LIVE
Study finds gender bias in GPT models is not reduced but reshaped across generations17/09/26 · OpenAI|Architectural tweaks may break conventional scaling law exponents16/09/26|OpenAI publishes a framework for reporting model misalignment16/09/26 · OpenAI|NVIDIA's Vera Rubin NVL72 Debuts in MLPerf Inference v6.116/09/26 · NVIDIA|OpenAI expands ChatGPT advertising with Sponsored Agents16/09/26 · OpenAI|OpenAI moves into advertising with 'Sponsored Agents'16/09/26 · OpenAI|Google DeepMind Introduces Gemini 3.8 Live and Its Extended Thinking Variant15/09/26 · Google DeepMind|What's at stake in AI's trillion-dollar infrastructure bet15/09/26|A Flaw in Chain-of-Thought Safety Monitoring14/09/26|Stellar Colosseum: A Multi-Agent System for Long-Horizon Mathematical Research14/09/26|Apple Code Hints Siri Could Be Swapped for ChatGPT or Claude14/09/26 · Apple|Anthropic says Houthi-linked actors used Claude Code for missile guidance software13/09/26 · Anthropic|Study finds gender bias in GPT models is not reduced but reshaped across generations17/09/26 · OpenAI|Architectural tweaks may break conventional scaling law exponents16/09/26|OpenAI publishes a framework for reporting model misalignment16/09/26 · OpenAI|NVIDIA's Vera Rubin NVL72 Debuts in MLPerf Inference v6.116/09/26 · NVIDIA|OpenAI expands ChatGPT advertising with Sponsored Agents16/09/26 · OpenAI|OpenAI moves into advertising with 'Sponsored Agents'16/09/26 · OpenAI|Google DeepMind Introduces Gemini 3.8 Live and Its Extended Thinking Variant15/09/26 · Google DeepMind|What's at stake in AI's trillion-dollar infrastructure bet15/09/26|A Flaw in Chain-of-Thought Safety Monitoring14/09/26|Stellar Colosseum: A Multi-Agent System for Long-Horizon Mathematical Research14/09/26|Apple Code Hints Siri Could Be Swapped for ChatGPT or Claude14/09/26 · Apple|Anthropic says Houthi-linked actors used Claude Code for missile guidance software13/09/26 · Anthropic|
Beginner🎬

Veo 3, Sora, Kling: creating videos with AI in 2026

Veo 3 (Google), Sora (OpenAI), Kling (China): 3 AIs that generate pro videos from text. We explain how they work, which to choose, costs and killer use cases.

13 min readPublished May 5, 2026· Updated September 17, 2026

In one sentence

Veo 3 (Google), Sora (OpenAI) and Kling (Chinese) are AIs that generate realistic videos from simple text. In 2026, we've gone from funny gifs to professional 1-2 minute videos almost indistinguishable from real footage. And it changes everything for content creation.

🎬
The analogy that works
Imagine hiring a director, a cameraman, an actor, a lighting technician, an editor : the entire crew for an advert shoot. Except you simply ask them: "Film me a golden retriever surfing on a wave in Hawaii at sunset". And 30 seconds later, they deliver the video. Without you, without a set, without travel. That's exactly what Veo 3 or Sora does.

🎭 Want to understand the risks of AI video generation?

The other side of video generation: deepfakes and how to protect yourself.

Read the deepfakes article

Where does the revolution come from?

AI video generation has existed since 2022 (first Meta Make-A-Video prototypes). But it was blurry 2-second gifs. The leap happened in 2024-2025 with:

The explosion of AI video

  1. Meta Make-A-Video

    First AI gifs, 4 seconds, 256×256 pixels. Very blurry. Technical curiosity.

  2. Sora (OpenAI)

    Demo that shocks the world: 1 minute, 1080p quality. OpenAI shows the potential.

  3. Kling (Kuaishou China)

    First massive consumer video model. Free, fast, popular in Asia.

  4. Veo 2 then Veo 3 (Google)

    Google catches up with OpenAI. YouTube integration, cinema quality, integrated sound.

