RunwayML Review

There is a specific type of tech product that is more dangerous than a bad product: it’s the one that’s good enough to hook you, but inconsistent enough to frustrate you. RunwayML fits into this category with almost surgical precision.

The platform has built a solid reputation in the world of AI-generated video, and part of that reputation is well-deserved. The Gen-4 model represents a real technical breakthrough, the interface is well-built, and for those just entering the world of AI-generated video, the first week is usually genuinely impressive. Short clips for social media, visual experiments, motion graphics for presentations. It works. The problem starts in the second week.

When you start demanding more from the tool—whether with more complex prompts, scenes featuring specific characters, minimally coherent physics, or continuity between clips—the house of cards begins to collapse. Hands floating at impossible angles, faces that change identity between frames, object physics that seems to have been calculated by someone who’s never seen gravity. And the cruel detail: each of these flaws consumes your credits as usual. You pay for the error. No discussion, no refund, no appeal.
The credit math is where Runway’s marketing really diverges from reality. The mid-tier plan delivers around 187 seconds of generated video in total per month, which would already be meager if every clip were usable. But it isn’t. Real-world tests with heavy usage show utilization rates in the range of 18 to 20 percent. This means that the actual cost per second of usable video is multiplied by five in practice. When you do the math, the price stops seeming reasonable and starts seeming exorbitant.

The Unlimited plan, at $95 a month, is where the situation gets most problematic. The promise of unlimited generations in relaxed mode sounds like the solution to all this, but heavy users report account suspensions for “suspicious activity,” without warning and without a refund of the amount already paid. Support, when contacted, responds with the bureaucratic elegance of someone clearly following a script: “it’s normal platform behavior.” For a tool that charges nearly $100 a month, this is a response that should embarrass any product team.

The comparison with competitors is where it becomes hardest to defend Runway. Pika Labs, which many still treat as a secondary alternative, delivers longer and more usable clips for the same prompt, at a significantly lower cost. Luma Dream Machine has its own limitations, but the relationship between what you pay and what you actually get is more honest. Runway has clearly positioned itself as the industry’s premium brand, but premium implies consistency—and consistency is exactly what’s missing.

The Trustpilot rating of 1.2 out of 5, based on hundreds of reviews, is neither a coincidence nor a coordinated campaign by haters. It is the accumulation of people who paid for something that didn’t deliver what it promised, lost credits due to failed generations, had accounts suspended without a clear explanation, and encountered support that treats legitimate complaints as mere inconveniences.

RunwayML has real technology under the hood. The potential is there. But in 2026, with the video generation market more competitive than ever, technical potential without operational reliability is just expensive marketing. If you need AI video for regular production, always calculate the cost per usable clip before signing up for any plan, and test thoroughly on the cheapest tier before considering any upgrade.