We backed Runway in 2024 as part of our thesis that the generative creative supply chain was being rebuilt at the infrastructure level. At the time, the quality of AI-generated video was impressive in laboratory conditions and unreliable in production conditions. Marketing teams experimenting with it were producing output that read as AI-generated — the temporal inconsistencies, the physics errors, the character coherence failures that were the tells of first-generation video generation models.
That quality picture has changed substantially. What has not changed is the broader commentary about the category, which is still largely stuck in the "experimental / promising but not ready" register. I want to be direct: for a well-defined set of commercial video applications, AI video generation is ready. Understanding where that line is — and where it genuinely is not — is more useful than either uncritical enthusiasm or reflexive skepticism.
What commercial maturity means here
Commercial maturity does not mean the technology is perfect. It means the output quality is above the threshold at which a commercial buyer would deploy it in production — not as an experiment but as a standard part of their workflow — and that the cost-to-quality ratio makes sense compared to the alternative.
For AI-generated video in marketing, that threshold is context-specific. A 15-second product showcase video for a paid social ad has a different quality bar than a 90-second brand film. A social media asset demonstrating a product feature has a different quality bar than a broadcast television commercial. The error in most commentary about AI video quality is treating it as a single question ("is AI video good enough?") when it is really a dozen different questions indexed against a dozen different commercial contexts.
For short-form paid social creative — the 6 to 30 second formats that dominate performance marketing spend — AI video generation has crossed the commercial threshold. The output is good enough to test, good enough to deploy, and good enough to differentiate from the baseline creative in a campaign. More importantly, it is good enough to generate the volume of creative variants that modern paid social optimization requires, which was the actual bottleneck that no amount of human creative production was going to solve.
The variant volume problem that video solves
The central economic problem in performance marketing video is not quality. It is volume. A performance marketing team running paid social campaigns at meaningful scale needs creative variants — different hooks, different visual approaches, different value proposition framings — to feed the platform optimization algorithms that determine which creative wins. The team that can test 30 variants finds a winner faster than the team that can test 5.
Human creative production does not scale to that variant volume without a linear headcount increase. A video production team producing at high quality can produce 5 to 8 finished video variants per week under normal working conditions. AI video generation, in the specific context of performance marketing short-form creative, can generate 30 to 50 variants in the same time with a creative director setting the direction and reviewing outputs. The creative director's role changes — less execution, more curation and direction — but the throughput per unit of human creative work increases by a factor of 5 to 10.
This is not theoretical. We are seeing it in the performance marketing teams that have adopted AI video generation tools. The ones that have integrated it well are running more creative tests, finding winning variants faster, and reducing creative fatigue (the degradation of ad performance over time as an audience tires of seeing the same creative) because they have a larger pool of variants to rotate through.
Where the quality ceiling is still real
We should be honest about what does not work yet. Character consistency — maintaining a recognizable human character across a generated video sequence — remains technically difficult. If your brand relies on a specific spokesperson, a consistent mascot, or any scenario where a human face needs to be coherent and recognizable across more than a few seconds of generated footage, current AI video generation is not reliable enough for production use. The models fail at this in ways that are visible and brand-damaging.
Long-form narrative video — anything over 60 to 90 seconds that requires coherent scene structure, consistent visual context, and narrative logic across cuts — is also not ready for commercial deployment. The temporal coherence problems that affect all current generation models become compounding failures in longer content. A 30-second failure is recoverable in editing. A 90-second failure means re-generating from scratch.
Brand film and brand identity work — the kind of video that a major brand invests in to establish its visual language for a year — is still a human creative problem. This is less a quality question than a creative direction question. AI video generation is good at executing creative direction quickly. It is not yet good at originating the creative direction that would genuinely represent a brand's aspiration. The prompting quality that drives the best AI video output still requires a human creative director who knows what they are trying to achieve. The tool amplifies that human direction; it does not replace it.
What this means for the companies building in this space
The market structure for AI video generation in marketing is still being established. The companies that have moved fastest from research-stage to production-grade output have done so by focusing relentlessly on specific commercial use cases rather than trying to solve video generation generally. The performance marketing use case — short-form, variant-heavy, direct-response creative — has driven more production-grade development than any other commercial application, because the demand signal from performance marketers is clearest and most measurable.
We see the next wave of development focused on two areas. First, brand consistency enforcement: tools that allow a brand to define its visual language — color palette, visual texture, character traits — in a way that is consistently applied across generated content. The underlying model capability exists; the tooling for brand teams to specify their requirements at the right level of abstraction is still underdeveloped. Second, template-to-video workflows for e-commerce: the product-on-background video variant generation use case, where a product image becomes a motion video in a defined visual context, is an area where AI video generation can close a very specific production gap in e-commerce marketing.
When we backed Runway, we made a bet that commercial-grade AI video for marketing would arrive on a two-to-three year horizon. We are at that horizon now for a defined subset of use cases. The remaining quality gaps are real and will be closed — the research progress in the underlying models is not slowing down — but they should not obscure the fact that for performance marketing short-form creative, the technology is ready. The teams that are waiting for perfection before adopting are ceding a creative testing advantage to the ones who are already running.