for long videos), and the heat pause only reads NVIDIA cards.
0.2.0 on Windows
The heat pause and the computer details in Settings (both from 0.2.0) were checked on a Windows PC for this release:
Settings shows the processor, graphics card, memory, what the AI models run on, free space and the card's
temperature read through nvidia-smi. A full tape-mode restoration of a short clip with faces and sound on finished
with every check passing. The card never got hot enough to trigger a heat pause during testing; that part was
tested on a Mac in 0.2.0.
tapewake 0.1.5: clean-up after a full check
There's no installer to download yet: build it from source with BUILDING.md. Full history in CHANGELOG.md.
Clean-up after a full check of the repository and the installer
Why: a file-by-file check of everything in the repository and the installer, its dependencies and their
licences.
What changed
- The installer carried 24 Python cache files (
__pycache__) left in the engine folder by test runs on the
build computer. Nothing needed them. Every build now deletes them first (scripts/clean-engine.mjs). THIRD_PARTY.mdlists the Rust crates' licences accurately (not only MIT / Apache-2.0: also Unicode-3.0,
MPL-2.0, Zlib and BSD-3-Clause, all permissive, none GPL) and the helper libraries the face and sound models
need (all permissive). Shipping the crates' licence texts with an installer is on the release checklist.- Two screenshots that nothing used were removed.
Nothing else changed: the app and the restoration work exactly as in 0.1.4.
Known limits (unchanged since 0.1.4)
- Only Face restoration and Audio can be switched; other settings and presets (including Animation)
don't change the restoration yet, and tapewake won't start until they're at the tested values. - Tape mode is for NTSC tapes (29.97 fps) only; PAL tapes can't be restored yet.
- Progressive mode has only been tested with face restoration off. Animated films get the general upscaler
and a light film grain. - Try a short clip before a whole film. Details in the README.
tapewake 0.1.4: progressive video mode
There's no installer to download yet: build it from source with BUILDING.md. Full history in CHANGELOG.md.
Progressive video mode
Why: every restoration used the home-video tape settings, which deinterlace the picture. A video that
is already progressive (each frame a whole picture, like a DVD rip of a film) has nothing to deinterlace, so
it failed: a 720×462 rip stopped with "SeparateFields: clip height must be mod 2", and its frame rate would
have been changed to the tape's 29.97 fps.
What changed
- New Progressive video · Tested preset. When you choose a video, tapewake reads its scan type, frame
rate and size, picks the preset that fits and says so under the preset menu. - Progressive mode skips deinterlacing and keeps the video's own frame rate (23.976 stays 23.976). It works
at a square-pixel size of its own (a widescreen 720×462 DVD frame becomes 856×464) and restores at twice
that. Everything after deinterlacing is the same as for tapes: the same gentle denoise (DFTTest), the
Real-ESRGAN upscaler with flicker smoothing, sharpening and grain. - Face restoration and sound restoration start off in this mode. Switching Audio off keeps the original
sound as it is (one track). - The home-video tape mode is unchanged: on a tape clip, the cut and deinterlaced pieces are identical,
frame for frame, to 0.1.3.
Settings that can't work are explained before anything starts
Why: the app let you choose settings and presets that the restoration doesn't use yet, and nothing
said so.
What changed
- Under the preset menu, a note says what the video is, which mode it gets and what the result will be,
and lists any setting that can't work, in red. - Pressing Restore with such a setting opens a window that explains each one and offers Use the tested
settings for this video. Nothing starts until the settings can work. It catches: Deinterlace on for a
progressive video or off for an interlaced one, interlaced videos that aren't 29.97 fps, presets that
aren't available yet, Enhance with AI off, and any other step switched or changed from the tested values. - The restoration engine refuses the same mistakes, with the same explanation.
Fixes
- The step switches only changed how they looked: switching Face restoration off didn't reach the
restoration in 0.1.3. They now do. - A running restoration with face restoration off showed an empty preset menu.
tapewake 0.1.3: stop and delete a project
There's no installer to download yet: build it from source with BUILDING.md. Full history in CHANGELOG.md.
Stop and delete a project
Why: there was no way to stop a restoration for good or remove a project, for example one started
with the wrong settings. Pause was the only option, so the project stayed in tapewake.
