Picture and video AI turns a written description into an image or a short clip. You do not need to draw, film or edit to use it. You need to describe clearly what you want to see. This page explains the basics in plain words and links to the site’s detailed pages.
- What it is: an image tool is like a sketch artist working from your description. A video tool is like a film crew: it needs a shot list, not a mood.
- Your first image prompt: say the subject, the setting, the style and the camera view, in plain sentences.
- Your first video prompt: add what moves, how the camera moves, and how long the clip is. Keep it to one main movement.
- Which tool: each one has strengths, and each vendor publishes its own prompt format. Start from the cheat sheets.
- Same face twice: use a reference image. Words alone let a character drift.
- Rights and labels: the tool may let you sell what you make, but in the US you usually cannot own the copyright in a purely generated picture.
How picture and video AI works
The plain point: these tools learned what things look like from a huge number of pictures and clips, and they build a new one that fits your words. They do not search for an existing photo.
Prefer paper? Print the one-page cheat sheet.
The comparison for images: think of a police sketch artist. You describe a face: “narrow chin, short grey hair, glasses.” The artist has seen thousands of faces, so they can draw one that matches. Leave something out, and the artist guesses. The guess is usually the most ordinary choice.
The comparison for video: think of a film crew. A crew cannot work from “make it feel exciting”. It needs a shot list: who is in it, what they do, where the camera is, and how long it lasts. Video AI is the same. A vague request can give a different-looking person in every frame, because the tool has to keep drawing the same scene, frame after frame.
Two words you will meet. A prompt is the text you type. A model is the AI system that makes the picture; the app or website is the shop window around it. AI model types goes deeper.
The AI fills every gap you leave with the most average choice. Good prompts are mostly about leaving fewer gaps.
How to write your first image prompt
The plain point: describe four things, in plain sentences. What is in the picture, where it is, what it should look like, and where the camera is.
- Subject: who or what, doing what. “An older man fixing a bicycle.”
- Setting: place, time and light. “In a narrow garage at dusk, lit by one bare bulb.”
- Style: photo, painting, cartoon, and the mood. “Documentary photograph, muted colours.”
- Camera: how close, and from which angle. “Close-up, from slightly below.”
OpenAI publishes almost exactly this structure for its image model, with one extra part for things to leave out.1
The comparison: it is like ordering a cake from a baker. “A nice cake” gets you whatever they usually make. “A two-layer lemon cake for ten people, white icing, ‘Happy 40th’ in blue” gets you your cake.
Words inside the picture are their own skill. Put the exact words in quotes and say where they go. Then check every letter. Text inside images covers the details, and camera grammar lists the shot types worth naming.
Subject, setting, style, camera. Four short sentences beat one long list of adjectives.
How to write your first video prompt
The plain point: a video prompt is an image prompt plus time. Add what moves, how the camera moves, and how long the clip should be.
- Subject and action: who is in it and the one thing they do. “A woman pours coffee into a white cup.”
- Setting: “A small kitchen, morning light from a window on the left.”
- Style: “Natural colours, like a home video.”
- Camera: one move, named plainly. “Slow push-in” (the camera moves closer) or “static” (it does not move).
- Length: set it in the tool or say it in the prompt, so the action fits the clip.
The comparison: a shot list for a real film crew has one line per shot. If you need three things to happen, that is three shots. AI video works best the same way: one main movement per clip, then join the clips together.
Start small. Runway’s own guide says to begin with only the most important motion and add details one at a time.2 Vendors disagree about word order, so check the cheat sheet for your tool before you settle on one.
The anatomy of an AI video prompt shows a full worked example, and what the vendors actually say puts every vendor’s formula side by side.
Ask for less motion than you want. A held shot with one small movement usually survives. A moving camera, a moving person and a changing background all at once usually falls apart.
Which tool for which job
The plain point: no single tool is best at everything, and the vendors say so about their own models. Pick by the job, then use that tool’s cheat sheet.
| Tool | Worth trying for | Read first |
|---|---|---|
| ChatGPT ImagesOpenAI | Try it for: editing a picture in conversation, step by step. | Sheet: ChatGPT Images cheat sheet |
| Nano BananaGoogle | Try it for: words inside images and working from several reference pictures. | Sheet: Nano Banana cheat sheet |
| MidjourneyMidjourney | Try it for: a strong artistic look. | Sheet: Midjourney cheat sheet |
| FLUXBlack Forest Labs | Try it for: running a model on your own computer. | Sheet: FLUX cheat sheet |
| Seedream and SeedanceByteDance | Try it for: designs with lots of layout (Seedream); video built from many reference files (Seedance). | Sheets: Seedream · Seedance |
| VeoGoogle | Try it for: video with sound you can direct, such as dialogue and effects. | Sheet: Veo cheat sheet |
| Kling, Runway, HiggsfieldKuaishou, Runway, Higgsfield | Try them for: short cinematic clips and camera moves. | Sheets: Kling · Runway · Higgsfield |
| Grok ImaginexAI | Try it for: images and short clips inside the Grok app. | Sheet: Grok Imagine cheat sheet |
The “worth trying for” column follows each vendor’s own description, as collected on which image model for which job. For example, Black Forest Labs publishes a small FLUX model you can download and use commercially.3
For video, Veo vs Gemini Omni vs Grok Imagine Video compares what Google and xAI publish, and Chinese AI video models compared covers Seedance, Wan, MiniMax and Kling. If you make video in the Gemini app, it now comes from Gemini Omni, which Google says “will replace Veo in the Gemini app”; Veo is still offered in Google’s Flow.45 One name you may remember has gone: OpenAI closed the Sora app and website in April 2026.6 What happened to its developer version is on the Sora API page.