  5. Democratization

    Tools accessible to all (£10-30/month), 1-2 min duration, quality indistinguishable from real video in 80% of cases.

How it works (in simple terms)

No need to be an engineer:

The principle in 4 steps
1. You write a prompt (text description) E.g.: "A red-haired woman in a red dress walks down a rainy Paris street, close-up, cinematic atmosphere." 2. The AI breaks down your request Characters, location, atmosphere, camera movement, duration, style. 3. The model generates the video frame by frame Like a painter drawing 24 images per second, maintaining coherence between each frame. 4. You get an MP4 file Downloadable, editable, usable wherever you want.

Comparison of the 3 stars of 2026

Veo 3 vs Sora vs Kling

 🎬Criterion🏆The best
Image qualityVeo 3 ★★★★★ / Sora ★★★★★ / Kling ★★★★Veo 3 (slightly)
Max durationVeo 3: 60s / Sora: 60s / Kling: 30sVeo 3 and Sora
Integrated soundVeo 3: yes (effects, voices) / Sora: no / Kling: noVeo 3 (game changer)
Movement (action, sport)Veo 3 ★★★★ / Sora ★★★★★ / Kling ★★★★Sora
Monthly priceVeo 3: £30 / Sora: £200 (ChatGPT Pro) / Kling: £10Kling (value for money)
Free versionVeo 3: limited / Sora: none / Kling: yesKling
Character coherenceVeo 3 ★★★★ / Sora ★★★★ / Kling ★★★Veo 3 and Sora
Generation speedVeo 3: 2-3 min / Sora: 5-10 min / Kling: 1-2 minKling

Use cases that are booming in 2026

📚6 ways to use AI video right now

1. Social media (Instagram, TikTok)

Idea: create 5-10 reels per week without filming. Examples: animated recipes, stylish transitions, fake "lifestyle shots", animated YouTube thumbnails. Cost: £10-30/month vs £500-2,000 for a real shoot.

2. Adverts and content marketing

Idea: prototype several advert versions before the real shoot. Examples: test 3 atmospheres for a product, rapid visual A/B testing. Practice: the Toys"R"Us brand released a 100% Sora advert in 2024. Coca-Cola, Mango likewise.

3. E-learning and tutorials

Idea: illustrate abstract concepts with generated videos. Examples: "how water circulates in a plant", "the solar system in motion". Advantage: visuals that can't be filmed in reality.

4. Presentations and pitches

Idea: replace static PowerPoint images with generated mini-videos. Examples: product presentation, company story, pitch deck intro video. Effect: it impresses investors/clients ("company of the future" vibe).

5. Storytelling and fiction

Idea: create music videos, short films, animations with no budget. Examples: independent artists releasing their video for £50 instead of £50K. Practice: "AI Film Awards" Festival since 2024.

6. Drafts and previz for cinema/TV

Idea: test a shot, an atmosphere, a camera movement before filming. Examples: "previz" for big budgets, animated mood boards. Pro: used by Pixar, Netflix to pre-visualize complex scenes.

How much does it really cost?

Monthly cost by usage

Intensive pro (multi-accounts)500£/month
Sora via ChatGPT Pro (unlimited)200£/month
Veo 3 via Google AI Pro30£/month
Sora via ChatGPT Plus (60 videos)22£/month
Kling Premium (60 videos/month)10£/month
Kling (free with limits)0£/month

Reading: with £20-30/month, you can produce the equivalent of £5,000-10,000 worth of traditional filming per month. ROI x100-x500.

