Repository navigation
v0.2.0 adds the OpenFX build: the same offset-litho press, as a filter for DaVinci Resolve, Vegas, Nuke and Natron, printed on the CPU. It is the same C++ the Resolume build uses for the separation, the spot functions and their threshold table, the dot gain, the ink model, the wander and the six presets. Its per-pixel print pass is a line-for-line mirror of the plugin's GLSL, and it is held against the GPU's output. The Resolume build is otherwise unchanged: its controls now reach the shaders through one function that both builds call, and that refactor changed no pixel of its render.
What differs in OpenFX
- No audio. OpenFX gives a plugin no audio, and a timeline renders frames in any order, so Audio Drive and the audio input are left out rather than left dead. No preset touches them, so every preset means the same press in both builds.
- Presets set the controls. Choosing one writes its values into the controls as one undo step, and moving a control it covers off the preset's value puts the menu back to Custom. Registration, Solo and Mix are left alone by every preset, as in Resolume.
- The press runs on the timeline. The wander is a pure function of time, so any frame renders on its own and in any order. Time is the frame number over the clip's frame rate. Resolve's Fusion page reports the frame rate on the effect but not on its clips; the plugin reads the effect's, so it runs at the composition's rate there, and assumes 24 fps only where a host reports no rate at all.
- The output is declared frame-varying, so a host re-renders the wandering press on every frame even over a still clip.
- The ink colours and the paper are colour pickers rather than three sliders each. Same values, same defaults.
- At a reduced render scale the screen and the press offsets scale with it, and a non-square pixel aspect is taken into account so the dots stay round. Both are written but not yet seen working in a host.
Downloads
- Resolume (FFGL), as before:
rosette-macos-universal.zipand a.dmgfor macOS (universal),rosette-windows-x86_64.zipand a setup.exefor Windows x64. The plugin goes in Resolume's Extra Effects folder. - OpenFX, new:
rosette-ofx-macos-universal.zip,rosette-ofx-windows-x86_64.zipandrosette-ofx-linux-x86_64.zip. CopyRosette.ofx.bundleinto/Library/OFX/Plugins/(macOS),C:\Program Files\Common Files\OFX\Plugins\(Windows) or/usr/OFX/Plugins/(Linux), then restart the host. The Linux build targets glibc 2.28, so it loads on Rocky 8.
The macOS builds are signed and notarised by the release pipeline after publication. The Windows builds are unsigned, so SmartScreen warns once.
What is measured
On an M4 Max, macOS 26.4:
- The CPU print against the GPU, on the test card, over 16 cases (the defaults at two sizes, all six presets, seven settings across every control group, a soft alpha ramp at half Mix): at most 0.77% of pixels differ at all, at most 0.0135% by more than one 8-bit step, mean at most 0.0021/255. The remaining pixels sit on a cell boundary. A control case, Screen 0.46 against 0.48, differs in 70% of pixels, so the comparison can fail.
- The real
Rosette.ofx.bundlein a CPU OpenFX test host gives the same numbers as the in-process check. - Frame 37 of a wandering press renders byte-identically alone, after frames 0–36, and after 60, 5 and 0. Mix 0 is declared to the host as an identity and is exact when rendered.
- 37 ms per frame at 1920×1080 on eight threads.
- In DaVinci Resolve Studio 21.1 on macOS, 2026-10-04, as a Fusion tool, it renders byte-identical to the test host at 24 fps. The first build had failed every frame there: Fusion reports the frame rate on the effect but not on its clips, and that build read a clip's. The fix reads the output clip, the source clip and then the effect, each guarded.
tools/verify.shpassed on a fresh universal build on 2026-10-04, covering the Resolume build's checks from v0.1.0 as well as the OpenFX ones. Those include a render under a test host stricter than Fusion, which withholds the effect's frame rate too; it matches the render at 24 fps.
Not done
- The OpenFX build has never been in Vegas, Nuke or Natron. In Resolve it has been run only as a Fusion tool, not on the Edit or Color page.
- The Windows and Linux OpenFX builds have never rendered a frame in a host. They have been compiled on GitHub's runners, and the Linux one loaded on Rocky 8.
- A real host's proxy render scale, pixel aspect, premultiplication and parameter panel are unconfirmed.
- The CPU print was matched against one GPU, an Apple M4. Another card's Resolume render may differ from both by an amount nobody has measured.
- The Resolume build is where v0.1.0 left it: never run on a GPU in Resolume, never instantiated in Arena on macOS, no real audio in a host, and its clock unit in Arena is unconfirmed. No user guide.
Full Changelog: v0.1.0...v0.2.0