Talking Head
Record yourself talking to camera, upload it, and VidRush builds the video around you. Your footage becomes the A-roll: VidRush tightens the pacing, cuts the dead air, layers B-roll over your narration, adds captions and your theme, and assembles the finished video.
This is the option when you want to be on screen. Everything after the upload works the way it normally does: prompt field, reasoning effort, Quote Statement, and editor.
How to use Talking Head
- Open the prompt window and select Upload Talking Head, either from the tab above the input box or the + button in the bottom left.
- Upload your recording.
- Write your prompt.
- Pick your reasoning effort.
- Review the Quote Statement and approve.
VidRush takes it from there.
What to put in the prompt
The prompt works differently here. You already supplied the content by recording it, so the prompt is context rather than a brief: what the recording is, who it is for, and what you want the finished video to do.
You do not need to ask for the editing itself. Cutting filler and dead air, choosing supporting footage, pacing, captions, and audio cleanup all happen automatically. Asking for them changes nothing.
What the prompt does control:
| You want | Say this |
|---|---|
| A specific length | ”Make this a 10 to 12 minute video.” Available bands are 6–8, 10–12, 18–20, and 30–40 minutes. |
| A tighter cut | ”Keep it tight” or “just the essentials” |
| To keep as much as possible | ”Comprehensive, keep the full argument” |
| Your intro removed | ”Remove the greeting and intro, start on the main point.” Nothing else will cut your intro. |
| A Top 10 structure | Include “top 10” in your prompt |
| A specific focus | ”Focus on X, cut the tangent about Y” |
Anything you state directly wins over what VidRush infers. A vague prompt gets default treatment. Length and format can still be changed in the Quote Statement before you approve, and what you approve is what gets made.
You cannot choose a length longer than your recording, and in practice the finished video runs to about 90% of it, because silences and weak clips are removed. A 20-minute recording can be cut to the 18–20 band but not to 30–40, and the same limit applies in the Quote Statement.
The prompt checker is more permissive on a talking-head project. A tutorial, a piece of commentary, or any other presenter-led format is accepted here, where the same request is refused on a standard prompt. Content rules still apply exactly as they do everywhere else.
Prompt templates
Condensed and refocused
Turn this into a tight 10 to 12 minute documentary-style video. Keep my
greeting and intro. Focus on the three pricing mistakes and cut the
tangent about the conference. Keep my closing call to action intact.No intro, Top 10
Make this a punchy top 10 video, around 6 to 8 minutes. Remove the
greeting and intro and start directly on the first item. Prioritise the
segments with specific numbers.What Talking Head cannot do
Asking for these in the prompt will not make them happen.
- Run on Mini. Mini is an image-only model, so talking-head projects use Deep Video v1 Pro.
- Cut your recording into clips. One recording produces one finished video. VidRush does not chop a long recording into shorts, highlights, or a set of social cuts.
- Turn a long recording into a short one. The minimum upload is six minutes and the output is long-form. Asking for a 60-second version, or naming a target length, will not shrink it. The edit removes dead air and filler, and that is what sets the final duration.
- Produce vertical or square video. Output is widescreen. Reframing for other placements is not available yet.
- Translate or dub. English in, English out. No translation or dubbing is provided. See Multi-Language.
- Generate a presenter. Talking Head edits footage of a real person. If you want someone on screen without filming, see AI Avatar.
- Handle more than one speaker. Podcasts, interviews, and panel recordings are rejected on upload, not trimmed down to one person.
Recording requirements
| Requirement | Details |
|---|---|
| Formats | MP4, MOV, WebM |
| Video encoding | H.264, HEVC, VP9, or AV1 |
| Audio encoding | AAC, MP3, Opus, or PCM |
| Min duration | 6 minutes |
| Max duration | 2 hours |
| Max file size | 5 GB |
| Files per project | One |
| Spoken language | English |
Talking-head uploads must be in English, even though prompts themselves are accepted in English, German, and Spanish. A German or Spanish recording is rejected. See Multi-Language.
How to record for best results
Your footage is the spine of the video, and nothing downstream can fix a recording that does not work.
| Do this | Why it matters |
|---|---|
| Pause between thoughts | Cuts land in the silences. Talk without breaks and there is nowhere clean to cut. |
| Redo a flubbed line after a pause | Both attempts are found and the better one kept. No need to restart the recording. |
| Stay silent about mistakes | ”Cut that” and “take two” are treated as speech. Silence is the signal. |
| Look at the lens | Holding the lens keeps you on screen. Looking away for long stretches cuts to other footage. |
| Stay in frame and on camera | Slideshows and screen recordings are rejected, and long stretches with nobody visible get flagged. |
| Light your face and check focus | Too dark or too soft is rejected outright. Merely soft gets a warning and proceeds. |
| Frame yourself off-centre | Captions and graphics need room, and are never placed over your face. |
| Get the microphone close | The transcript drives captions, footage choice, and pacing. Weak audio degrades all three. |
| Open with a real intro, close with a payoff | Your intro survives the edit unless you ask for it to be removed, and a clear close keeps you on screen for it. |
| Drop strong lines anywhere | Bold claims are found wherever they occur and can be promoted to the opening, so nothing is wasted by arriving late. |
| Record more than you need, stay on topic | The edit condenses to your target length. Structure earns a longer video; rambling gets cut hard. |
What happens automatically
None of this needs to be asked for.
- You appear on screen for the moments that carry you: hooks, personal statements, your closing. Supporting footage covers the rest.
- Re-hooks bring you back on screen at turning points to reset attention.
- Text and graphics are placed so they never cover your face.
- Audio is cleaned, levelled, and brought to broadcast loudness.
How your recording is checked
Your recording is checked as it uploads, before anything is generated. The check ends in one of three ways.
| Outcome | What it means |
|---|---|
| Passed | Nothing to flag. Your video continues into the queue. |
| Passed with a warning | Something is worth knowing, but the upload proceeds. |
| Rejected | The recording cannot be used as it is. Fix the issue and upload again. |
Three things are examined: the file itself, what the camera shows, and what you say. Each message names the problem in one sentence, with the fix directly underneath it.
- The file. Length, size, format, encoding, and whether it has both a picture and a sound track.
- The camera. Whether a single presenter is visible, in focus, adequately lit, and doing nothing prohibited.
- The speech. Whether it is English, whether it is a real spoken presentation, and whether anything said is prohibited.
Only three things produce a warning rather than a rejection: low picture quality, a variable frame rate, and black bars around the picture. All three are safe to proceed with, and the fix is offered as optional.
When to upload the same file again. Some rejections end with a line inviting you to upload again to be re-checked. Take it literally. Those are judgement calls, they are not identical every time, and a second attempt often clears one. Rejections without that line are measured facts about your file, so the same file will always give the same answer, and re-exporting or trimming is the only way forward.
Good to know
- A problem in one stretch still blocks the upload, but you only have to fix that stretch. The message names roughly when it happened, accurate to about five minutes: “The picture is too blurry or too dark around 5:00–10:00.” Re-record or trim that section rather than starting over. If more than two stretches are affected, the message names the first two and says “and elsewhere”.
- Your Brand Profile still applies. Theme, backgrounds, captions, compliance settings, and blacklists all shape the B-roll and the styling laid over your footage. Your own footage is yours and is not subject to sourcing tiers, but everything added around it is. See Compliance & Footage Sourcing.
- Uploading someone else means having their permission. If the person on camera is not you, make sure you have the right to use their likeness before you upload.
- A rejected upload costs nothing. Credits are deducted when a video completes. See Credits & Billing.