Input Validation
Before VidRush spends a single credit, it reads what you gave it. If the input would produce an unwatchable video, you find out in seconds instead of an hour later.
Nothing you submit is charged until a video completes, so a blocked check costs you nothing but the time it takes to fix it.
What validation does and does not tell you. It confirms two things: that VidRush can physically produce a video from what you supplied, and that the content is allowed. It says nothing about whether the result will be good. A prompt can pass every check and still produce a weak video, and a Custom Script can pass and still fail later in generation if it is not written the way the narration pipeline needs. Passing means “nothing blocking found”, not “this will work well”. See Prompting and Custom Script.
There are four separate checks, one for each way you can supply content. They share the same three outcomes and the same generous instincts, but their rules really do differ. A script that passes as narration can be rejected as a prompt, and fiction is accepted from an audio upload but not from a typed script.
| What you supplied | What is checked | Where the limits live |
|---|---|---|
| A prompt | Your brief | This page |
| A Custom Script | Your narration text | Custom Script |
| A Custom Voiceover | Your recording, transcribed | Custom Voiceover |
| A Talking Head upload | Your file, your footage, and your speech | Talking Head |
If your video uses an AI Avatar, a stricter rule set is added on top of whichever check applies. See Avatar rules below.
Three possible outcomes
Every check ends in one of three states.
| Outcome | What it means | What happens next |
|---|---|---|
| Passed | Nothing to flag | You continue. No message appears. |
| Warning | Something is worth knowing, but it will still produce a video | You choose: refine your input, or proceed as-is |
| Blocked | The result would be unusable, or the content is not allowed | Generation does not start. Fix the issue and submit again. |
Each message tells you the problem in one sentence and the fix underneath it, under How to fix this. You will normally see one issue at a time rather than a full list, so clearing one can surface another on your next attempt.
Every check is deliberately generous. If something is borderline, it goes through. A block means VidRush is confident the output would be wasted, not that it is being cautious.
Three or more warnings at once become a block. Individually these issues are survivable. Stacked, they reliably produce an unusable video. When this happens you are asked to simplify rather than being shown the full list.
Supported formats
All four checks draw the same line on what VidRush can produce. Five formats:
Documentary · Explainer · Essay · News · Top 10
Rejected everywhere: compilations, commentary and reaction videos, 3D and animation, advertisements, tutorials, and fictional stories.
The line is drawn by what the video is, not what it is about. A documentary about a fictional universe is fine. An explainer about how 3D films are made is fine. See Choosing Format & Length.
Tutorials are not supported, including on talking-head projects. You can film yourself explaining something and have it edited as a documentary or an explainer, but a recording structured as a step-by-step tutorial is rejected on its speech content.
One exception, in one direction: Custom Voiceover accepts fiction. Narrative stories, character monologues, and serialised episodes all work when you upload your own narration, because your voice is paired with atmospheric footage rather than a literal depiction of the story. The same story typed into Custom Script is rejected. See Custom Voiceover.
Content rules
Prohibited content. Graphic violence for entertainment and explicit sexual material are refused. Educational and documentary framing of difficult subjects is explicitly accepted, so a serious historical treatment of a violent event is fine.
One rule has no exemption. Content involving child sexual abuse, child nudity, the sexualisation of minors, or child exploitation is refused in every framing without exception, including news, documentary, court reporting, scandal coverage, and awareness content. The same applies to explicit descriptions of sexual violence. There is no wording, no context, and no purpose that makes this pass.
Requests about VidRush internals. Asking the AI to reveal its own instructions or system configuration is not a video topic and is refused.
Political propaganda. Refused only in clear cases, meaning explicit calls to action or recruitment. Politically informative content and commentary are accepted.
Warnings you can proceed past
These do not stop generation. They are worth reading, because most of them predict a weaker video. They apply to prompts, scripts, and uploaded narration alike.
| Warning | What it means | How to fix it |
|---|---|---|
| Vague or unclear input | There is not enough direction to build a script around | Add a specific topic, angle, or conflict |
| Low footage availability | Your topic is niche enough that suitable B-roll may be thin | Broaden the angle slightly, or add research links |
| Hyper-specific visuals | You asked for a shot that almost certainly does not exist as footage | Describe the subject rather than the exact shot |
| Fictional scenario | The premise has no real-world footage to draw on | Frame it as a documentary-style explainer |
| Partial script pasted | A stretch of finished narration is sitting inside an idea-level prompt | Move the full script to Custom Script, or trim it to a description |
| Camera or shot directions | Stage directions are sitting inside the narration itself | Remove them. Footage is matched to what is said |
| Over-specified audio or motion graphics | You are directing details the pipeline assigns itself | Leave these to your Brand Profile |
| Unfilmable or private topic | The subject is not something footage exists of | Pick a publicly documented angle |
| Over-specified editing style | The editing instructions contradict each other, or cannot be done | Simplify to the effect you actually want |
| Content and framing mismatch | The tone you asked for does not fit the subject | Align the two, or accept the result |
| Not ready to be read aloud | URLs, timestamps, or formatting artefacts that will be narrated literally | Strip them out |
| ”As you can see here” | The text points at visuals it cannot control | Describe the thing instead of gesturing at it |
| Language differs from your Brand Profile | Your input is in a supported language, but not your profile’s | Match the profile, or proceed if intentional |
Language
VidRush validates input in English, German, and Spanish. Anything else is blocked and you are asked to rewrite it in one of the three.
