Video

Full-screen, full-motion cutscenes with synchronised audio, in NextDAAD's own NXV format.

Drop a video source in VIDEO\, named by video number. VIDEO\001.mp4 is encoded by the build to VIDEO\001.vid and staged as RELEASE\001.VID; a pre-encoded VIDEO\001.vid with no .mp4 beside it is staged as it is. .mkv sources are accepted the same way, with the same numeric naming (VIDEO\001.mkv).

Your game plays one with GFX n 13 (play once) or GFX n 14 (loop until any key), where n is the video number - GFX 3 13 plays 003.VID. The classic DOS DAAD symbols SFX n 9 (PLAYFLI) and SFX n 10 (PLAYFLIL) do the same thing - see the SFX sub-commands table in Audio.

In a multi-part game, PART<n>\NNN.VID shadows the root copy for that part only.

Video needs a 2MB Next. Playback buffers through a large pool of RAM banks, and a 1MB machine has too small a pool to hold or stream clips reliably.

Video needs real hardware to watch. Cutscenes do not play correctly under emulation: the player reads the card directly and paces itself off a hardware timer, and no emulator currently reproduces both well enough to run a clip through. The rest of your game tests in an emulator as usual - only the cutscenes need a real machine.

The kit can encode the same piece of footage two completely different ways, and on this hardware that choice decides more about how a cutscene looks than every other setting put together. Pick the right route and the defaults are usually enough; pick the wrong one and no amount of tuning rescues it. Sections 1 and 2 below are that decision.

An NXV file is fully self-describing - there is nothing to select at the DSF level. A clip is encoded to a shape: any Layer 2 width (256 or 320) by any height from one line to the mode's full 192 or 256. Five preset shapes ship:

Every shape carries 256-colour pictures with adaptive palettes and full-rate stereo audio at 15625 Hz. A mono source is converted to stereo for you. Letterbox bars cost no file bytes at all - the player draws them black. Sources are centre-cropped to the shape's displayed aspect, never distorted.

Settle the shape before anything else. Because the crop is a crop, a 16:9 source encoded into the 4:3 full shape loses its sides. Match the shape to how your footage is framed first, then pick a route.

Smaller shapes come out better. The player's data budget is fixed, so quality scales inversely with pixel count: classic has three-fifths the pixels of full, so at the same data rate each pixel gets about 1.7 times the bytes. If a clip looks rough at full, re-encode it letterboxed. Two things qualify that: hard content - constant whole-frame motion, noise, fades - eats budget whatever the shape, and the letterboxed 320-wide shapes carry a decode surcharge for their gapped surface, which makes 16:9 the usual best trade for a widescreen source.

Free heights. Beyond the presets, any WIDTHxHEIGHT is a valid shape (320x150, 256x100 and so on), and --aspect derives the height for a target displayed aspect ratio - --aspect 2.35 gives true cinema scope, with the 320-wide mode's non-square pixels corrected for automatically.

Compressed is the default and needs no options. It sends only what changed between frames, so it is very efficient on footage that largely holds still, and it can run at the full 25 fps at the full 320x256 screen size on that kind of material. Its weakness is fine detail moving everywhere at once: when the frame asks for more than the card can deliver, parts of the picture are left to catch up on a later frame, and that shows as banding - stripes or blocks that lag behind the rest of the image, usually vertical on the full-width shapes and most obvious in water, weather and gradients.

Uncompressed (--direct) sends every frame whole, straight from the card to the screen. It cannot band, because nothing is ever left behind - every pixel of every frame is exactly what the encoder produced. Its weakness is size: it costs about 1 MB for every second of screen time, which limits both how long a clip can be and how much picture you can have at 25 fps.

Neither route is better. They fail in different directions, and the footage decides which failure you would rather have.

Answer these in order and stop at the first one that fits.

1. Is the clip longer than about 15 seconds? Compressed. Uncompressed costs about 1 MB for every second of screen time, so past this point it is spending more of the card than most games can spare on one scene (section 6). Go to Preset 5.