What about cost? This page gives no prices. See free vs paid creative tools and what generation actually costs. Every model, with its prompt format, is on the model guides.
Choose the tool for the job, then write in that tool’s format. A prompt that works on one model can underperform on another.
Keeping the same character and look
The plain point: each picture is made fresh, so the AI does not remember the face it drew last time. To keep a character the same, show it a picture, called a reference image.
The comparison: on a real film set, a continuity person keeps photos of every actor’s hair, clothes and props, so the next day’s shots match. A reference image is that photo, handed to the AI.
Meta describes the fix many tools use: anchor a whole series on a few reference images.7
- Use a clean reference. A clear, well-lit, front-facing photo works better than a blurry one.
- Do not contradict it. If the picture shows dark hair and your prompt says blonde, the words usually win.
- Change one or two things per step. Ask for a new background, keep everything else.
Character consistency covers every method, and how to use reference images explains why a tool sometimes ignores yours.
When it comes out wrong or looks fake
The plain point: a bad result usually means a gap or a clash in your prompt. Change one thing at a time, so you learn which change fixed it.
Why AI pictures look “too perfect”: image models were tuned toward pictures people like, not pictures that look real. One widely used training dataset was built around a single rating question.8 Pretty is not the same as real. So ask for real-world details by name: skin texture, one light source, a little mess. Why AI output looks fake gives the words, and the realism words explains why “hyperrealistic” often backfires.
The comparison: it is like a mechanic. Swap the battery, the tyres and the spark plugs at once, and the car may start, but you will not know what was broken. Why your generation came out wrong matches six common failures to the line of the prompt that causes each.
Negative prompts are lists of things you do not want, such as “blurry, extra fingers”. Some tools have a box for them and many do not. Midjourney has one; Black Forest Labs says FLUX.2 has none.9 Where there is no box, describe what you do want instead: “visible pores” rather than “no plastic skin”. Give the taxi driver the address, not a list of streets to avoid. Negative prompts by problem and negative prompts for every model say which tools read them.
Read the failure before you rewrite. If three rewrites change nothing, try a reference image, a different tool, or a quick fix in an editor.
Who owns it, and how AI pictures get labelled
The plain point: a tool’s terms may let you sell what you make. That is not the same as owning the copyright, which is the right to stop others copying it.
The comparison: a licence is a permission slip; copyright is a fence. Most AI pictures come with the slip and not the fence. In the US, the Copyright Office protects only work that comes from a person’s creativity, and a prompt on its own does not count.10 Who owns what you generate explains where the line falls. Copying a real person’s face or voice raises different laws, covered on cloning a voice or a face.
Labels. Some tools hide a mark inside the picture itself. Google’s is called SynthID, and Google says it is built to survive common edits.11 But a missing label proves nothing: many tools add none, and a screenshot can strip some kinds. How AI content gets labelled explains both kinds of mark, and how to spot AI-generated images, video and voices covers checking someone else’s.
You can usually use and sell what you make on a paid plan. You usually cannot stop others making something similar. If you publish AI pictures, say so yourself: that label cannot be stripped.
Where to go next
Pick the line that sounds like you.
- “I want to see everything on this topic.” Open the Filmmaking hub.
- “I know my tool and want its rules.” Go to the cheat sheets.
- “I want to make a short video.” Read the anatomy of a video prompt, then camera grammar.
- “My pictures look fake.” Start with why AI output looks fake.
- “What shape should it be for social media?” See aspect ratio, resolution and format.
- “I want sound or music too.” Try prompting music and voice.
- “I am new to AI in general.” Begin with AI explained for complete beginners.
Questions people ask
How do I write a good AI image prompt?
Describe four things in plain sentences: the subject, the setting, the style and the camera view. OpenAI publishes almost exactly this structure for its image model, with one extra part for things to leave out.1
How do I keep the same character in AI images?
Use a reference image. Each picture is made fresh, so the AI does not remember the face it drew last time. Meta describes the fix many tools use: anchor a whole series on a few reference images.7
Do I own the copyright to AI-generated images?
