AI video for social media and ads: a practical guide for small businesses in Andorra (2026)

A reel every week, three versions of the sale ad, a shot of the new product on a terrace with mountain views. Producing all of that with a camera takes time and money, and that is where generative AI is now genuinely useful. But the market changes every few months: OpenAI shut down Sora in April 2026 and Google has just launched Nano Banana 2.1. In this guide we explain which tools to use as of October 2026, which formats work, what they cost, how to label them and when it is still better to get the camera out.

Which AI tools are people using to create images and video today?

As of October 2026 it helps to think in four families rather than an endless list of apps. First, the general-purpose assistants: ChatGPT creates images with Images 2.5, launched on 8 September 2026 for all plans, and Google's Gemini includes Nano Banana for images and Veo 3.1 for video, which produces 8-second clips with native sound that you can then extend.

Second, the image specialists, such as Midjourney, with a very polished look and video generation too. Third, the video specialists: Kling 3.0, from the Chinese company Kuaishou, generates up to 15 seconds with audio in several languages, and Runway bundles its own Gen-4.5 model with others such as Kling 3.0 and Nano Banana Pro in a single subscription. Higgsfield works along the same lines: it brings together models from several companies (Kling, Veo, Seedance…) and adds a marketing studio that turns a product URL into a video ad.

Fourth, the editors where you put the piece together: CapCut, with automatic captions and the Seedance video model on paid plans, and Canva, which adds AI video creation to the templates many small businesses already use.

And Sora? OpenAI closed the app on 26 April 2026, so any tutorial recommending it is out of date.

Don't marry a single tool: models are replaced every few months. Choose by the type of piece you need and review your choice every quarter.

AI avatars and voice: what are they good for in a small business?

Avatars are digital presenters who read your script to camera. HeyGen offers more than 700 stock avatars, can create an avatar of you and translates videos into more than 175 languages and dialects. Synthesia works with more than 140 languages and voices and is geared towards corporate and training video.

For voiceover, ElevenLabs is one of the most widely used options: its Eleven v3 model supports 85 languages, including Catalan, Spanish, French and English, exactly the four languages an Andorran business works in. Note that the commercial licence starts on the Starter plan; the free plan does not include it.

Where do they fit? In explainer videos (how to book at the clinic, how hotel check-in works), in the English or French version of a video you have already filmed, and in internal tutorials. On social media, an avatar presenting your restaurant's offer is obvious, and in a business built on trust it tends to work against you. Cloning your own voice or image is fine; an employee's, only with their written consent.

Which AI formats work on social media and in ads?

AI performs best when it starts from something real (your product, your premises, your face) and multiplies it. These are the formats with the best effort-to-result ratio for a small business:

• Product reels: from two or three good photos, you animate the product (the ski jacket with snow falling, the steaming dish) and edit a 10-15 second reel with text and music.
• Ad variations: the same ad with different backgrounds, formats (1:1, 4:5, 9:16) and hooks to test on Meta Ads or TikTok Ads without producing it again.
• Product shots in new settings: the perfume on a rock with the mountains behind, the sofa in a chalet living room. You start from the real product photo and AI creates the surroundings.
• Captions and language versions: automatic captions in CapCut or Canva make the video work without sound, and with an AI voice you can produce the French version for customers coming from across the border.

Always start from the real product photo. Generating it from scratch looks impressive but is inaccurate, and in an ad that ends up being misleading advertising.

A realistic workflow: from script to publishing

What separates AI content that sells from content that embarrasses you is not the tool, it is the process. For a small business publishing two or three pieces a week, this workflow works:

• 1. Script: the goal of the piece, the hook in the first three seconds, the message and the call to action. An AI assistant can give you ten hooks; you pick the right one.
• 2. Source material: real photos of the product or premises, logo and brand colours. The better the reference, the fewer failed generations.
• 3. Generation: run several versions of each shot and keep the best. Expect some to go wrong; that is normal and part of the cost in credits.
• 4. Human review: someone checks product, prices, text, hands, faces, logos and spelling in Catalan and Spanish. If anyone looks like a real person, it needs a label.
• 5. Editing and publishing: captions, licensed music, the right format for each platform and the AI content box ticked where required.
• 6. Measurement: compare retention and clicks with your filmed pieces. If an AI version performs as well in ads for less money, scale it; if not, go back to the camera.

Keep a simple log: piece, tool, date and who approved it. If a platform or a client asks tomorrow, you have it to hand.

How much does AI video cost? Plans as of October 2026

Almost every tool works on a monthly subscription with credits: each image or second of video uses up credits, and video uses far more than images. These are the prices published on the official websites as of October 2026, in US dollars and excluding tax unless stated. Treat them as a guide: they change often.

