How are youtube caps paused on the clearest and
 

most interesting frame?

Magic....

it's called i am big fucker

you have no clue do you? hehe it's magic:1orglaugh

what ev

It's relatively simple... if you have a log/petitions processing and thumbnailing server because using the same server for streaming and thumbnailing dinamically would make your server go a bit crazy.

As you may have seen youtube uses streaming like video delivery, you can look at the video end without loading the entire flv.
Every time yo click and load a specific video position the server delivering the stream gets a petition for that specific position. To save server resources videos are segmented in key positions that are several seconds long, usually between 5 and 10 seconds, but this is completely adjustable value.
After that it's just a matter of math... calculate how many times a video position is requested... then make a screencap of the most requested position for a given video.

Can you do that too? Probably, if your coder knows how to do it and your server admin is smart enough.
Is it worth it? I know how to do it and I don't do it on my sites... so my opinion is that it's not worth it :)

i'll ask again. Why it look good?:(

Maybe you are looking at YouTube Partner videos, which partners are able to upload their own thumbnails and not rely on the Youtube's new algorithm that selects the thumbs from random spots.

Or maybe you are looking at older videos which did not use this new algorithm and selected the thumbnail from 3 predetermined points within the video, thus enabling the video producer to insert still images into those predetermined spots, thus producing a nice clear thumb of their choice.

cool but why do you never see a wall or other unrelated frames then? knowng there's no time to look abd choose?

I''l

not sure what your question is and what instance you are referring too

you do occasionally see a wall or unrelated frames

Because their thumbnail software can take good thumbs from a movie... you know... take a thumb, then apply a couple of filters and its done.

get a basic video/image editing course and you will know to do it too.

There is quite a bit more to it than that, they don't have 100s of PhDs sitting there doing nothing.... :2 cents:

I'm not sure but I guess they might be using something like an algorythm to search for noiseless keyframes that include some sort of movement or something like that. Then again, just assuming, I know it can be done that way, now doing it is a whole different story