What changed
- New red Stop and delete project button (Delete project once it has stopped) under each
restoration's progress. Deleting takes three clicks: the red button opens a red confirmation window
that says how many GB of working files go, you tick "I understand ... can't be undone", and only then
does the red Delete project button work. - It stops the restoration if it is running (freeing the graphics card), deletes its working files and
any half-written restored video, deletes the player cache and removes the project from tapewake. - Kept: the original video, a finished restored video and saved samples. Not offered for restorations
that run from their own scripts.
Every project gets its own working folder
Why: a second project for the same video used the first one's working folder, so the two would
have overwritten each other's files.
What changed
- New projects get a folder no other project uses:
<name> - tapewake work (2),(3)and so on. - For projects made before this update that share a folder, Delete only removes that project from
tapewake and keeps the files, because the other project still needs them.
The progress screen shows the steps a project really runs
Why: it always listed finding faces, restoring faces, upscaling and assembling, whatever was chosen.
What changed
- A project started with Face restoration switched off now restores without faces, and its progress
shows only upscaling and assembling. Face restoration is the first setting the app passes on to the
restoration; the others still use the tested values (see the Status section of the README).
Progress no longer looks stuck at the start
Why: before the first piece starts, tapewake prepares the sound, cuts the video into pieces and
deinterlaces the first ones. That can take a while (the sound alone took about 8 minutes for a two-hour
film), and the progress screen showed 0% and no steps the whole time.
What changed
- A Preparing stage shows these three steps, each with its own percentage, and they fill the first
2% of the overall bar. The sound step now reports after every block of audio. - If cutting or deinterlacing fails during the preparation, the progress screen says so in red straight
away (with the error), instead of looking stuck until the sound is finished.
tapewake 0.1.2: the Restoration finished prompt
There's no installer to download yet: build it from source with BUILDING.md. Full history in CHANGELOG.md.
A "Restoration finished" prompt
Why: when a restoration finished, nothing told you. The job just sat in its tab, and you had to
find the file and tidy up by hand.
What changed
- When a job finishes, tapewake asks once: Finish and show the file or Later.
- Finish opens File Explorer with the restored video selected, deletes the working files (as Free up
space does) and the player cache, and removes the job from tapewake. The restored video, the saved
samples and the log are kept. - Jobs that run from their own scripts keep their working files; only the job is removed from tapewake.
tapewake 0.1.1: disk space
There's no installer to download yet: build it from source with BUILDING.md. Full history in CHANGELOG.md.
Disk space: working files are deleted once nothing needs them
Why: restoring a tape makes large working files at every stage, and the lossless ones are huge.
Before this update every one of them was kept until the very end, and none were deleted after the
finish. A two-hour tape used about 180 GB and kept all of it until you deleted it by hand. The
README said about 150 GB.
What changed
- Deinterlacing now stays at most 8 pieces ahead of the graphics card instead of doing the whole tape
first (settingqtgmc_ahead). - Each deinterlaced piece is deleted as soon as it and both of its neighbours are restored.
- Once the finished video passes its checks, whatever is left of the deinterlaced and in-progress
files is deleted automatically. - New Free up space button for finished restorations. It deletes the remaining working files (the
cut source and the restored pieces) and keeps the restored video. You click it twice to confirm. It
refuses while a restoration is running and before the finished video exists. Afterwards the
before/after player for that video explains that its files were deleted. - Resuming a finished restoration no longer rebuilds its finished video.
- New setting
keep_work_fileskeeps everything, for debugging. qtgmc_aheadcan't go below 2. Restoring a piece needs the next one deinterlaced too, so a smaller
value would have stalled the restoration (found while testing this update).
Disk space needed now, per hour of tape:
| Before | Now | |
|---|---|---|
| Peak while restoring | about 90 GB | about 50 GB |
| Left after Free up space | (no such option) | about 17.5 GB, just the restored video |
The README, BUILDING.md, the in-app setup check and engine/README.md
now give these figures.
Testing
- Simulated jobs: the deletion rules, a failed piece being redone, a hand-deleted piece being rebuilt,
progress, the button's refusals, and no stalls at anyqtgmc_aheadvalue. - A real restoration of a 12-second NTSC clip, cut into 4 pieces with
qtgmc_aheadset to 2:- Deinterlacing waited for the graphics card as intended, with at most 3 deinterlaced pieces at once.
- The deinterlaced and in-progress folders were empty after the finish.
- Free up space left only the restored video and the logs, and Resume then refused to start.
- The output had the same 722 frames as a restore of the same clip with 0.1.0, and averaged 42.9 dB
PSNR against it (lowest frame 41.4 dB, joins between pieces no worse), so the picture is the same.
- Not yet run on a full-length tape.