Two narrower limits sit on top:
- Talking-head uploads must be spoken in English.
- Avatar videos must be in English, including any script or audio you attach to them.
When input is blocked for language, that is the only thing you are told about. Anything else wrong with it surfaces on your next attempt, after you have translated it.
If your input’s language differs from your Brand Profile’s voiceover language, you get a warning, not a block. English never triggers this, so an English prompt on a German profile passes quietly and produces a German voiceover.
Language is detected by the AI but the decision is made in code, so it cannot be talked around. On talking-head uploads the language is checked twice, from the audio and from the transcript, so a recording that opens in English and switches is still caught.
→ See Multi-Language for supported video languages.
Prompts
Length. Between 200 and 10,000 characters.
A finished script in the prompt field. The prompt box takes a brief, not narration. A word-for-word script pasted here fights the research and scriptwriting stages and comes out garbled. If you have a finished script, attach it properly: click + in the bottom left of the prompt window and select Custom Script.
A prompt that instructs VidRush (“Create a documentary about…”, “Open with this hook…”) is always read as a brief, however long it is, and can quote sample narration freely. Only text written to be spoken aloud to a viewer is treated as a script.
A listicle with the wrong number of items. VidRush allows roughly 1.3 list items per minute. Too many and each item gets a few seconds; too few and each is stretched past what the footage supports.
| Video length | Works well | Maximum items |
|---|---|---|
| 6–8 min | 3–10 | 10 |
| 10–12 min | 4–16 | 16 |
| 18–20 min | 6–26 | 26 |
Lengths between these tiers scale in proportion, so a 14-minute video tops out around 18 items. This only applies when you explicitly request a Top N or a numbered countdown. A general outline with several points in it is not treated as a listicle.
Custom Scripts
Minimum 780 characters, roughly a minute of narration. Below that there is not enough material to build a video from.
A prompt instead of a script. “Make me a video about the Roman Empire” is an instruction, not narration, and is rejected. So is a script with instructions mixed into it. This is the mirror image of the prompt rule above.
Duplicated content. A block of 1,000 characters or more repeated inside the script is rejected with a request to remove it. This is almost always an accidental double paste.
Full formatting rules, length caps, and credit costs: see Custom Script.
Custom Voiceover
Your recording is transcribed first, and the transcript is then checked the same way a typed script is, with the fiction exception noted above.
Minimum 6 minutes. Checked before anything else runs.
Near-silent recordings are rejected: if almost no words come back, there is nothing to validate or build from.
Severely garbled recordings are rejected when the transcript is too corrupted to use.
Spoken instructions are rejected. Recording yourself describing the video you want (“make me a video about…”) is not narration.
Only the opening of a long recording gets the content check, roughly the first 7 to 8 minutes of speech. Length, language, and transcription cover the whole file. A passed check is not a statement about your last hour of audio.
If the content check itself breaks, your upload proceeds without a verdict. Talking-head uploads make the opposite call, where an unfinished check blocks.
File requirements and recording tips: see Custom Voiceover.
Talking Head uploads
The most thorough of the four, because there is a file, footage, and speech to judge. Full detail, including every file and encoding requirement, is on Talking Head. What is specific to validation:
Sensitive live topics are rejected when the recording is centred on one: a current or upcoming election, live public-health guidance, financial advice or stock and crypto promotion, or breaking news. This is deliberately narrow. Historical documentaries and educational explainers that touch these areas are accepted.
Official documents read aloud are rejected when the recording is substantially a verbatim read-out of a contract, terms of service, earnings report, or court statement delivered as if it were an official announcement. A documentary about such a document is fine.
Extreme distress is rejected, meaning crying or screaming for most of the recording.
The “most of the video” principle
Your upload is raw material that gets edited afterwards, so the visual rules judge what the recording is most of the time. A second person stepping into frame briefly, or you walking away for a moment, does not block anything.
One exception: prohibited visual content blocks on a single stretch, however short.
When a problem covers only part of a long recording, the upload is still rejected, but the message names roughly when it happens, accurate to about five minutes, so you know whether to trim ten seconds or re-record forty minutes.
Uncertainty never passes
If the check cannot reach a verdict, your upload is rejected rather than waved through. Two versions of this, and they behave differently:
- The check could not finish. Nothing is known to be wrong with your video. Uploading the same file again can genuinely succeed.
- Something specific was seen but could not be confirmed. Re-submitting the same file returns the same answer. The recording itself has to change.
Some rejection messages end with a line inviting you to upload again to be re-checked. Take it literally: those are judgement calls and a second attempt often clears one. Rejections without that line are measured facts about your file.
