Useful: the Hiroska blog

Why AI video clips are only a few seconds long, and how to make a longer video anyway — The first thing everyone asks after their first generation is "how do I make it longer". The honest answer is that you do not, and that the question is the wrong one. A film is not a long clip. It is short ones, arranged.
How to remove the music from a video and keep the voice — A video with music baked into it used to be finished, for better or worse. Now the music comes out. Here is how the separation works, where it fails, and what to do with the voice you get back.
Why your AI video sounds dead, and the three layers that fix it — You can tell an AI video with your eyes closed. The picture may be flawless; the sound is a voice, some music, and a silence under both of them that the audience feels as "nothing is really there".
How to mix voice over music so both can be heard — Every beginner mix has the same two faults: the music fights the voice, and the whole thing plays quieter than the video before it. Both have simple causes, and neither is fixed by turning something up.
How to cut the ums and pauses out of a recording without cutting the person — A recording with every pause and every um removed does not sound professional. It sounds like a hostage tape. The skill is in knowing which silences are dead and which ones are the person thinking.
How to keep the same character across shots — Prompting the same description twice gives you two different people, because a description is not an identity. The fix is structural: make the character once as an image, and let every shot afterwards start from that image rather than from words.
How to export for YouTube, TikTok and everywhere else — Every platform re-encodes whatever you upload, which decides most of the export dialog for you: give it the best master you can and let it be destroyed once instead of twice. The rest is frame rate, codec and knowing when Match source is the answer.
How to put your webcam on a screen recording — Two files, one timeline: sync on the audio, float the webcam as an overlay, and keep it small enough that it never competes with the thing you are demonstrating. The mistakes are all in sizing and placement.
When to use a transition — and when to just cut — The transitions menu is where new editors spend their time and experienced ones spend almost none. Not because transitions are bad, but because a cut is invisible and everything else is a statement — so each one has to be saying something.
How to stop your generations looking like everyone else's — Everyone is prompting the same handful of models, so everyone gets the same house look. A style is the fix, and there are three ways to get one — only the most expensive of them involves training anything.
How to blur a face or a licence plate in a video — A blur region does not follow a moving subject, and that single fact decides the whole technique: split the clip where the subject moves and blur each piece. Also, what blurring genuinely hides — and what it only appears to.
Titles that do not look like a template — White text, centred, fading in, in whatever font came first: the look of a video nobody art-directed. Fixing it is four decisions — typeface, size, contrast against the picture, and being ruthlessly consistent about all three.
How to make slow motion that does not stutter — Slow motion is not a speed setting, it is a frame-supply problem: a 24 fps clip at quarter speed has six real frames a second to show you. Interpolate first, slow down second — the other order gives you a smooth slideshow.
How to score a video without licensing a track — Library music sounds like library music and a commercial track gets your upload claimed. Generating the bed solves both — but the part that actually changes how a video feels is the ambience underneath it, which no generated clip comes with.
How to narrate a video in your own voice — Voice cloning is not really about sounding convincing — it is about what happens when a sentence changes. The value is the re-record you never have to do, and the setup takes about thirty seconds of clear speech.
How to translate a video into another language — There are three levels of translating a video, and they are not upgrades of each other — subtitles win on cost and reach, a lip-synced dub wins on attention. The traps are length drift, on-screen text, and dubbing before the edit is locked.
How to clean up a talking-head recording — A forty-minute self-recorded take becomes a tight twelve minutes in seven passes. The order matters — two of the passes are free, one is billed by length, and doing them backwards costs you both money and a re-run.
How to remove an object from a video — There are two ways to take something out of a shot and four ways to make it stop mattering. The two removals are the expensive ones, and they are not always the ones that look best.
How to remove a video background without a green screen — You shot at your desk, not in front of a green wall. There are three ways to lift the subject off that background, they cost wildly different amounts, and the right one is decided by the footage rather than by preference.
How to make a vertical video from horizontal footage — Cropping 16:9 to 9:16 throws away two thirds of the width. There are three ways to fill a vertical frame, and choosing per shot is the difference between a reframe and a decapitation.