How to edit a talking-head video into a 4K YouTube master and shorts

Last updated 2026-08-23 · 6 min read

What is the short version?

A finished YouTube upload is not one file, it is a set: the full-resolution master, the vertical clips that feed it, and the thumbnails that get it opened. All three come out of the same edit if the edit is built at full resolution from the start.

1. Go to Vizard Agent and upload the take at its original resolution.
2. Say what must be preserved — the format, the frame rate, the colour, the audio.
3. Say what you need: the master, the shorts, the thumbnails.

What do you need before you start?

The take and your constraints. Vizard Agent finds the edit points from the transcript and sources the supporting material, and the things it cannot infer are your non-negotiables — the resolution, the colour, the processed audio you already mixed.

• The take. At its native resolution; 4K is fine.
• What must not change. Format, frame rate, colour, audio.
• Your logo. For the corner watermark.
• The language. It decides the caption typography.
• The deliverables. Master, shorts, thumbnails, subtitle tracks.

What do you type into Vizard Agent?

List your constraints explicitly in the instruction. Vizard Agent treats a line like "keep the original 4K, 30fps, colour and processed audio" as a hard limit on every stage of the build, which is what stops a pipeline quietly normalising footage you already graded and mixed.

Prompt

Variants worth knowing:

• Two subtitle tracks. Burned in one language, muxed as a selectable track in another.
• Infographic overlays. For anything the speaker explains geographically or numerically.
• Karaoke captions. Word-level highlighting suits the shorts more than the master.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real YouTube edit. Two long stretches of it stand out: five separate passes refining a transparent logo until the matte was clean, and a full typography test before a single caption was written.

1. Probes the file and inspects the attached logo.
2. Transcribes the speech and extracts detailed word timestamps, then verifies their length against the take.
3. Looks at a frame of the speaker to assess the composition and framing.
4. Builds a transparent logo through five passes — cropping, refining the matte, optimising the alpha and cleaning isolated specks — checking each result.
5. Designs the edit points for clean speech and evaluates the proposed spans against the transcript.
6. Searches and downloads stock b-roll and builds a map graphic template.
7. Downloads high-grade typefaces for the language and installs the text-shaping libraries, then tests each font's rendering on a frame.
8. Tests karaoke highlighting and phrase-level styling and inspects the rendered results.
9. Renders the 4K master, reviews quality-control frames, re-lays the infographic so it never overlaps the speaker's face, then renders five vertical shorts, three thumbnail concepts and a second subtitle track.

Step 9's infographic fix is a rule worth generalising. An overlay that covers the speaker's face breaks the one thing a talking-head video has going for it, and checking that means looking at the rendered frame rather than the layout file.

What does the result look like?

From the run this page is written from, probed on the delivered master: 3840x2160, H.264, 30fps, 205.0 seconds, AAC audio. Full 4K, three and a half minutes, with the original colour and processed audio preserved, plus five vertical shorts, three thumbnail concepts and a selectable second subtitle track.

Delivering at 3840x2160 rather than a downscaled proxy is the point of the brief. Everything — the overlays, the graphics, the b-roll — was built at that size so the master never has to be upscaled back to it.

When does this not work well?

Working at full resolution throughout is slower and considerably stricter than working at a proxy size and conforming at the end. Vizard Agent handles it without complaint, and the constraints that choice creates are real ones worth knowing about.

• 4K rendering takes time. Every overlay is composited at full size.
• Complex scripts need font work. Vizard Agent downloads and tests typefaces, and unusual scripts are still the risk area.
• Preserving the audio means preserving its flaws. If your mix has a problem, it survives the edit.
• Cutting for clean speech changes the pacing. Removing every stumble can make a delivery feel clipped.
• Thumbnails are concepts. They are a starting point, not a finished design system.

How do you fix a result that came back wrong?

Name the element. Vizard Agent keeps the word timings, the keep-spans configuration, the transparent logo, the tested typography and the overlay layouts, so a change to the cut or a caption style is a re-render against work it already did.

• "Do not cut that pause." Restored from the spans configuration.
• "The overlay covers my face." Re-laid out and verified on frame, as here.
• "Use the other font." Already tested and rendered for comparison.

How does Vizard Agent compare to doing it yourself?

By hand this is cutting every stumble out manually, keying a logo in a compositor, discovering that your captions render in the wrong shape for the language, and then exporting the whole set over again for the shorts.

Common questions

Why build everything at 4K instead of conforming later? Because an overlay composited at 1080 and scaled up to 4K is visibly softer than the footage around it. Vizard Agent generates the graphics, the infographics and the b-roll at full size so the master never has to upscale anything.

Will Vizard Agent keep my 4K? Yes. Vizard Agent builds every element at full resolution rather than upscaling anything at the end.

Does Vizard Agent preserve my audio mix? Yes, if you say so. Vizard Agent carries the processed audio through the edit untouched.

Can Vizard Agent make the shorts too? Yes — five on this run, cut from the same master edit.

Does it make thumbnails? Yes, as concepts — a starting point rather than a finished design.

Why test the fonts first? Because a typeface that cannot shape your language produces text that is present and wrong. Rendering a frame and looking at it is the only way to know, and it costs a minute against a full 4K re-render.

Does Vizard Agent build the shorts from the master? Yes. The five vertical shorts come out of the same edit rather than being cut separately, so the framing and the captions stay consistent with the master.

Can I get two subtitle languages? Yes — one burned in, one muxed as a selectable track.

Will overlays cover my face? Vizard Agent checks the rendered frames and re-lays anything that does.