2. Does the picture mostly hold still? A slow zoom, a slow pan, a held shot, a talking head, a title card, a mostly-static scene with one thing moving in it. Compressed, at full screen, at 25 fps. This is the cheapest good result in the kit and it is already the default. Go to Preset 1.

3. Is fine detail moving across the whole frame? Water, rain, falling snow, smoke, foliage, crowds, a camera moving over textured ground. This is the class compressed video handles worst, and the one where uncompressed wins outright. Then ask which you would rather keep:

   • The big picture - full 320x256, at half the motion rate. Go to Preset 2.
   • The motion - full 25 fps, at a smaller picture. Go to Preset 3.

4. Is the source grainy, noisy or heavily textured? Old film, high-ISO footage, heavy compression artefacts in the source. Whichever preset the answers above gave you, add the line from Preset 4.

5. Anything else - start compressed with the defaults (Preset 1). If the result bands, work down Preset 6.

Three ways to apply them, all equivalent.

Per video, with VIDOPTS_NNN. Set it for the video number, and the next build re-encodes that clip:

SET VIDOPTS_001=--direct --shape full --fps 12.5

This is the normal place for anything in this document, because these are per-clip decisions - one cutscene may want one route and the next another. VIDASPECT, VIDFPS and VIDOPTS apply to every clip in the game, so use those only for something you genuinely want everywhere.

A VIDOPTS_NNN line works in CONFIG.BAT or in CONFIG.local.BAT. Any setting in CONFIG.local.BAT overrides the same setting in CONFIG.BAT (every key, last wins, and an empty SET NAME= clears it), so a VIDOPTS_NNN line in the local file masks the same-numbered line in CONFIG.BAT. Lines you write by hand in CONFIG.BAT stay valid.

By hand, for a one-off encode or to cut a clip from a longer source:

tools\vidtools\videnc.exe source.mp4 VIDEO\001.vid --direct --shape full --fps 12.5 --start 00:01:12 --duration 8

Keep the source file outside VIDEO\ when you do this. The build re-encodes any VIDEO\NNN.mp4 whose settings it does not recognise, so a VIDEO\001.mp4 sitting next to your hand-made VIDEO\001.vid will overwrite it on the next build. A .vid with no .mp4 beside it is used exactly as you made it.

In the tuner. VIDTUNE.BAT puts the same settings behind a preview window; when you accept an encode it saves the matching VIDOPTS_NNN line into CONFIG.local.BAT (created on the first accept, or by Encode Stale + Edited when it saves a clip) and never edits CONFIG.BAT, so the settings survive a kit update. That local line overrides any VIDOPTS_NNN line for the same video in CONFIG.BAT. Resetting a clip whose setting comes from CONFIG.BAT writes an empty SET VIDOPTS_NNN= to the local file, which clears it.

Before you drop your own footage into VIDEO\001.mp4 or 002.mp4, read the VIDOPTS_NNN lines already in CONFIG.BAT. Those were written for the demo clips that ship with the kit. Options that suited a cartoon will not suit your footage, and they apply silently to whatever file you put in that slot. Editing CONFIG.BAT does not last, because a kit update brings the lines back. Instead reset the clip in the tuner, which writes an empty SET VIDOPTS_001= to CONFIG.local.BAT to mask the line, or write that empty line there yourself. The local file survives kit updates.

For most repeated options the encoder takes the last occurrence, so VIDOPTS_NNN wins over VIDOPTS. Shape is guaranteed rather than left to order: a VIDOPTS_NNN that sets its own --shape, --width or --aspect always wins, because the encode pass suppresses VIDASPECT for that video.

The .vid beside the source is an encode cache, keyed on both the .mp4's timestamp and the effective VIDASPECT/VIDFPS/VIDOPTS/ VIDOPTS_NNN settings for that video (kept in a sidecar, VIDEO\NNN.vid.args). Change any of them and the clip re-encodes by itself - there is nothing to delete by hand.

Encode time is quality. The encoder is deliberately slow: it spends its time squeezing the best picture into the fixed playback budget. Expect minutes per clip, once per source change.

Each one is a complete answer for a class of footage. Copy the CONFIG.BAT line, adjust the video number, build.

For: slow zooms and pans, held shots, talking heads, title cards, establishing shots, anything where most of the frame is the same from one moment to the next.

