Interpolation models
10 interpolation models. Each section says what the model is, whether it may be used commercially and how fast it is; licences and checks are under Details.
Input: two RGB images (consecutive frames). Output: the frame between them at timestep (0.5 by default; 0 = the first, 1 = the second), RGB in [0, 1].
Measured on RTX 5060 Ti, torch 2.14.1+cu130; fastforks bench MODEL --size HxW measures yours. The summary uses one input, 1080x1920: kernels time with its speed-up over the faster original run, and kernels peak reserved memory with its ratio to the smaller original one. Where measured (one variant per family), a variant's table also lists portable under torch.compile and the fp16 TensorRT engine of portable (export; memory: the engine's device memory + torch's peak). – = not measured; no kernels = the family has no kernels backend (reason under its table). Reserved memory is the allocator's cache: kernels (CUDA graphs, cuDNN workspaces) can reserve a few percent to some tens of percent more than portable at the same allocated peak.
Other devices: Apple M1 Pro 16 GB, NVIDIA L4, NVIDIA L40S, NVIDIA A10, NVIDIA A100-SXM4-40GB, NVIDIA RTX PRO 6000; across devices.
| model | commercial | kernels time (ms), speed-up vs original | kernels peak memory (MiB, reserved), ratio to original |
|---|---|---|---|
ema-vfi | yes | 207, 1.4x | 3566, 0.83x |
ema-vfi-small | yes | 84.4, 1.4x | 1914, 0.8x |
ema-vfi-small-t | yes | 84.3, 1.4x | 1914, 0.8x |
ema-vfi-t | yes | 207, 1.4x | 3566, 0.83x |
film | yes | 332, 1.4x | 3080, 0.87x |
gimm-vfi-r | no | 1629 | 5362 |
gimm-vfi-r-p | no | 1628 | 5282 |
rife-4.25 | yes | 21.4, 1.1x | 916, 0.79x |
rife-4.25-lite | yes | 18.3, 1.2x | 896, 0.74x |
rife-4.26 | yes | 21.4, 1.1x | 916, 0.79x |
All models
Modes, licences, training data and outputs per model
Commercial status per variant: yes if the code and the weights licence both allow commercial use (the training data is listed but does not change it; weights without a licence of their own are assumed to carry the code licence); no if either forbids it; unclear for custom licence terms (treat as no).
| model | modes | code licence | weights licence | training data | commercial | outputs |
|---|---|---|---|---|---|---|
ema-vfi | portable, kernels | Apache-2.0 | Apache-2.0 | vimeo90k | yes | frame |
ema-vfi-small | portable, kernels | Apache-2.0 | Apache-2.0 | vimeo90k | yes | frame |
ema-vfi-small-t | portable, kernels | Apache-2.0 | Apache-2.0 | vimeo90k | yes | frame |
ema-vfi-t | portable, kernels | Apache-2.0 | Apache-2.0 | vimeo90k | yes | frame |
film | portable, kernels | Apache-2.0 | Apache-2.0 | vimeo90k | yes | frame |
gimm-vfi-r | portable, kernels | LicenseRef-noncommercial | LicenseRef-noncommercial | vimeo90k | no (code: LicenseRef-noncommercial (noncommercial); weights: LicenseRef-noncommercial) | frame |
gimm-vfi-r-p | portable, kernels | LicenseRef-noncommercial | LicenseRef-noncommercial | vimeo90k | no (code: LicenseRef-noncommercial (noncommercial); weights: LicenseRef-noncommercial) | frame |
rife-4.25 | portable, kernels | MIT | MIT | vimeo90k | yes | frame |
rife-4.25-lite | portable, kernels | MIT | MIT | vimeo90k | yes | frame |
rife-4.26 | portable, kernels | MIT | MIT | vimeo90k | yes | frame |
Training datasets
| id | dataset | licence | commercial use |
|---|---|---|---|
vimeo90k | Vimeo-90K (septuplets / triplets for frame interpolation) | not stated | unclear |