How we used AI to give 800 Wibra fans a personalised video

The concept sounded simple: every new Wibster receives a tailor-made, personalised video. In practice, it became a small marathon of tech, testing, nuance and efficiency. AI was not a gimmick here, but a real lever for personalisation at scale.

Our objectives were clear:

• Speak to more than 800 people in a warm, personal way.
• Maintain consistency in style and brand expression.
• Combine fixed, general shots with personalised AI-generated fragments.

Building a digital Elise

As host, we chose our colleague Elise. We created a digital version of her using a single video and audio recording. The video was used to train an avatar in HeyGen, the audio modelled her voice for realistic text-to-speech and lip-sync. Result: Elise only had to stand in front of the camera once but appeared in more than 800 different AI-personalised videos.

The AI workflow: static + dynamic

Each video was built from two types of material: static AI-generated footage and images that were identical for every Wibster, and dynamic personalised segments. In those personalised parts, the Wibster’s name appeared on a cake and on a gift card and was spoken out loud by Elise’s AI voice. Thanks to some custom coding, we automatically stitched all these elements together into more than 800 unique final videos.

The AI tech under the bonnet

Techniques used:

• Text-to-image
• Image-to-image
• Text-to-speech
• Image-to-video (with start and end frames)
• Lip-sync

Models used:

• Main image model: Seedream 4
• For image editing: Nano Banana (Pro), Seedream and Qwen
• Main video models: Seedance 1.0 Pro, Veo 3.1, Wan 2.5 and Kling 2.5
• For lip-sync: Sync Labs Lipsync 2 Pro and Pixverse Lipsync

All credits combined? A few thousand euros. Plus, a couple of days of GPU firepower in the cloud via Runpod.io. Because generating 800 videos doesn’t happen in a snap.

We often say it in our AI talks: AI isn’t a super-assistant. It’s a stubborn intern. This case proved that once again.

When things look a bit too AI

Despite tweaks and retries, some shots stayed too plastic. Then you simply have to delete, regenerate and fine-tune. Working with AI also means having the guts to say no when something is ‘almost, but not quite’.

Names are the ultimate boss fight

Especially when it comes to long names, non-Dutch names and anything with a Flemish twist (AI seems to have a natural anti-talent for this). So, we manually checked every single video for mistakes. AI is clever, but pronouncing Ghislaine correctly is still harder than quantum physics.

Flemish is a language of subtleties

Elise’s AI voice sounded surprisingly good, but pitch, emphasis and informal Flemish sounds remain tricky. Even with training, we have to be realistic: AI speaks Standard Dutch, but you can hear when it’s not the real Elise.

Combining is winning

The real power doesn’t lie in AI alone, but in the mix. Combine AI with real video and a thorough manual review and you get scalable personalisation without sacrificing quality. AI brings the speed by taking over the repetitive work. Humans safeguard taste, tone and credibility.

Automation doesn’t work without control

AI can generate over 800 videos. But you still need to check that Ghislaine isn’t coming out of the speakers as ‘Djislainie’. Automation without quality control is just a faster way to make mistakes at scale.

Don’t expect AI to replace craftsmanship

AI speeds things up. AI personalises. AI makes things possible that used to be unaffordable. But the soul, nuance and credibility of a real person, captured on a real set? Those still don’t come pre-installed on your GPU.

Budget: transparency helps

Spending thousands of euros on credits is not nothing. But it’s still peanuts compared to producing more than 800 individual videos from scratch. The financial return is crystal-clear.

AI helps us work faster, personalise at scale and bring creative ideas to life that were previously practically or financially out of reach. But authenticity? That remains a human craft. A real host, real emotions, real nuance: you still film those the classic way. The real magic happens when these two worlds meet. In projects like this one, where technology doesn’t replace creativity, but amplifies it.