google-research/frame-interpolation

Health v2 · maintenance only

Flags: no_releases archived

Adoption not part of the score

3150 stars · 323 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

FILM is the official TensorFlow 2 implementation of a state-of-the-art frame interpolation neural network from Google Research, presented at ECCV 2022. It synthesizes smooth slow-motion video from near-duplicate photos or sparse frames, handling large scene motion with a unified single-network approach that requires no pre-trained optical flow or depth models.

Use cases

• turn near-duplicate photos into slow motion video
• interpolate between two images to generate in-between frames
• smooth out low-framerate video by generating intermediate frames
• high resolution frame interpolation with patch subdivision
• research baseline for large motion frame interpolation

When to choose

• you need high-quality interpolation across large motion between frames
• you want a TensorFlow-based model with pre-trained checkpoints and a CLI
• you're converting photo sequences into slow-motion footage

When to avoid

• you need real-time interpolation on CPU-constrained hardware
• you prefer PyTorch tooling
• you need active feature development rather than a maintained research artifact

Facets

library · maturity maintenance

image-processing video-processing machine-learning deep-learning computer-vision machine-learning media python cross-platform frame-interpolation optical-flow slow-motion tensorflow eccv-2022 video-synthesis video gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="google-research/frame-interpolation")