SET VIDOPTS_001=

Nothing at all - the kit defaults are the right answer for this material. Full 320x256, 25 fps, stereo.

Optional, and worth trying on this class if the motion is continuous (a zoom or pan that never stops rather than a static shot):

SET VIDOPTS_001=--tile-slack 0.5

You get: the whole screen, at full motion rate, with room to spare. Footage in this class typically uses less than half of what the card can deliver, so the picture is not fighting for space.

You pay: nothing. Do not send this kind of clip to the uncompressed route - it looks the same and costs roughly three times the card space.

For: detailed sustained motion filling the whole frame - sea, weather, falling snow, crowds, smoke - where you want the picture at full size and can accept less fluid motion.

SET VIDOPTS_001=--direct --shape full --fps 12.5

For a 16:9 source, use its own shape instead - the half-rate envelope has room for every named shape:

SET VIDOPTS_001=--direct --shape 16:9 --fps 12.5

You get: the full 320x256 screen with no banding whatsoever, every frame pixel-exact. On the footage this route is meant for, the difference is not subtle - detail that the compressed route turns into flat stripes stays as detail.

You pay: two things, both real. About twice the card space of the compressed version of the same clip, and half the motion rate - 12.5 frames a second instead of 25. Motion is noticeably less smooth. It is still a clean cadence, so there is no stutter or beat on top of the reduced rate, but it reads as half-rate and you should look at it before committing.

Expect the encoder to print an at-capacity warning about wire utilization on this preset. Full-screen uncompressed sits at the top of what the card can carry, by design; that is what makes it possible at all. It plays at rate on a healthy, defragmented card.

For: the same detailed motion as Preset 2, when smooth movement matters more than picture size.

SET VIDOPTS_001=--direct --shape 256x167

For a cinematic wide framing at full width instead:

SET VIDOPTS_001=--direct --shape 320x105

You get: the full 25 fps, no banding, every frame pixel-exact. Motion looks the way it looked in the source.

You pay: the picture is smaller. 256x167 plays in the bordered 256-wide frame; 320x105 uses the full width but is heavily letterboxed. There is no full-screen version of this preset - 25 fps uncompressed and 320x256 do not fit down the same wire.

These two shapes are the most that fits at 25 fps with stereo audio. If you ask for more, the encoder refuses and prints the exact sizes that do fit for your chosen width - use one of those.

For: film grain, sensor noise, heavy source compression, or any footage that looks busy even where nothing is moving. Grain is motion as far as the encoder is concerned, and it is paid for on every single frame.

Add --prefilter to whichever preset you are already using - it is an extra option, not a route of its own:

SET VIDOPTS_001=--shape 16:9 --tile-slack 0.5 --prefilter

--prefilter runs a light temporal denoise before encoding. It can be given an explicit ffmpeg filter string if you want to control it, but the bare form is deliberately conservative and is the one to try first.

You get: a modest improvement on genuinely grainy material - enough to be worth an encode, not enough to change a bad route into a good one. Compare the two encodes before keeping it. It does most on the compressed route, where every speck of grain is charged for on every frame; on the uncompressed route a frame costs the same whatever is in it, so denoising there only changes how the picture looks.

You pay: a little fine texture. On clean sources it takes away more than it gives, so this is not a default.

For: anything over about 15 seconds. At that length uncompressed costs more card space than a game usually has to give (section 6), so the compressed route is the practical one and the quality decision moves to the shape.

SET VIDOPTS_001=--shape 16:9 --tile-slack 0.5

Or, for a 4:3 framing that still needs the room:

SET VIDOPTS_001=--shape classic --tile-slack 0.5

You get: the compressed route working with headroom instead of against a wall. A smaller shape has real room to spare where full 320x256 has none, and that headroom is what stops banding. Drop --tile-slack if the clip is a series of static shots rather than continuous movement - it does nothing there.

You pay: picture area. A letterboxed or bordered clip is smaller on screen. This is the trade the compressed route offers and there is no way around it at length.