Warnings on uploads
Five things warn rather than reject, and all five are safe to proceed with:
| Warning | Note |
|---|---|
| Black bars around the picture | Optional to crop |
| Variable frame rate | Optional to re-export |
| Blurry, soft, or poorly lit but usable | Merely soft warns; too dark or fully out of focus rejects |
| Presenter in an official or professional uniform | Advisory only. Scrubs, a lab coat, military or police uniform, or judicial robes. A real professional recording in work clothing is a legitimate upload |
| Presenter clearly appears to be under 18 | Advisory only, high-confidence cases only. A youthful-looking adult is not flagged |
There is no human review queue on uploads. The message you get is the whole resolution path, and the fix is always to change the recording and upload again.
Avatar rules
When an AI Avatar delivers your content to camera, a stricter set of rules switches on. These apply to the prompt and to an attached Custom Script, identically in both.
A Custom Voiceover cannot be used with an avatar at all. The avatar speaks with the voice you selected for it, so uploaded audio and avatars are mutually exclusive. See AI Avatar.
They exist because a synthetic presenter saying something carries weight a narrator’s voice does not. A viewer who sees a person say it treats it as a person saying it.
Avatar mode relaxes one thing first. First-person, presenter-led delivery is the expected input here. “Hey everyone, let’s get into it”, direct address, and monologue-style writing are never rejected as an unsupported format.
It then adds nine rules. Unlike everything else on this page, these are not generous. When one is clearly triggered, it blocks.
| Rule | What it covers |
|---|---|
| Fake authority | The avatar gives advice, orders, or official announcements as a doctor, lab worker, military or police officer, judge, or government official. The persona does not have to be declared. Prescriptive medical or legal instructions delivered in an expert tone are enough, because the avatar delivering them becomes a fake expert. A general informational health documentary is not this. |
| Unauthorised spokesperson | Reading out a legal contract, terms of service, corporate earnings report, official corporate announcement, or court statement. Reading such a document at all is the failure, because the reading itself implies authorisation. |
| Phishing-style interaction | Simulating a live agent such as a bank teller, government agent or tech support line, prompting viewers to verify accounts, share credentials or payment details, call a number, or install software. |
| Shock-value distress | Directing the avatar to cry, scream, or act hyper-panicked, including content written to be performed that way without saying so. |
| Impersonating a real person | Delivered as a specific named politician, celebrity, or executive in first person. Talking about real people in third person is fine. |
| Emergency sound effects | Air-raid sirens, official emergency-alert tones, police sirens, or disaster warning sounds behind the avatar’s speech. Always a block in avatar mode, never softened to an audio warning. |
| Actionable regulated advice | Direct financial or medical instructions tied to real-time events or specific tradable assets: “buy this coin now”, named tickers to trade, concrete treatment or dosage instructions. |
| Cloned real voice | Asking for delivery in a specific real person’s voice, or pairing the avatar with a cloned voice of a public figure. |
| Live volatile topics | Current or unfolding instances of the above: live price movements of named assets, a new or ongoing disease outbreak and its emergency medications, or ongoing wars and disasters covered as breaking news. Historical and retrospective coverage is fine, so “the history of Bitcoin” or “the 2008 financial crisis” is at most a warning. |
Framing does not exempt anything
This is the rule people get caught by. All nine apply to every word, including:
- passages inside quotation marks
- passages framed as a “sample”, an “example”, or a “reading”
- content carrying a disclaimer such as “this is fictional”, “for testing purposes”, or “not investment advice”
- the same content scattered among many unrelated topics
The finished video delivers the same message to a viewer regardless of how the input was framed, so it is judged as if the quotation marks and the disclaimer were not there.
The sensitive-topic warning
An avatar video substantially about elections, public health, finance, or breaking news gets a warning and proceeds. It is a heads-up that platforms scrutinise AI-presenter videos on these topics heavily.
These four are read strictly. “Finance” means markets, investments, and money advice, not any story where money is mentioned. “How LEGO became a $10B empire” or “the rise of electric vehicles” is a company or industry story and passes with no warning.
Avatar mode is switched on by your project, not by your words. Writing “avatar” inside a prompt or a script does nothing, so a documentary about AI avatars is checked under the normal rules.
If the check cannot finish
Occasionally a check does not reach a verdict and you are told the validation service is temporarily unavailable. Nothing is known to be wrong with your input. Wait a moment and submit again.
On talking-head uploads you may also be told the check timed out or that a temporary system issue occurred. Same thing: nothing is wrong with your file. Our own failures are never reported to you as a problem with your content.
Good to know
- Blocked input costs nothing. Credits are deducted when a video completes. Input that never enters the pipeline never draws down your balance. See Credits & Billing.
- Silence is a pass. If no message appears after a check runs, you are clear to continue.
- Warnings are advice, not verdicts. Proceeding past one is a legitimate choice. You know your niche better than the checker does.
- A pass is not a quality guarantee. The check confirms your input can produce a video, not that it will produce a good one. → See Prompting for what makes a strong brief.
- You are shown one issue at a time, the most severe one, rather than a full list. Fixing it can reveal the next, so a submission may take more than one pass.
- The rules are not the same across the four inputs. If something was accepted as a Custom Voiceover and refused as a Custom Script, that is the design, not a bug.