For a small business, a sensible combination is one general assistant plus one editor: ChatGPT Plus and Canva Pro come to around €30 a month, and Google AI Pro with CapCut Pro around €40. If you make a lot of video for ads, add a specialist such as Runway or Kling (Kling and Higgsfield change their plans often, so check the official pricing table before paying). And don't forget the biggest cost: the hours of whoever writes, generates and reviews.

Try one month with a specific goal, such as 8 reels and 10 ad variations. If you are not using it every week, cancel: on plans such as Runway's, unused credits do not roll over.

Do you have to label AI-generated content? Can you use it commercially?

You need to label it when it looks real. On top of that, since 2 August 2026 the EU AI Act requires you to disclose when an image, audio or video is a deepfake, advertising included. Andorra is not in the EU, but the regulation also reaches those outside it when the output is used in the Union: if your ads are seen in Spain or France, apply the same standard.

Each platform handles it in its own way (as of October 2026):

• Meta (Facebook and Instagram): adds the "AI info" label to ads created or significantly edited with its AI tools, and if a generated photorealistic person appears, the label sits next to "Sponsored". In 2026 it also began detecting ads made with third-party tools using industry-standard technical signals.
• TikTok: requires clear disclosure, with a sticker or caption, of synthetic content showing realistic scenes. It bans using the likeness or voice of a private individual, and of public figures for commercial endorsements.
• YouTube: when uploading you must disclose AI use if a real person says or does something they didn't, if you alter a real place or event, or if you generate a realistic scene that never happened. It is not needed for captions, scripts, colour filters or your own cloned voice in a voiceover.

Many tools already embed provenance metadata (the C2PA standard) or invisible watermarks, as OpenAI does, and platforms read them. Don't remove them.

Commercial rights and what NOT to do with AI

The main tools let you use what you generate in your business, with caveats. OpenAI assigns you ownership (and does not allow you to pass it off as human-made); Google does not claim ownership of generated content; Runway does not restrict commercial use; Higgsfield, under terms in force since 27 August 2026, recognises that outputs are yours, although it uses content to improve its models. Canva gives you ownership unless the piece includes content from its library. Midjourney requires companies with more than $1 million in annual revenue to be on the Pro or Mega plan, and on lower plans your images are public by default.

Also, other users may get similar results, as OpenAI and Runway warn: if you want exclusivity, for example for your logo, don't hand it to an AI. And most problems come down to judgement:

• Real people without permission: not celebrities, not customers, not employees. Andorra's data protection agency (APDA) publishes an image release consent template, based on Law 29/2021 on personal data protection, that separates social media, website and advertising. Use it for avatars and cloned voices too.
• Other people's logos, brands and characters: no football club shirts, no well-known logo in the background, no cartoon characters. Check every frame, because they sometimes appear without you asking.
• False promises: the dish, room or flat you show must match the real thing. Showing views that don't exist in a property ad, or a generated before-and-after for a clinic, is misleading the customer.
• Customer data in tools without a contract: don't upload customer photos or internal documents to generate content.
• Publishing without review: a spelling mistake in Catalan, a six-fingered hand or a made-up price is spotted in a second and erodes trust.

Quick rule: if a customer would feel misled on seeing the real thing, don't publish it.

Real filming or AI? When each makes sense

AI doesn't replace the camera: it complements it. Film when what you sell is trust and authenticity: your team, your premises, real testimonials, the chef cooking, the real views from a flat. People want to see what they will find, and generated content works against you. That is why we have our own photo and video kit in our social media management service.

Use AI when you need volume, variations or something impossible to film: ten versions of an ad, the product in five settings, a seasonal animation, the French version. That is what we do in AI content and video: pieces generated with AI and reviewed by people.

Often the best approach is to mix both: a real photo and video shoot every so often, which AI multiplies into reels, formats and variants for your Meta Ads and Google Ads campaigns. To see which formats your competitors use, check their live ads in our competitor radar.

Frequently asked questions

Free

3 AI content ideas for your business, free

Tell us what you do and where you post. Someone from our team will email you, within 3 working days, three ideas ready to produce: a reel, an ad and a post, each with its script and the tool we recommend.

• A reel, an ad and a post designed for your sector and audience.
• A script for each piece: hook, message and call to action.
• The recommended tool for each one, and whether AI or real filming suits you better.

A real person from the team replies, not a bot.

Name

Email

Your website (optional)

I have read and accept the privacy policy. Privacy

I want my 3 ideas Rather talk? WhatsApp · +376 866 044

Kim Capel

Founder of Visualtec (2005). AI, automation and SEO consultant with 20+ years in digital marketing.