ffmpeg-filter-compose
v1.3.0
Published
FFmpeg filtergraph generator
Downloads
8
Readme
FFmpeg-Filter-Compose
Filtergraph generator with composing API for FFmpeg.
API References: https://xeroalpha.github.io/ffmpeg-filter-compose/
Install
npm i ffmpeg-filter-compose
We recommend to use it with ffmpeg-filter-compose-extension!
Examples
Blue parts of the graph is auto-generated by
filterComplex
.
Basic chain:
filterComplex(({ from, input, filter }) => {
const [videoOutput] = from(input[0].v)
.pipe(filter.crop([400, 200, 12, 34]))
.pipe(filter.scale({ w: 200, h: 100 }));
return { videoOutput };
})
// => `[0:v]crop=400:200:12:34,scale=w=200:h=100[videoOutput]`
Multiple inputs and outputs:
filterComplex(({ from, input, filter, split }) => {
const [{ v: video }] = input;
const [main, tmp] = split(video);
const [flip] = from(tmp)
.pipe(filter.crop(['iw', 'ih/2', 0, 0]))
.pipe(filter.vflip);
const [out] = from(main, flip).pipe(filter.overlay([0, 'H/2']));
return { out };
})
// => `[0:v]split[_1][_2];[_2]crop=iw:ih/2:0:0,vflip[_3];[_1][_3]overlay=0:H/2[out]`
Apply filter conditionally:
await filterComplex(async ({ from, input, filter }) => {
let video = input[0].v;
if (await fetchNegate()) {
[video] = from(video).pipe(filter.negate);
}
return { video };
})
// => `[0:v]negate[video]` or `[0:v]null[video]`
Apply fade-in effect, then concatenate videos:
filterComplex(({ from, input, filter, concat }) => {
const inputArr = input.toArray(videoCount);
const [outv, outa] = concat()
.video(...inputArr.map((e) => e.v).map((v) => {
const [fade] = from(v).pipe(filter.fade({ t: 'in', d: 1 }));
return fade;
}))
.audio(...inputArr.map((e) => e.a));
return { outv, outa };
})
// => `[0:v]fade=t=in:d=1[_1];[1:v]fade=t=in:d=1[_2];[_1][0:a][_2][1:a]concat=n=2:v=1:a=1[outv][outa]`
Complex example:
filterComplex(({ from, use, pipe, filter }) => {
const L1 = pipe('L1');
const L2 = pipe('L2');
use(filter.testsrc)
.link(use(filter.split).connect(L1))
.link(use(filter.hflip).connect(L2));
const [out] = from(L1, L2).pipe(filter.hstack);
return { out };
})
// => `testsrc,split[L1],hflip[L2];[L1][L2]hstack[out]`
Automatic split insertion:
filterComplex(({ from, use, filter }) => {
const [testVideo] = use(filter.testsrc).mark('video');
const [flipped] = from(testVideo).pipe(filter.hflip);
const [out] = from(testVideo, flipped).pipe(filter.hstack); // testVideo is used twice!
return { out };
})
// => `testsrc[_1];[_3]hflip[_2];[_4][_2]hstack[out];[_1]split[_3][_4]`
Send commands:
filterComplex(({ from, input, filter, command }) => {
const atempo = filter.atempo().ref('my');
const asendcmd = command(({ when }) => {
when(4.5).onEnter(atempo, 'tempo', 1.5);
}).toFilter('audio');
const [out] = from(input[0].a).pipe(asendcmd).pipe(atempo);
return { out };
})
// => `[0:a]asendcmd='4.5 [enter] atempo@my tempo 1.5',atempo@my[out]`
Deferred generation:
const context = filterComplex();
let chain = context.from(context.input[0].v);
chain = chain.pipe(context.filter.setpts('N/(60*TB)'));
// ......
const [out] = chain;
context.complete({ out });
// => `[0:v]setpts=N/(60*TB)[out]`
Extension:
declare module 'ffmpeg-filter-compose' {
interface FilterComplexContext {
select: (input: Pipe, expr: string) => Iterable<Pipe>;
}
}
FilterComplexContext.select = function*(input, expr) {
const filter = this.filter.select();
const chain = this.from(input.mark('video')).pipe(filter);
let outputCount = 0;
for (const pipe of chain) {
outputCount++;
filter.setArguments({ n: outputCount, e: expr });
yield pipe;
}
}
filterComplex(({ from, input, filter, select }) => {
const [odd, even] = select(input[0].v, 'mod(n,2)+1');
const [tmp] = from(odd).pipe(filter.pad({ h: '2*ih' }));
const [out] = from(tmp, even).pipe(filter.overlay({ y: 'h' }));
return { out };
})
// => `[0:v]select='n=2:e=mod(n'\\,'2)+1'[_1][_2];[_1]pad=h=2*ih[_3];[_3][_2]overlay=y=h[out]`
FAQ
Q: Why can't I use the spread operator (...
)?
A: The spread operator (...
) cannot be used in this context because we don't have knowledge of the exact input count or filter pad count. Since they are infinite iterable, using the spread operator could potentially produce an infinite number of objects. To address this, you can use the toArray()
or fork()
methods to explicitly specify the count of objects.
Q: What is the purpose of pipe.mark()
? When should I use it?
A: The pipe.mark()
function is used to specify the media type of a pipe, which helps determine the relevant filter to be applied. For example, when using the split()
function, specifying the media type with pipe.mark()
ensures that the appropriate filter, such as split
for video or asplit
for audio, is applied.
Q: How can I resolve the error message Pipe [pipe] is not bound to any input
?
A: This error message indicates that the [pipe]
object is neither passed to from().pipe
nor returned from the callback function. To resolve this issue, you should explicitly call the recycle()
method with the [pipe]
object to ensure it is properly handled.