AI video tools and how they’re changing business communication [Q&A]

The use of AI video has exploded in the past year. But while it’s deepfakes that make the headlines the technology also has the potential to change how businesses create and use video content in their messaging.

We spoke to Victor Erukhimov, chief executive officer of CraftStory, to find out more about AI video and how businesses can exploit it to their advantage.

BN: How do you see AI video tools changing the businesses create and use video content?

VE: AI video tools allow businesses to radically reduce both the effort and the time required to create video content. This spans everything from ads and launch videos to explainers, tutorials, and educational content. What used to require specialized teams, motion graphics, or custom animation can now be produced in hours -- and updated on demand.

At the same time, the demand for video has exploded. With mobile-first consumption and social media shaping how people learn and buy, companies need to communicate far more through video than ever before.

But there’s a problem. Most on-camera videos are boring, repetitive, and look identical across companies. They blend together. Anything more dynamic -- animated sequences, product walkthroughs, cinematic shots -- is usually too expensive and time-consuming to produce at scale.
This is exactly what CraftStory changes. We let companies create rich, expressive, human-centric videos that go far beyond talking-head content -- but without the high cost or complexity of traditional production. As a result, brands can stand out, publish more frequently, and build video channels that actually grow.

AI video isn’t just making production faster; it’s unlocking entirely new formats of storytelling that used to be out of reach for most businesses.

BN: Most AI models today still focus on short clips. Why has long-form, five-minute video been such a hard problem for the industry to solve?

VE:Long-form video is fundamentally harder because diffusion models struggle to maintain consistency over long timelines. When you try to generate several minutes of footage in one pass, the model needs enormous amounts of training data, memory, and compute just to keep a character’s appearance, gestures, and environment stable. Beyond a certain duration, the video simply starts to drift -- faces shift, lighting changes, and motion becomes inconsistent. That’s why most models today cap out at short clips.

Our research team solved this by rethinking how long video is generated. Instead of forcing a single diffusion process to cover a long interval, we break the video into shorter segments and run multiple diffusion processes in parallel -- while preserving character identity, motion, and visual coherence across all segments. This lets us scale to minutes instead of seconds without losing quality.

There’s also a practical challenge: render time. A five-minute video is an order of magnitude heavier to compute than a 20-second clip. Long render times make iteration painfully slow for creators.

We’ve spent a huge amount of effort optimizing our pipeline so creators can actually iterate. Today, we can generate a one-minute video in about 30 minutes, which makes tweaking scripts, shots, or gestures feasible.

In short, long-form video is hard because it requires both algorithmic breakthroughs and massive engineering optimization -- and that’s exactly where we set our sights and innovation.

BN: There’s a big debate about data sources for AI. How important is it for video models to be trained on curated or proprietary footage rather than scraped internet content?

VE: For us, this made an enormous difference. We built our own multi-camera capture system that records high-frame-rate (HFR) footage synchronized across several angles. This lets us capture the subtle dynamics of human movement that standard 30 fps internet video simply misses.

For example, human fingers move incredibly fast, and at 30 fps they appear as a blur -- meaning a model trained on that footage can never learn the correct motion. Our high-frame-rate, synchronized captures produce crisp, detailed hand and facial movements, which dramatically improves motion realism.

The result is that we can train a far better model with far less data, because the data we train on is clean, consistent, and physically accurate.

BN: A lot of companies are looking at AI video for training, marketing, and product demos. What kinds of use cases do you think will see real adoption first?

VE: We’re already seeing strong adoption in training and instructional videos. AI avatars have taken off in L&D because they let teams update content easily and instantly without reshooting -- and consistency matters a lot in training environments.

The next wave is product demos, explainers, and lightweight marketing videos. These are high-volume, repetitive formats where companies need to ship updates fast, localize across markets, and keep the messaging consistent. AI video is a perfect fit.

We’re also seeing early experimentation in advertising. The most visible example is the controversial Coca-Cola Christmas ad, which showed the industry that brands are willing to explore AI in high-stakes creative work -- even if the execution is still evolving.

BN: The AI video market already feels crowded, with both Big Tech and startups jumping in. How do you expect the space to evolve over the next few years?

VE: The AI video market looks crowded today, but most of what we’re seeing is an explosion of tools -- not complete solutions for brands. Generating a clip is easy; producing an actual marketing video still requires multiple iterations, creative decisions, and a team that knows how to pull all the pieces together. That gap between ‘a model’ and ‘a finished video’ is where most current products fall short.

Over the next few years, the winners will be the platforms that make end-to-end video creation truly simple. Brands want to go from idea to script to finished video without needing a motion graphics team, a director, or someone stitching assets together in post-production.

That's exactly what we’re focused on. We’re building a system where the model follows both the script and high-level director instructions -- including dynamic camera movement for shots like walk-and-talk. Our upcoming text-to-video model will let creators specify tone, pacing, framing, gestures, and camera choreography in natural language.

Image credit: phonlamai/depositphotos.com