Film (24p) content¶

Nearly all commercial film DVDs store 24 fps material and mark it for 3:2 pulldown rather than storing 60 fields per second. The core handles that two ways, and Film 24p Out (Debug page, default Auto) picks between them.

Default path — the core performs the 3:2 pulldown itself, following the flags in the stream, and outputs at the display's native 59.94 Hz (50 Hz for PAL).

Film 24p path — the core instead outputs a true 23.976 Hz progressive raster (25.000 Hz for PAL) and lets the framework scaler do the pulldown. Because 23.976:59.94 is exactly 2:5 and the clocks are locked, that conversion is exact.

This second path also cuts framebuffer re-reads from 60 per second to 24, which hands the decoder a much larger uninterrupted memory window each frame — so it helps throughput on demanding discs as well as cadence.

When to change it¶

Leave it on Auto. The one case that needs On is a hard-telecined disc, where the pulldown was baked in when the disc was authored — Auto genuinely cannot see those, for the reason in What it cannot see below. If a disc looks like film and Auto is not engaging after a few seconds, set Film 24p Out = On.

How Auto detects film¶

A DVD does not label itself as film. What it carries instead are two flags the encoder writes into every coded picture, and the detector reads the pattern they make.

progressive_frame: The encoder's claim that this picture is a whole progressive frame rather than two interlaced fields.
repeat_first_field: Show one of this picture's fields a second time. This is how 24 fps becomes 60 fields per second — alternate frames get an extra field, giving the 3:2 pattern.

NTSC film shows up as progressive_frame set and repeat_first_field toggling from the previous picture. That toggle is the signature: it is what alternates the 2-field and 3-field pictures. PAL film is simpler — 25 fps needs no pulldown, so progressive alone is enough. True interlaced video is the opposite: progressive_frame clear, sustained.

It builds confidence rather than counting a run¶

Each picture nudges a confidence score — up for a confirming picture, down for a contradicting one, with a hard drop for a definitely-not-film picture and a gentle one for an ambiguous case. Film mode engages at the top of that range, roughly 40 clean film frames, about 1.7 seconds into a title, and disengages only after a long run of contradictions.

The reason it is a score and not a consecutive run: a strict run gets reset to zero by anything that interrupts the cadence, and reaching a title through the disc's menus does exactly that — nav packs, cell and PGC boundaries, and brief still frames all inject hiccups. A run-based detector locked fine on a disc played straight through and never locked at all through a menu. A score decays through a hiccup and recovers.

It also cannot be fooled into engaging by accident. 30 fps progressive video never toggles repeat_first_field, so it produces no confirming pictures at all — its score only ever falls. The bias is deliberately toward not film.

Pictures that carry no evidence are ignored¶

progressive_frame is a claim the encoder wrote, not a measurement — and on a near-black picture there is no field structure to describe, so encoders fall back to the MPEG-2 default and mark it interlaced. Believing that is what used to make fading credits knock the detector out of film lock: Apollo 13's opening changed output mode nine times in the first 46 seconds, re-locking the display each time.

So the detector now scores a picture only if it carries real evidence, using how many bytes the picture took to code as a stand-in for how much is happening in it, judged against what is typical for that disc — so it works the same on a heavily compressed disc as on a lavish one. A picture that fails the test updates nothing at all, not even the repeat_first_field history, so the 3:2 toggle test picks up across the gap instead of seeing a false edge.

What it cannot see¶

Hard-telecined discs. If the pulldown was baked in when the disc was authored, the frames really are interlaced and there are no flags saying otherwise — there is no pattern left to detect. Nothing can infer it from the stream, which is why the manual On exists.

Frame Drop must stay On. Advancing past a frame is how the player recovers time it has lost, and film content on a demanding disc is exactly where it needs to. With it off, the picture runs progressively later and lip sync goes with it. See Settings.
A/V Offset defaults to 0 ms. There should be no need to change it. (It was +100 ms in v0.4.0 and earlier, nulling an offset the current clock no longer has.)

Known limitations¶

Unreleased — fixed

The offset that used to appear when Auto engaged a couple of seconds into a title, and cleared on the next seek, is gone. It was not a film-detection problem: the player's clock counted displayed frames from a point in the demux, so everything buffered between there and the screen turned into an audio offset, and engaging the 24p raster was simply where it became audible. The clock now runs on its own and both picture and sound are presented against it, so there is nothing for a seek to fix.

Film 24p needs Video Output = Progressive (or Auto on an HDMI rig) — a 23.976 Hz raster cannot carry interlaced fields, so it is unavailable while Interlaced mode is active. Film on a CRT plays with its normal 3:2 field cadence instead.