Learn AI /Generative Media /Video generation
Video generation
- Starter
- Generative Media
- Communications teams
Also called: text-to-video, AI video.
Back to the milestone you came from
In five seconds
What it is
A system that creates moving pictures from prompts, images or other guidance, producing short video clips that did not previously exist as recorded footage.
In plain words
It makes a new video from instructions, instead of only editing footage that already exists.
An everyday example
A communications officer asks an illustrator to make a 5-second flipbook showing a cyclist crossing a market square at sunrise. The illustrator has seen many films, photos and drawings before, so they sketch one page after another to keep the bicycle, the square and the light fairly consistent. If the officer adds, "make the camera move closer," the illustrator redraws the pages to create that effect. At the end, the flipbook shows motion even though no one filmed a real cyclist in that square.
This analogy uses a human illustrator who deliberately plans each page. A video generation model does not work like a person drawing with full understanding of the scene. It statistically predicts visual patterns from training data and may produce inconsistencies, impossible motion or invented details. Real systems also use mathematical representations, not literal paper pages.
Remember thisVideo generation creates new moving images from prompts or other guidance. Even a convincing result is synthetic media, not verified footage of an event.
Understand it
Understand it
Why it matters
Non-specialists need to recognise two things at once: video generation can save time in creative work, and it can also produce convincing false visuals. That matters when you approve communications, assess evidence, brief colleagues, moderate online content or decide whether a clip is safe to share. A generated video may be useful as an illustration, simulation or draft, but it should not automatically be treated as proof of an event. Good decisions depend on asking where the clip came from, how it was made, whether it is labelled clearly, and whether other sources confirm it.
How video generation turns guidance into a clip
After viewing this, a reader should be able to say what can go into a video generator, what happens inside at a high level, and what comes out.
01
1. Give guidance
The user provides a text description, image, storyboard or reference video to steer the result.
Draw a box with three small icons labelled text, image and clip feeding into the pipeline.
02
2. Generate frames
The model creates candidate frames, often starting from random visual noise and shaping it toward the prompt.
Draw a central box with faint noisy thumbnails becoming clearer from left to right.
03
3. Keep time consistent
The system tries to make objects, lighting and motion line up across neighbouring frames so the clip looks continuous.
Show three adjacent frames connected by arrows, with matching coloured outlines around the same object.
04
4. Output and verify
The result is a new video clip that may look convincing, so people still need to check source, context and any provenance information.
Draw a play-button video thumbnail and a small adjacent badge labelled check source/context.
A left-to-right flow diagram with four boxes. Box 1 shows possible inputs: text prompt, image, reference clip. Box 2 shows the model building frames from noise or drafts. Box 3 shows frame-to-frame consistency checks for subject, motion and style. Box 4 shows the output clip plus a warning label that realism does not prove authenticity. Read from left to right.
Hovering or tapping each box reveals a short note with one example input or output; keyboard focus moves through the four steps in order.
Describe this without the diagram
This diagram explains video generation as a four-step flow. First, a person gives guidance such as a written prompt, an image or a reference clip. Second, the model generates frames, often beginning with random noise and gradually shaping them to match the guidance. Third, it tries to keep the scene consistent over time so objects, lighting and movement do not change abruptly from frame to frame. Fourth, it produces a video clip, but the viewer is reminded that realistic appearance does not prove the clip is authentic or documentary evidence.
With reduced motion: A static left-to-right diagram with four labelled boxes and arrows: guidance in, frames generated, consistency across time, video out with a source-check reminder.
Video generation creates new video clips with software models trained on many examples of videos, images and text. The input can be a written prompt, a still image, a rough storyboard, or an existing clip to transform. The output is a sequence of frames, often with motion, style and camera movement that match the guidance.
A prompt might say, "A red bus drives through light rain at dusk, viewed from street level." The system then predicts visual patterns across space and time so the clip looks coherent. In many systems, a model called a diffusion model starts from visual noise and gradually turns it into frames that fit the prompt.
This differs from ordinary video editing, which mainly arranges or modifies existing footage. Video generation can make scenes, people, objects or movements that were never filmed, although it can also transform an existing clip. It differs from image generation because it must keep things reasonably consistent across time rather than make one still picture.
The results can be useful for storyboarding, mock-ups, training materials and visual ideas. They can also be misleading. A realistic clip does not prove that an event happened, so generated video raises authenticity questions and may need labels, provenance information and human checking before use.
How “A flipbook made from a written brief” maps onto the real thing
| In the story | In the AI system |
|---|---|
| The written brief for the flipbook | The text prompt or other guidance given to the model |
| Many pages drawn in sequence | Many video frames generated in order |
| Keeping the bicycle and square consistent across pages | Maintaining visual continuity over time |
| Changing the pages to create a closer view | Controlling camera movement or style from the prompt |
A worked example
An exhibition designer needs an eight-second concept clip for a trailer storyboard before the next morning's production review.
Starting from: Prompt: "8-second shot of a blue ceramic vase rotating slowly on a wooden table, warm window light, plain background, camera moves slightly closer."