Usually not in the US, even when the tool’s terms let you sell what you make. The US Copyright Office protects only work that comes from a person’s creativity, and a prompt on its own does not count.10
Why do AI pictures look fake or too perfect?
Image models were tuned toward pictures people like, not pictures that look real; one widely used training dataset was built around a single rating question.8 Ask for real-world details by name, such as skin texture, one light source and a little mess.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. Which four things does the page say to describe in a first image prompt?
- The price, the size, the file type and the deadline
- The subject, the setting, the style and the camera view
- A long list of adjectives such as stunning and detailed
Show the answer
Answer: The subject, the setting, the style and the camera view. The page says four short sentences on subject, setting, style and camera beat one long list of adjectives.
2. What is the page’s advice for a first video prompt?
- Ask for several big movements at once to get your money’s worth
- Leave out the camera, because the tool always picks well
- Keep it to one main movement and add time, camera move and length
Show the answer
Answer: Keep it to one main movement and add time, camera move and length. The page says a video prompt is an image prompt plus time, and a clip works best with one main movement.
3. How do you keep the same character looking the same across pictures?
- Show the tool a picture of the character, called a reference image
- Describe the face in more and more detail each time
- Ask the tool to remember the face from last time
Show the answer
Answer: Show the tool a picture of the character, called a reference image. The page explains each picture is made fresh, so a reference image works like a continuity photo on a film set.
Related on this site: AI Video Platforms Like Google Flow and Higgsfield, Sorted by Job · Image Edit Prompts That Change Only One Thing · Old Photo Restoration Prompts · Art Styles for AI Prompts · Make Your First AI Video · How to Write Your First AI Video Prompt
- 1OpenAI, Image prompting guide, read at source 16 Sep 2026“For complex requests, organize the prompt as scene, subject, details, and constraints, using labeled sections.” ↑a“For complex requests, organize the prompt as scene, subject, details, and constraints, using labeled sections.” ↑b
- 2Runway, Gen-4 Video Prompting Guide, read in a browser 17 Sep 2026“Begin with a foundational prompt that captures only the most essential motion to the scene.” ↑
- 3Black Forest Labs, FLUX.2 [klein] 4B model card, read at source 18 Sep 2026“Open weights available for commercial use under the Apache 2.0 license.” ↑
- 4Google, Gemini Omni (Gemini app overview), undated, read at source 13 Oct 2026“Gemini Omni is our latest video editing and generation model that will replace Veo in the Gemini app.” ↑
- 5Google, Learn about Google Flow models & supported features, Google Flow Help, read at source 13 Oct 2026sections “Supported features in Veo 3.1 - Lite”, “Veo 3.1 - Fast”, “Veo 3.1 - Quality” and “Supported features in Gemini Omni Flash 1.1”. ↑
- 6OpenAI Help Center, What to know about the Sora discontinuation, read in a browser 1 Oct 2026“The Sora web and app experiences were discontinued on April 26, 2026.” ↑
- 7Meta, Build with Muse Image, 25 Aug 2026, read in a browser 18 Sep 2026“Anchor a whole series of generations on a small set of reference images so the character, style, and setting stay consistent from one image to the next.” ↑a“Anchor a whole series of generations on a small set of reference images so the character, style, and setting stay consistent from one image to the next.” ↑b
- 8LAION, LAION-Aesthetics, re-read at source 11 Sep 2026the rating question was “How much do you like this image on a scale from 1 to 10?” ↑athe rating question was “How much do you like this image on a scale from 1 to 10?” ↑b
- 9Black Forest Labs, FLUX.2 prompting guide, read at source 18 Sep 2026“FLUX.2 does not support negative prompts.” ↑
- 10US Copyright Office, Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, Federal Register, effective 16 Mar 2023, read at source 23 Sep 2026copyright can protect “only material that is the product of human creativity”. ↑acopyright can protect “only material that is the product of human creativity”. ↑b
- 11Google DeepMind, SynthID, read at source 17 Sep 2026“It’s added the moment content is created, and designed to stand up to modifications like cropping, adding filters, changing frame rates, or lossy compression.” ↑
OpenAI, Image prompting guide (16 Sep 2026) · Runway, Gen-4 Video Prompting Guide (17 Sep 2026) · Black Forest Labs, FLUX.2 [klein] 4B model card and FLUX.2 prompting guide (18 Sep 2026) · Meta, Build with Muse Image (18 Sep 2026) · OpenAI Help Center, Sora discontinuation (1 Oct 2026) · LAION, LAION-Aesthetics (11 Sep 2026) · US Copyright Office, registration guidance (23 Sep 2026) · Google DeepMind, SynthID (17 Sep 2026). Dates are when this site read each source, as cited on the linked detail pages.
The comparisons on this page (sketch artist, film crew, cake order, continuity photos, mechanic, taxi driver, permission slip and fence) are this site’s own ways of explaining, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK