// the take
ab-av1 automates the tedious part of AV1 encoding: finding the right CRF value to hit a target VMAF score without guessing. It samples short clips at different CRF values, interpolates, then runs the full encode — skipping the usual trial-and-error loop. For anyone batch-encoding video archives or doing quality-preserving compression, this is the tool to reach for.
1. The binary search over sample clips is genuinely smart — it avoids encoding the whole file dozens of times, which makes CRF tuning practical instead of painful. 2. VMAF model auto-selection (4K vs 1080p) and auto-threading means you don't have to think about those parameters. 3. JSON output mode makes it composable in scripts and pipelines — not just a GUI-adjacent tool. 4. Statically linked musl binary for Linux means no dependency hell; it just works on any distro.
1. Requires a specific ffmpeg build with libsvtav1, libvmaf, and libopus enabled — most distro-packaged ffmpegs won't have all three, and the README doesn't tell you how to get one that does. 2. VMAF scoring is CPU-intensive; on long or high-resolution sources the sample phase can still take a while, and there's no GPU-accelerated VMAF path. 3. No batch/queue mode — you have to shell-script your own loops over multiple files, which means managing partial failures yourself. 4. The CRF search assumes a monotonic VMAF-to-CRF relationship, which holds in practice but can misfire on pathological inputs (heavy grain, animation with flat areas) where samples aren't representative of the full file.
We read the README, the directory tree and a key source file in June 2026. Nobody pays to be listed. How we judge →