I never thought we would direct a live-action video for a leading dental brand without a single day on set. Yet here we are – our team at Now We Collide just delivered a cinematic-quality series of videos using our GenAI pipeline. This wasn’t a sci-fi experiment or hype for hype’s sake; it was a real client campaign, completed in weeks instead of months.
I’m excited to share how our Collide-AI post pipeline – using a quirky tool called Nano Banana alongside the latest video gen models – made it possible. In this post, I’ll pull back the curtain on how we achieved shot-to-shot consistency in AI-generated video and what it means for agencies, brands and the future of content production.
The Character Consistency Problem
Every video producer knows the pain points: desired production and creative value, cost and time. Our dental client needed a series of short product videos with consistent characters, settings, scenes and props across multiple shots – think a friendly dentist in a clinic, close-ups of a patient, explainer graphics and product pack-shots. Typically, we’d shoot everything live, in-camera. But due to tight timelines and logistical hurdles, live-action simply wasn’t an option.
The catch? For characters and product shots Generative AI video has a notorious consistency problem. One scene might render our dentist with different facial features or the clinic decor changing colour shot-to-shot – obviously a brand safety and scene continuity nightmare. Early text-to-video models struggled to maintain the same character or environment across cuts. We also had to ensure the output met strict brand guidelines (free from unusual AI artefacts or off-brand elements) and complied with regulatory limits (no unproven product claims in visuals). The challenge was clear: how could we leverage AI’s speed and creativity without sacrificing continuity or quality?
The Breakthrough
Our breakthrough came through 'test and learn' R&D, looking at how multiple AI tools 'in the AI tech stack' can work seamlessly toegther for a desired outcome – while adding some old-school production discipline to the mix. We treated the AI models as “collaboration tools” that needed strict guidance. We started by building a “character bible” for our characters – a reference image set locking down appearance, wardrobe and even lighting. Likewise, we created a “scene bible” for the clinic and props. These weren’t just moodboards; they became the ground truth images we would feed into the generative models to anchor their outputs.
We then mapped out a shot graph – essentially a storyboard of the sequence – and carefully crafted text prompts for each shot, referencing our hero character and set pieces by name. Crucially, we used seed control (fixing random seeds) and image references so the AI started from the same visual DNA each time. Picture a “consistency stack” diagram: at the base are our character and scene bibles, above them a shot-by-shot plan, all feeding into the AI generator with consistent seeds. This stack ensured that as we generated each scene, the dentist and talent were consistent in every frame and the clinic setting didn’t magically morph. Shot-to-shot continuity – solved.
When the first fully AI-generated draft came through with a cohesive look and narrative flow, it felt like a moon landing. We had a smiling dentist character who remained identical across an entire 30-second spot – same face, same outfit, same office – without re-shoots or VFX tweaks. Cinematic framing and camera moves were on point, too. And thanks to careful prompting, even the product packaging appeared accurately in the final pack-shot, with no misspelt labels or off-brand colours. For the first time, we saw AI video consistency reach a level suitable for a brand ad. It was a genuine “aha” moment for our creative team and the client alike.
Pipeline
How did all these pieces come together in practice? Our Collide-AI pipeline closely mirrored a traditional production workflow, with a twist of AI at its core. We kept our creative process practitioner-led – our human team remained the directors and quality controllers at every step. Here’s how a project like this flows from brief to final QC:
Pipeline at a Glance:
In short, our AI pipeline compressed what normally might be a two-month project into a matter of weeks. We moved from ideation to final cut in under a month, and at perhaps one-third the cost of a live-action shoot. The client got their content faster, without compromising on quality or brand consistency. And our creative team? We got a taste of a new, more agile way of working – one where we could try wild ideas (want an alternate background? just prompt it) without expensive re-shoots. It’s not less work, just different work – more upfront planning and iterating with AI, but far less logistics and physical constraints.
Implications
As a Creative Technologist who’s spent decades producing content in new and innovative ways, this project was eye-opening. The implications for agencies and brands are huge. AI video production isn’t a novelty anymore; it’s a viable option when traditional shoots are impractical or impossible. We’ve proven that with the right approach, AI can deliver scene consistency and character consistency on par with a live shoot – which has been a major barrier until now.
For agencies, this means we can be more nimble in serving client needs. Imagine responding to a client brief by generating a content prototype in 48 hours – something unthinkable with physical production. We can also tackle projects that were once budget-prohibitive. Need 100 personalised video variations for a campaign? With AI, scaling up doesn’t linearly scale the cost like it would with 100 separate shoots.
Brands, on the other hand, gain a creative superpower: the ability to produce brand-safe AI content quickly without sacrificing quality. It’s important to note, this doesn’t eliminate the need for live action or real creative work – but it adds a powerful tool to the arsenal. For instance, sensitive scenarios (like a medical explainer that’s tricky to film) or imaginative storytelling (fantastical visuals, global settings) become easier to execute. We’re already discussing with clients how AI pipelines could create content libraries for always-on social media, international market versions, or rapid-response topical ads – things that used to be constrained by production lead times.
However, agencies will need to evolve. Our producers and art directors had to become prompt designers and AI curators. We spent as much time art directing the AI as we might with a human crew. This is a new skill set, and it will be in high demand. We also learned that cross-functional collaboration (creative, tech, strategy, legal) is key – you need everyone aligned when you venture into AI-driven production.
Bottom line: AI isn’t coming for filmmakers; it’s collaborating with them. The agencies and brands that embrace this, with the right ethical guardrails (more on that next), will be able to tell stories in ways and at speeds we never could before. The future of production will be a hybrid one – and after this project, I’m convinced that’s a good thing.
If you’re as intrigued as we were about the potential of AI in video production, want to see this campaign or want to find out more about our Collide-AI tool chain for a project just like this, let’s talk. We’re offering hands-on demos and workshops for brands and partners to experience Collide-AI first-hand. Come see how a brand-safe, cinematic AI video comes together, and imagine what it could do for your content strategy.
Whether you have a campaign that can’t wait for a full shoot, or you’re just curious about the workflow, we’d love to collaborate and explore the possibilities with you. The future of filmmaking is here, and it’s time to get your teams ready for it.