If a long sequence really needs to be full-screen and full-motion, consider cutting it into separate numbered clips played one after another. Each play is its own load, so expect a beat between them - this works for a sequence with natural cuts, not for one continuous shot.

Work down this list, re-encoding and looking at each step. Stop at the first one that fixes it. The order is cheapest-first in terms of what it costs you on screen, not by how much each one helps - step 2 is the biggest lever of the six, but it is also the first one that takes something away.

1. --tile-slack 0.5. The one picture knob worth trying on any clip with continuous motion. Costs a little of the encoder's safety margin, helps pans and zooms, does nothing on static material. (Compressed encodes only - there is nothing for it to act on in an uncompressed one.)

2. A smaller shape. 16:9, then scope, then classic-wide. This works because shape is the real problem: the full screen leaves nothing spare, and a smaller one has genuine room. Match the shape to your framing so the crop does not hurt.

3. --prefilter, if the source has any grain or noise in it (Preset 4).

4. --fps 12.5. Halves the motion rate and gives every remaining frame twice the room. The same trade Preset 2 makes, on the compressed route.

5. The uncompressed route - Preset 2 or Preset 3. If the clip is short enough to fit, this ends the argument: banding is not possible there.

6. Shorten or re-frame the shot. A four-second version of a difficult shot at a shape that works beats twelve seconds of stripes.

What not to do: do not set --stream-budget by hand. It is not a quality dial. It tells the encoder how much the card can supply, and lowering it does not compress harder - it hands the encoder less room, so more of the picture goes unpainted and everything gets worse at once. The encoder works out the right value for each clip and prints it. See Video delivery.

25 fps is the rate for almost every clip. The Next composites at 50 Hz, so only a rate that divides 50 evenly gives frames that each occupy a whole number of display frames - anything else beats against the 50 Hz composite and shows it. 25 is the default and there is no quality to be won by nudging it. The one alternative worth considering is 12.5, which also divides 50 evenly and hands every frame twice the room: half the motion rate, bought deliberately, for a demanding clip that cannot be made to look right at 25.

There is a floor. One frame may carry at most 3072 audio bytes, so 15625 Hz stereo needs 10.17 fps or more. The encoder refuses an encode below the floor and names the remedy - raise the frame rate.

Your footage is retimed for you. Almost no source material is 25p: film is 24 (or 23.976), and most cameras and stock libraries shoot 30 or 29.97. Every one of those has to be resampled in time on the way in, and how that is done is visible. The encoder blends to the target rate, automatically, whenever the source rate is not the target rate. A source that is already at the target rate is not touched at all.

Two overrides, both per title via VIDOPTS/VIDOPTS_NNN:

This matters more on the uncompressed route. Uncompressed playback reproduces every frame exactly as the encoder made it, blended in-between frames included, so if you can see the blend you will see it clearly. On a slow uncompressed shot try --retime mci; on a fast cut one try --retime drop.

Uncompressed costs about 1 MB for every second of screen time - roughly 60 MB a minute at full screen. A smaller shape or a lower frame rate costs less, roughly in proportion to the pixels sent each second.

Compressed clips are typically a third to a half the size of the same footage uncompressed, and quiet material is far cheaper still - it varies with the content, which is the whole point of the route. The encoder prints the finished size, and the .vid file in VIDEO\ is the file that goes on the card.

Budget accordingly. Three uncompressed cutscenes of twelve seconds each is most of 40 MB - fine on a modern card, but it is the largest thing in most games by a wide margin, and it is worth deciding early which scenes deserve it.

A video file can be up to 256 MB, and no clip may run past 65535 frames - roughly 43 minutes at 25 fps, 87 at 12.5. Divide 256 MB by the rate in the table above for the longest uncompressed clip each shape can carry: about four minutes at full screen, and longer at the smaller shapes. Compressed clips run at a much lower byte rate, so the same 256 MB carries several times as much of them - usually ten minutes or more, and the frame count only becomes the limit on very cheap material.

You do not have to work either limit out yourself: the encoder checks both before it writes anything, and if a clip is over it says so, and names the longest clip your shape and frame rate can reach. Nothing it writes is too big to play.