How to write a good video prompt

The prompt makes 80% of the final quality. Effective method:

✍️
The magic formula for video prompts
[SUBJECT] + [ACTION] + [LOCATION] + [ATMOSPHERE] + [SHOT] + [STYLE] Basic example: > "A cat running." Optimized example: > "A white Persian cat runs through a field of sunflowers at sunset, seen from the front, close-up shot following the movement, beautiful golden light, cinematographic style, 4K." Difference: the optimized version will give a video 5x better. Tips: - Specify the shot (close-up, wide, low angle) - Specify the atmosphere (cinema, documentary, animation) - Specify the lighting (golden, dramatic, natural) - Avoid too many simultaneous movements (AIs handle them poorly) - For dialogue, use Veo 3 (the others don't do audio)

The limitations in 2026 (and there are some)

What AI videos can't (yet) do
1. Long-term coherence Beyond 30-60 seconds, the character can "mutate" (change hair colour, face). For a 5-min video, you need to generate several clips and stick them together. 2. Hands and fingers Hands remain a challenge. Often 6 fingers, bizarre gestures. Avoid in close-up. 3. Readable text AIs struggle to write coherent text in videos (signs, books). Add text in post-production. 4. Realistic physics Generated videos can have bizarre physical behaviour (water flowing backwards, objects defying gravity). 5. Synchronized voice (except Veo 3) Sora and Kling don't have sound. You need to generate voice separately (ElevenLabs) and synchronize. 6. Famous locations Censored models don't generate certain trademarked locations (logos, brands). Eiffel Tower yes, "inside Apple Store" no.

Important legal questions

What you need to know before publishing
🎭 Real people You CANNOT generate a video of a real person (politician, celebrity, your neighbour) without their written consent. It's image rights. → Penalties: up to 1 year in prison + £45,000 fine in France. 🏛️ Brands and logos You CANNOT include logos, brands, copyrighted characters (Mickey, Pikachu) without authorization. → Intellectual property infringement. 🏷️ Mandatory labelling Since the AI Act (2026), any AI-generated video must be clearly labelled "generated by AI" if it could mislead. → Penalties: up to 35 million euros for companies. 📄 Copyright on your creations A 100% AI video = not protected by copyright (US law, debate in EU). To have rights, you need a significant human contribution (editing, assembly, original idea). 👥 Minors TOTAL PROHIBITION on generating videos of real or realistic minors. Very heavy criminal framework (CSAM).

Veo 3 vs Sora vs Kling: which to choose?

Which tool for which need

 🎯You want...✅Choose
Test for freeNo budget, just discoverKling
Pro advert with dialogueFinalized video with synchronized voiceVeo 3 (audio included)
Action / sport / movementFighting, running, dancingSora
Long format (>30s)Short film, complete sceneVeo 3 or Sora
High volume / daily10+ videos per dayKling Premium
YouTube integrationPublish directlyVeo 3 (integrated)
Research / experimentationTest limits, prototypesKling (free, fast)

The metaphor that sums it all up

📸
The iPhone moment for video
Before 2007, taking a good photo required an expensive camera + technical know-how. The iPhone democratized it: everyone has a pro photographer in their pocket. In 2026, AI videos are the iPhone moment for video. Before, making a pro video required: camera (£5K), crew (3-10 people), editing (30+h). Now: a prompt + £30/month + 5 minutes. As with the iPhone, it doesn't eliminate pros (pros remain better at the high end). But it allows anyone to produce decent quality video content. Massive democratization = new creative opportunities + new risks (deepfakes, disinformation).

Absolutely remember

  • ✅ Veo 3 / Sora / Kling are the 3 leaders in AI video in 2026
  • ✅ Cost: £10-30/month for regular use
  • ✅ Massive use cases: social media, adverts, e-learning, storytelling
  • ✅ Limitations: duration < 60s, sometimes odd hands, narrative coherence
  • ✅ Legal: don't generate real people, label "AI", prohibition on minors
  • ✅ Veo 3 = leader if you want integrated sound; Kling = best value for money

It's a new profession being created: video prompt engineers. If you're creative, it's a huge opportunity in 2026.

🧠 Quiz
Question 1 of 3

Which model has the MAJOR advantage of also generating sound and synchronized voices?

To go further

Tags
Vidéo IAVeoSoraKlingCréation
All articles

Read next

Beginner🎭

Deepfakes: how to protect yourself in 2026

12 min read
Beginner🛠️

10 truly essential free AI tools

10 min read
Beginner💰

Pay or not for an AI?

10 min read