- The designer enters the prompt and selects an eight-second output.
- The model generates candidate frames conditioned on the description, attempting to preserve the vase, lighting and camera movement across the sequence.
- During review, the designer notices that the handle changes shape and the table edge flickers between frames.
- The prompt is refined and another version is generated. The designer selects the most coherent clip and records the remaining visual defects for the production review.
Result: The output is a newly generated planning clip, not footage of a vase that was actually filmed. It can guide the storyboard, but its objects and motion still require human review.
In a United Nations workplace
At a shared services centre, a chief of section is clearing the briefing note due at 6pm on a proposed staff onboarding module. The vendor has supplied a generated clip showing a generic service request moving through an approval workflow, with no real personnel or case data depicted.
Video generation can turn a script, reference image or storyboard into a short concept clip before production resources are committed. The draft can help the chief assess pacing and visual logic, while checking whether objects, text and motion remain coherent from frame to frame.
Before you try thisNo personal, personnel, procurement-sensitive or operational information may be entered into a tool that has not been approved for it. Generated clips must be reviewed for fabricated details, unstable text and misleading imagery, and must not be presented as recorded footage of an actual office, person or event.
Where you may hear it
- A vendor demo during a procurement meeting
- Clearance of a public information campaign
- A briefing on manipulated media and verification
- A production meeting for a side-event explainer
Avoid confusion
Avoid confusion
Often confused with
Synthetic media
Synthetic media is text, images, audio or video generated or substantially altered by software, often AI, so it can resemble authentic material.
Deepfake
A deepfake is an AI-generated or heavily altered image, audio clip or video that convincingly imitates an identifiable real person, usually in a recording or event that did not happen as shown.
Image generation
Software can create new pictures from prompts or other inputs by learning visual patterns from many examples, then producing a fresh image rather than retrieving one file.
Common misconceptions
Common misconception: If a generated clip looks realistic, it must be based on real footage.
Correction: Realistic appearance only shows that the model produced believable visual patterns. It can synthesise scenes that were never recorded, combine invented details, or imitate camera behaviour without any real event behind it.
Common misconception: Video generation is just image generation repeated many times.
Correction: Single images do not need to stay consistent over time. Video generation must also handle motion, continuity and timing across frames, which is why flicker, drifting details and unnatural movement are common failure points.
Common misconception: All AI-generated video is a deepfake.
Correction: Deepfake usually refers to synthetic media that imitates a real person's face, voice or identity, often deceptively. Video generation is broader: it can create landscapes, objects, diagrams, animations or fictional scenes without impersonating anyone.
Quick check
Which statement best describes video generation?
Video generation produces new moving images from prompts or other controls. Some tools also transform existing clips, but the key idea is synthesis of video, not simple transcription or proof of reality.
Go deeper
Go deeper
The simple version
It makes a new video from instructions, instead of only editing footage that already exists.
The technical version
Video generation uses generative models to predict a sequence of video frames conditioned on inputs such as text, images, masks, depth maps or reference clips. Many current systems adapt diffusion-based methods, which iteratively denoise a latent representation toward frames that match the conditioning signal. The model must handle both spatial quality within each frame and temporal coherence across frames, so the same object, lighting and motion remain reasonably stable over time. Some systems generate short clips directly; others generate key frames and interpolate motion between them. Outputs can also be guided by camera controls, style references or video-to-video transformation. The exact architectures vary, and quality depends on training data, compute, duration and control mechanisms.
Reviewed external resources
Primary sourceRecommendation on the Ethics of Artificial Intelligence (opens in a new tab)UNESCO
This UNESCO recommendation gives an international ethics framework for AI use and suits policy, governance and communications readers.
Intermediatelink checked 2026-07-31
Primary sourceContent Credentials (opens in a new tab)Content Authenticity Initiative
This Content Authenticity Initiative resource explains content credentials and provenance ideas, and suits readers concerned with verifying media origin.
Starterlink checked 2026-07-31
EncyclopaediaText-to-video model (opens in a new tab)Wikipedia
This Wikipedia page gives a broad overview of text-to-video models and suits readers who want a quick background before going deeper.
Intermediatelink checked 2026-07-31
VideoVideo explanations: Video generation (opens in a new tab)YouTube search
This YouTube search results page helps readers find different video explanations of video generation and suits people who learn best by watching, but it is a search page rather than one vetted lesson.
search results, not one pagelink checked 2026-07-31
DiscussionCommunity discussion: Video generation (opens in a new tab)Reddit search
This Reddit search results page shows community discussions and questions about video generation and suits readers who want to see common public concerns, but it is a search page rather than an authoritative source.
search results, not one pagelink checked 2026-07-31
Where this came from
Drafted with a language model through an offline pipeline, checked against this site's schema, and every external link above fetched to see whether it answers. Treat it as a starting point and check anything that matters against the primary sources. How this site is built and what that means.
The one thing to remember
Video generation creates new moving images from prompts or other guidance. Even a convincing result is synthetic media, not verified footage of an event.