AI video generation: Sora, Veo, and what to know
AI can now generate video from a text description. Type "a golden retriever running through a wheat field at sunset" and 30 seconds later you get a video clip. The technology is impressive in some shots, obviously fake in others. Here's what's actually possible in 2026, which tools to use, and what to expect.
The short version
- Best quality: Sora (OpenAI), Veo (Google), Kling (Chinese).
- Best for free: Veo through Google's Gemini, Kling free tier.
- Best for editing: Runway Gen-3 (more controls).
- Realistic length: 5-20 seconds. Anything longer is stitching clips together.
- Not good for: complex action, consistent characters across scenes, anything requiring real story.
The main AI video tools
OpenAI Sora
Available to ChatGPT Plus ($20/month) for limited generations and Pro ($200/month) for more. Best general-purpose AI video. Produces 5-20 second clips at up to 1080p.
- Text-to-video: type a description.
- Image-to-video: upload a still photo, AI animates it.
- Video-to-video: upload a clip, AI restyles it.
- Storyboards: chain multiple clips together for longer videos.
Google Veo
Google's competitor. Available through Gemini Advanced and via the Veo dedicated app. Free tier has limits; paid (Google AI Pro) increases generations and length.
- Generally produces more photo-realistic results than Sora.
- Better at landscapes and nature.
- Tied into the Google ecosystem.
Runway (Gen-3)
The most pro-focused option. runwayml.com. Free trial with watermarks; paid plans from $12/month.
- More editing controls (camera angle, motion intensity).
- Lip-sync features.
- Favored by filmmakers and YouTubers.
Kling AI
Chinese-developed, surprisingly good free tier. klingai.com.
- Generous free generations daily.
- Good quality for the price.
- Some users report privacy concerns about Chinese-hosted services.
Luma Dream Machine
Free generations daily. lumalabs.ai/dream-machine. Good for quick experiments.
Pika Labs
Older, has been overtaken by newer competitors but still produces decent results. Free tier available.
What AI video can actually do
Short B-roll for videos
You're making a YouTube video about coffee. You need 5 seconds of "beans pouring out of a bag." Generate it. Cheap, fast, no licensing issues.
Animated greeting cards
A still photo of your dog turns into a 5-second clip of your dog looking up at the camera and wagging its tail. Send to family.
Marketing clips
Small business needs a 15-second social media video. Generate it without hiring a videographer or finding stock footage.
Storyboard previews
Filmmakers use AI video to visualize scenes before filming them. Cheaper than animatics.
Illustrations of concepts
A presenter explaining how nuclear fission works can generate an animated illustration of the process.
Personal video projects
Animate your old family photos. Create a video version of your kids' artwork. Reimagine a favorite vacation memory.
What AI video can NOT do (yet)
- Long, coherent scenes. 20 seconds is the limit for most tools. Beyond that, characters and settings drift.
- Consistent characters across clips. The same person looks slightly different each generation.
- Complex action. Sports, fights, dancing all break down.
- Realistic hands and fingers. Still problematic.
- Text in videos. Signs and labels often look like gibberish.
- Real people accurately. Most tools refuse to generate named celebrities or specific real people.
- Replace movies or TV shows. Hollywood is using AI for specific tasks; full AI-generated films are years away.
Step by step: make your first AI video
The fastest free option (Google Veo via Gemini)
- Go to gemini.google.com.
- Sign in with Google.
- Click the photo/image icon next to the chat box.
- Type: "Generate a video of a hummingbird hovering near a red flower in slow motion, 5 seconds, photorealistic."
- Wait 30 seconds to 2 minutes.
- Download the result.
Sora (ChatGPT Plus or Pro)
- Go to sora.com while signed into ChatGPT Plus.
- Click Create video.
- Type your prompt.
- Choose duration (5-20 seconds) and resolution.
- Generate. Wait 1-3 minutes.
How to write a good AI video prompt
Same recipe as AI images, with motion added:
[subject] + [action / motion] + [setting] + [style] + [camera details]
Example: building a good video prompt
- Bad: "A dog."
- Better: "A golden retriever puppy running."
- Better still: "A golden retriever puppy running through tall grass in a meadow."
- Best: "A golden retriever puppy running through tall summer grass in a meadow at sunset. Slow motion. Camera tracking alongside. Warm golden light. Cinematic, like a Pixar film."
Useful style and camera terms
- Camera motion: static shot, panning, tracking, zoom in, zoom out, dolly, drone overhead
- Speed: slow motion, normal speed, time-lapse
- Lighting: golden hour, blue hour, harsh sunlight, soft window light, candlelight
- Mood: cozy, dramatic, peaceful, mysterious
- Style references: "in the style of a Wes Anderson film," "like a 1970s home video," "Pixar style," "documentary footage"
Where AI video shines for small business
- Instagram Reels and TikTok: short clips for product launches, behind the scenes, customer education.
- Google Business Profile videos: short clips to add to your business listing.
- Website hero animations: a 10-second loop on your homepage.
- Holiday greetings to customers: animated message that looks more interesting than a still card.
- Internal training videos: visualize a process for new hires.
Cost comparison (2026)
| Tool | Free tier | Paid |
|---|---|---|
| Sora | No | $20/mo (limited) or $200/mo (Pro) |
| Veo via Gemini | Yes, limited | $20/mo Google AI Plus |
| Runway Gen-3 | Watermarked trial | $12-$76/mo |
| Kling | Yes, generous | $10/mo and up |
| Luma Dream Machine | Yes, daily limit | $10/mo and up |
| Pika Labs | Yes, limited | $10/mo and up |
Editing AI videos
The AI generates raw clips; you usually need basic editing tools to make them useful:
- CapCut: free, easy. Web and mobile. Perfect for combining AI clips, adding music, captions.
- iMovie (Mac/iPhone): built in. Free.
- DaVinci Resolve: pro-level, free.
- Adobe Premiere Rush: simpler than full Premiere, paid.
Ethics and commercial use
- Sora, Veo, Runway, Pika all grant commercial use rights for content you generate.
- Don't generate videos of real people (especially celebrities or public figures) without permission.
- Don't generate copyrighted characters (Mickey Mouse, Marvel characters) for commercial use.
- Be careful with deepfakes; many jurisdictions are passing laws against unauthorized deepfakes of real people.
- Disclose AI-generated content where appropriate (paid ads in many countries now require AI disclosure).
Common AI video tells (to spot or avoid)
- Hands and fingers wobble or merge
- Background morphs as the camera moves
- Reflections don't match the scene
- Hair fades unnaturally at edges
- Mouth and lip movement looks slightly off
- Crowds blur into unrecognizable masses
- Text on signs is gibberish
- Physics is slightly wrong (water, fabric, smoke)
Use these as a checklist. If your AI video has multiple tells, regenerate or try a different prompt.
Best uses for personal projects
- Animate an old photo of a deceased loved one (treat with respect; family discussion first).
- Make a 5-second animated logo for a hobby YouTube channel.
- Generate a slideshow video for a birthday party.
- Create a "what if" video reimagining a memory.
- Make an art piece from a written description.
5 things to try this week
- Try Veo through Gemini for free. Generate one 5-second clip.
- If you have ChatGPT Plus, try Sora.
- Combine 3 AI clips in CapCut to make a 15-second story.
- Try image-to-video: upload a photo, ask AI to animate it.
- Compare the same prompt across two tools and see which you prefer.
Video walkthrough
Video by Kevin Stratvert on YouTube
Want help with AI video for your business?
If you have a small business and want to use AI video for marketing or social media, Isaac can sit with you and walk through tools and workflow.