AI News Video Generator

Documentary stock footage is the default here and the cheapest look, at about 12 credits a video. It builds the video from real stock clips, licensed for commercial use with nothing to attribute, which suits stories set in real places: a city, a port, a cathedral, an airport. Its limit matters more in news than anywhere else: stock libraries hold the place and the subject in general, not the event. A clip of a flooded street shows some other flood. Use footage for the setting, and keep anything that could pass for the event itself out of the video.

Mixed, typically about 14 credits, puts AI images between the clips for the parts of a story that have no footage: an idea, a mechanism, a number that changed. Keep those images illustrative. A realistic AI image next to real footage is easy to take for a photo of the event, which is the case YouTube's AI disclosure and TikTok's AI rules are written for (see below).

Flat Vector draws clean shapes in the science-explainer register, and it suits the 'how does this work' half of a story: how a ballot is counted, how a software update spreads, what a new law changes. A flat drawing can show a mechanism without anyone mistaking it for footage. About 25 credits.

Neon Cyberpunk is the tech-news look: glowing screens, server rooms and a futuristic city at night. Its scenes look like neon-lit film stills, so keep real people and real events out of its images and name them in the narration instead. Miniature World turns each scene into a tilt-shift toy diorama, which makes economy news easy to follow: a port, a market, a town's shops changing their price tags. Both typically cost about 25 credits.

Retro Poster's mid-century screen-print look suits 'this week in history' recaps, such as Sputnik 1's launch on 4 October 1957. The style tells viewers at a glance that the story is from the past, not today. About 25 credits.

Flat Vector, Neon Cyberpunk, Miniature World and Retro Poster can also turn every image into a short AI video clip, at about 464 credits a video; news dates quickly, so that's rarely worth it. Photoreal looks such as Cinematic or AI Images are the weakest fit: a realistic AI scene of a real event is exactly what the platforms' AI rules are about.

"News recap: fire brought down Notre-Dame's spire on 15 April 2019, and the cathedral reopened in December 2024. Explain what was rebuilt, citing the restoration body's figure of about 1,000 oak trees for the spire and transept frameworks. Documentary footage of the cathedral and Paris today, calm narrator, end on the work that continued after the reopening."

"Tech news recap: at 04:09 UTC on 19 July 2024, a faulty CrowdStrike update crashed Windows computers worldwide. Microsoft estimated 8.5 million devices were hit, under 1% of all Windows machines. Say it was a bad update, not a cyberattack, as CrowdStrike stated. Neon Cyberpunk look, symbolic scenes only, no logos or real people. End on how one update reaches millions of machines."

"Economy news recap: on 1 January 2026 Bulgaria switched from the lev to the euro at a fixed 1.95583 lev per euro; name the ECB as the source. Miniature World look: a toy town where price tags show both currencies, bank accounts convert automatically, and lev and euro cash both work for one month. Friendly narrator, explain what changed for shoppers, no forecasts."

Each fits the 400-character limit of the box above and covers something that has already happened, with its date. Ask for the source in the narration, as these do, then check that the draft kept it and that every figure matches the source before you render.

One story per video, in a fixed order: what happened, with its date and its source said in the narration; why it matters; then what's confirmed and what isn't. 'On 19 July 2024, a faulty update from the security company CrowdStrike affected 8.5 million Windows devices, by Microsoft's estimate' is a first line a viewer can check. Work from the source closest to the event, such as the agency's statement, the company's incident report or the official results, rather than another outlet's article about it.

Write it in your own words. YouTube's monetisation policies list 'content that exclusively features readings of other materials you did not originally create, like text from websites or news feeds' as reused content, and the policy applies 'even if you have permission from the original creator'. An AI voice reading out an article is that; your account of what the source says, in your own sentences, isn't. Read the source, close it, write the story the way you'd tell it, then check each fact against the source.

Keep the words that carry uncertainty. Microsoft wrote 'we currently estimate' when it put the number of affected devices at 8.5 million, so the narration says 'Microsoft estimated', not '8.5 million computers crashed'. Give the time zone when the hour matters: Microsoft's post dates the faulty update to July 18 and CrowdStrike's to 04:09 UTC on July 19, and both are right, because that was 9:09 p.m. on July 18 in Redmond, where Microsoft is based.

Correct in public. If a fact turns out to be wrong, post a follow-up that says what was wrong and what's right, and point to it from the original, rather than deleting the first video: the people who saw the mistake only see the fix if it's posted. In ShortFast that means generating a corrected video, because a rendered narration can't be re-voiced.

TikTok's rules are the most specific about news. Even with a label, it doesn't allow AI-generated content that shows 'fake authoritative sources or crisis events', and its Community Guidelines don't allow AI-generated or edited content that misleads about a matter of public importance, naming 'content made to look like it comes from a real news source' and 'a crisis event, like a natural disaster or conflict'. So never borrow a real outlet's name, logo or look, and never present an AI scene as footage of a real event. YouTube's impersonation policy covers channels the same way: copying an entity's 'profile pictures, banners, or specific color schemes' counts, and an AI disclosure 'is not a free pass to impersonate a person, entity, or channel'.

Both platforms also act on real footage shown out of context. TikTok doesn't allow misinformation that poses a risk to public safety or incites panic, 'including falsely presenting past crisis events as recent', and keeps out of the For You feed 'sharing unedited media that is presented out of context', with the example 'showing a crowd at a music concert and falsely calling it a political protest', as well as 'unverified claims about a crisis or major civic event'. YouTube's misinformation policies don't allow content that 'may pose a serious risk of egregious harm by falsely claiming that old footage from a past event is from a current event', or content 'technically manipulated or doctored in a way that misleads users'. YouTube removes such content; a first violation usually brings a warning, and three strikes within 90 days can end a channel.

Elections have their own rules. YouTube doesn't allow content 'aiming to mislead voters about the time, place, means, or eligibility requirements for voting', and 'telling viewers an incorrect voting date' is one of its examples. TikTok doesn't allow misinformation about 'how, when, and where to vote, or register to vote' or about election results, and it doesn't allow paid political advertising, including 'creators being compensated for making branded political content'. Take every election date and rule from the election authority itself.

On monetisation, YouTube's policy on 'AI Personas Related to Sensitive Topics' covers 'any content that presents itself as a human expert providing advice to viewers on topics such as health, legal issues, finances, or politics', and 'channels uploading this content will not be allowed to monetize'. YouTube doesn't define a persona further, so give the narrator no name, title or credentials, keep it to what happened and who said it, and never have it tell viewers how to vote. Under 'Unsatisfying or Off-putting Content', YouTube also excludes 'realistic visuals tricking viewers into believing a fake celebrity death or natural disaster has occurred'. And its advertiser-friendly guidelines treat civil emergencies, natural disasters, conflict and mass acts of violence as sensitive events: news reporting about them may still earn ad revenue, but content that 'profits from or exploits a sensitive event', such as 'using keywords related to a sensitive event to attempt to drive additional traffic', earns none.

For AI labels, YouTube's core cases for news are AI content that 'alters footage of a real event or place' or 'generates a realistic scene that didn't actually occur'; its examples include a realistic tornado heading for a real city that never happened, and making it look like a real person has been arrested. You answer Yes under 'AI use' when you upload; for videos ShortFast posts for you, the setting is in YouTube Studio. YouTube's examples that need no disclosure include non-realistic content, animation, and AI help with the script or captions, and it says disclosing alone won't limit a video's audience or its eligibility to earn money. Stock clips are real footage, not AI. Mixed and the photoreal looks can produce realistic AI scenes, and whatever the look, when a realistic AI scene shows a real place or event, answer Yes.

TikTok requires a label on AI-generated content that contains realistic images, audio or video. Its guidelines say disclosure isn't needed for artistic styles, like anime, or for generic text-to-speech narration that isn't a recognizable voice of a known individual, so the AI narrator alone doesn't call for one. TikTok encourages labeling content that's completely generated or significantly edited by AI and says the setting won't affect distribution, so turning it on costs nothing. Labeled or not, the fake news sources and crisis scenes above stay off-limits.

It has no news feed. ShortFast doesn't fetch, follow or verify the news: it writes a script from your prompt, or narrates yours as written, and neither step knows whether a date, a figure or a name is right. You bring the facts, from the source, on the day you make the video.

Stock footage won't show the event. Stock libraries hold places and everyday scenes, not the press conference, the outage or the flood you're covering. If you have the rights to a photo of the event, upload it into the scene; otherwise show the place and say what happened there.

It adds captions and nothing else on screen: no headlines, lower thirds, tickers or name labels, and no map animation. Names, numbers and dates go in the narration, and the burned-in captions show them word by word.

The narration is fixed once rendered. You can swap scenes, edit captions, change music, title and post text, and re-render, but a fact that changes after the render means generating the video again. For a story that's still developing, make the video once the facts you state are confirmed.

An automation doesn't follow the news. For each posting slot it writes a new topic, script, title, description and footage in the same tone, format and length, and checks recent titles so it doesn't repeat itself, but it doesn't read the day's news, and there's no approval step. For a news series, choose email delivery and check every episode before you post it. The next two episodes are queued: you can edit their captions, scenes and post text, re-time or delete one, but not change its script.

None of these channels use ShortFast; they show what a news explainer looks like when one story gets one video. TLDR News says it 'aims to make news and politics easier to understand': presenters at a desk explain one UK story per video between news clips, under titles such as 'The Holborn & St Pancras Election Explained', and TLDR News Global does the same for world news. RealLifeLore takes one current event per video, usually as a 'why' question, narrated over news and archive footage of the places and people involved. Economics Explained narrates economy stories, such as a country's housing or population policy, over stock footage and photo-collage graphics. Fireship, which describes itself as 'code tutorials and tech news', covers tech stories at speed, with a narrator over screenshots, logos and memes.

What they share is one story or one question per video, explained rather than read out. Their presenters and their footage of the events themselves aren't something stock footage can give you. The structure is the part to copy; the scenes, the voice and the captions are the part ShortFast handles.