Modelpublic

kevinqz/RIFE-Frame-Interpolation-CoreAI

sourceHugging Facemitupdated 2mo agoView on Hugging Face

0likes

Model Card

Canonical: `kevinqz/RIFE-Frame-Interpolation-CoreAI` — source of truth.

RIFE Frame Interpolation (fabric)

An Apple Core AI conversion of TensorForger/RIFE-safetensors — a frame interpolation network that takes two stacked RGB frames and synthesizes the intermediate frame between them. Produced by coreai-fabric and indexed by coreai-catalog.

Frame interpolation, not a generator. This is a single-forward interpolator — two frames in, one synthesized frame at t=0.5 out. No temporal model, no audio. The host owns frame decode/encode and resizing each frame to a multiple of 32 before feeding the graph.

Model facts

Field	Value
Parameters	0.01B
Architecture	cnn/transformer
Capabilities	super-resolution
Input	1×6×256×256 — concat(img0[:3], img1[3:]) in [0,1]
Output frame	1×3×256×256 @ t=0.5
Quantization / precision	none / float32
On-disk size	17 MB
Asset kind	single-graph frame interpolator (two frames -> middle frame)
assetVersion	2.0

Use it — this needs host code you supply

The bundle is a single static-size graph: frames (1×6×256×256, img0 and img1 concatenated on the channel axis, RGB in [0,1]) in → the interpolated frame (1×3×256×256) out. You supply the video demux/mux, frame resizing, and any multi-t or recursive-interpolation loop in your host code (Swift or Python).

bash

pip install coreai-catalog && coreai-catalog install rife-4-interp

Requirements

—Deployment: macOS 27.0+ / iOS 27.0+, Xcode 27+. The asset serializes with minimum_os v27, so the on-device Swift runtime requires macOS/iOS 27+. A Mac on macOS 26 can convert and inspect it but not run it on-device.
—Apple Silicon.

Verification (output parity)

—Gate A (structure): passed — the bundle's layout + metadata were validated; the graph loads.
—Gate B — graph_output_cosine: 1.000000 min output cosine (median 1.000000) vs the fp32 torch reference network over 8 seeded frame pairs, measured on apple_silicon. Certifies the export computes the SAME output as the source — a conversion-fidelity metric, not task accuracy.
—This certifies the export is numerically faithful to the source network — it does NOT certify perceptual interpolation quality on your footage. Reproduce with coreai-fabric verify.

Provenance

Field	Value
Base model	TensorForger/RIFE-safetensors @ `78a62b7c2dd910536432d6c2c3a25e76f14fbf78`
Converted by	`models/rife/export.py` (version not reported)
Recipe	rife-4-interp (recipe_source: fabric)
Precision / quantization	float32 / none
Conversion date	2026-07-10

Machine-readable, in this repo: `parity-report.json` · `reproduce-manifest.json` · `LICENSE`.

License and attribution

Weights licensed mit — see the bundled LICENSE. This artifact is a converted derivative of the base model: its weights were converted to Apple Core AI format. The conversion itself is community work.

The on-device Core AI ecosystem

—coreai-fabric — the reproducible recipe → .aimodel pipeline that produced this asset.
—coreai-catalog — the index of Core AI models with provenance and integration snippets.
—apple/coreai-models — Apple's official exporters and runtimes.

Not affiliated with Apple

Community conversion. Not produced, hosted, or endorsed by Apple. Apple and Core AI are trademarks of Apple Inc., used here only to describe the target runtime/format.