tlottie

Library for drawing lottie animations, written in Rust.

• Micro-optimized and benchmarked across 17k+ animations from Telegram, against rlottie and thorvg.
• SIMD on ARM NEON, x86_64 SSE2, and WASM: web demo. All three are baseline features of their target, so there is no runtime dispatch.
• Support of fitz modifier and color replacements.
• RGBA or BGRA output, chosen per instance at parse time and free at render time — desktop compositors (Qt ARGB32_Premultiplied, DXGI B8G8R8A8_UNORM, Cairo ARGB32) take BGRA without a conversion pass.
• Support of rendering into only alpha channel bitmap.
• Safe. Lottie JSON is treated as untrusted input.
• Zero dependencies¹.

Building (Native)

cargo rustc --release --features c-api --lib --crate-type staticlib
# library will be built at target/release/libtlottie.a
# include headers are at include/tlottie.h

Building (Web)

rustup target add wasm32-unknown-unknown
cargo build --target wasm32-unknown-unknown --release --no-default-features --features wasm --lib
# .wasm will be built at target/wasm32-unknown-unknown/release-size/tlottie.wasm

Example on how to work with tlottie.wasm is at examples/web/.

Benchmarks

Frame time, mainly compared to rlottie2019.

• android arm, Samsung Fold7: -64.7% at 64px, -55.8% at 320px, -52.0% at 720px
• android arm, Samsung F15: -62.7% at 64px, -42.0% at 320px, -23.0% at 720px
• macOS arm: -61.5% at 64px, -47.6% at 320px, -27.3% at 720px
• linux x86²: -70.0% at 64px, -50.2% at 320px, -36.4% at 720px

1. Listed dependencies are only for GPU renderers, that are currently work in progress, expected to be useful only on large canvas sizes. ↩

2. Tested on Threadripper PRO 9995WX at full core count, running tlottie’s SSE2+AVX2+AVX-512 backend against rlottie’s. Relative parity with rlottie2019 is worse here than on smaller cores because the vector span kernels are memory-bound at 720px; tlottie is faster than this on typical x86_64 laptops (see the AVX-512 breakdown in the report). ↩