Bands or stripes across the picture. Parts of the frame are lagging behind, most visible in water, weather, gradients and fine texture. This is the compressed route running out of room. Work down Preset 6.

Juddery or stuttering motion, at a regular beat. The source frame rate is being resampled badly. Check section 5 - if you have set --retime drop, remove it, and if you have not set anything, try --retime mci on slow material.

Motion that steps rather than flows, but at a steady even cadence. A half-rate encode looks like this and is meant to - Preset 2 trades motion for picture size on purpose. If you did not ask for --fps 12.5, check that no VIDFPS or VIDOPTS line in CONFIG.BAT is setting it for every clip.

The whole clip runs slow, with the audio popping. That is a 60Hz display, not the encode. Playback is designed for a 50Hz display; on a 60Hz one, expect slower playback and audio popping. No setting fixes it.

A pause, or a frame that holds, in the middle of a clip. Usually the card, not the encode - a fragmented file cannot be read fast enough to keep up. Defragment the card, or copy the file onto a freshly formatted one, and try again. If it survives that, and the clip is uncompressed, it is sitting at the top of what your card can deliver: move to a smaller shape or to the compressed route.

Gradients step visibly - skies, fades, soft lighting. Raise the dither with --dither 0.8. If instead what you are seeing is the dither pattern itself, as a fine speckle over flat areas, lower it with --dither 0.3. The default is 0.5 and suits most footage.

Everything looks worse than it did, across every clip. Check VIDOPTS and VIDASPECT in CONFIG.BAT - those apply to every video in the game. Per-clip settings belong in VIDOPTS_NNN.

A cutscene that cannot start prints a short message on screen and the game carries on without it. It is a diagnostic rather than part of the display, so the game's own text can scroll over it a moment later - look for it as soon as the cutscene fails to appear.

If the encoder itself refuses to make the file, it says why and names the remedies in the same breath - it will not produce a video that cannot play. The two you are most likely to meet are a frame rate below the audio floor (raise --fps) and an uncompressed shape that will not fit at rate, which prints the exact list of shapes that do fit at your chosen width. Take one from that list.

The build's encode pass needs an encoder and ffmpeg, both resolved from TOOLSDIR (see Getting started):

• videnc.exe (tools\vidtools\videnc.exe) - the standalone encoder, which ships with the kit. No Python needed, nothing to download, and it is preferred automatically. It shares tools\vidtools\ with vidtune.exe and runs from that folder's _internal\, so keep the folder whole.
• lib\videnc.py - the same encoder as a script, the fallback if videnc.exe is ever missing. Needs Python 3, Pillow and numpy (pip install Pillow numpy).
• ffmpeg (tools\ffmpeg\bin\ffmpeg.exe) - required either way, for reading the source video. tools\README.txt has the download.

None of this is needed for pre-encoded .vid files - a kit with only .vid sources, or with no VIDEO\ at all, builds with no encoder present.

Run the encoder by hand for per-file control or to cut a clip from a longer source:

videnc.exe INPUT.mp4 VIDEO\001.vid --shape classic --start 00:00:03 --duration 4

python lib\videnc.py ... takes the same options. The ones this document uses:

Video format lists the full set, and documents every option and the NXV file layout; Video delivery covers how a clip is loaded and played, and what happens to the screen and to game audio around it.

Per-clip tuning has a GUI too: run VIDTUNE.BAT. Pick a clip, preview a segment, adjust shape, frame rate, dither and the other knobs above, encode and accept - the settings land in VIDOPTS_NNN in CONFIG.local.BAT, so the next build reproduces the same encode.

A clip whose settings you have changed but not saved shows as edited in the clip list. Revert throws those changes away and goes back to the saved settings. Encode Stale + Edited encodes every stale or edited clip with the settings you gave it, and saves them to CONFIG.local.BAT the same way Accept does.

VIDEO\001.mp4 and VIDEO\002.mp4 build with the kit so the video condacts work out of the box. 001 encodes at the full 320x256 default; 002 has SET VIDOPTS_002=--shape 16:9 in CONFIG.BAT, the kit's one worked example of a per-video option. Try the verbs MOVIE, which plays 001.VID once, and REEL, which loops 002.VID until a key is pressed.