NNFlowVector is free to use for producing regular motion vectors, and no license is required to run it (just download and install it using the included instructions). To be able to use the smart vector output feature, you have to obtain a license (either buy one or request a trial license first).
Please do try the software on your own system before ordering a license using the shop.
If you do wish to test out any of the softwares without limitations, you can request a free trial license using this page.
By downloading our software you are accepting our end user license agreement.
If you are unsure of what CUDA version of the different builds that you should download and run, please have a look at our new CUDA compatibility chart page.
NNFlowVector – v2.2.1 (currently Linux only, Windows builds to come)
CUDA11.8.0* | CPU only | |
NVIDIA cuDNN | v8.4.1* | – |
NVIDIA driver | 520.61+ | – |
Nuke14.1 (Nuke Indie compatible) |
![]() |
![]() |
CUDA11.8.0 | CUDA11.1.1* | CPU only | |
NVIDIA cuDNN | v8.4.1 | v8.4.1* | – |
NVIDIA driver | 520.61+ | 455.32+ | – |
Nuke14.0 (Nuke Indie compatible) |
![]() |
![]() |
![]() |
For instant Ampere/RTX30xx compatibility, please use the CUDA11.2 (or CUDA11.8) builds.
But for maximum compatibility, please read below: * This combination of CUDA & cuDNN matches what Nuke 13.x is built against. This means that if you do need NNSuperResolution to work well together with Foundry AIR nodes (such as CopyCat) in the same Nuke scripts, you have to use this version. If you got a compute capability 8.0, 8.6, 8.9 or 9.0 GPU (i.e. Ampere/Ada Lovelace/Hopper architecture, for example RTX3080), you will then need to wait for the plugin to auto JIT-compile the kernels for that GPU the first time it’s used. Please see the section about “environment variables” in the documentation to read about how to increase the CUDA_CACHE_MAXSIZE to successfully use this version with modern GPUs.
Hot fix update 2023-10-09: NNFlowVector Utils v2.1.0R2
This fix update for NNFlowVector Utils implements a newly found workaround to get the MotionVector_DistortFrame and MotionVector_FrameBlend gizmos working correctly in Nuke 13.x and Nuke 14.x. They have been broken since Nuke13.1 and above because of a bug in Nuke that is still due to be fixed by Foundry. However, if you download this fix above the gizmos start working as intended even without a fixed Nuke. (For the interested, the bug that relates to using time nodes upstream of an IDistort node and can be tracked here)
NNFlowVector – v2.1.0
CUDA11.1.1** | CPU only | |
NVIDIA cuDNN | v8.4.1** | – |
NVIDIA driver | 455.32+ on Linux 456.81+ on Windows | – |
Nuke14.0 (Nuke Indie compatible) | ![]() ![]() | ![]() ![]() |
* This combination of CUDA & cuDNN matches what Nuke 13.x is built against. This means that if you do need NNFlowVector to work well together with Foundry AIR nodes (such as CopyCat) in the same Nuke scripts, you have to use this version. If you got a compute capability 8.0 or 8.6 GPU (i.e. Ampere architecture, for example RTX3080), you will then need to wait for the plugin to auto-compile the kernels for that GPU the first time it’s used. Please see the section about “environment variables” in the documentation to read about how to increase the CUDA_CACHE_MAXSIZE to successfully use this version with modern GPUs.
Important! If you are downloading the Windows versions of the plugin, you have to add the full path to the chosen plugin installation folder to the PATH environment variable on your system (in addition to the NUKE_PATH environment variable). This is needed for the plugin to be able to find additional included libraries that it is dependent on. Please refer to the “Plugin Installation” section in the included Documentation for more details. (This is handled automatically by the system when you are using the Linux versions.)