Why does AI-generated stop motion animation have inconsistent frame rates?

AI stop motion has inconsistent frame rates because video models synthesize continuous, interpolated motion — they imitate the stop-motion look without locking a temporal cadence, so held frames drift. In testing Google Omni Flash, stop-motion clips oscillated between 12 FPS and 8 FPS. The reliable fix: generate stills and sequence them at a fixed frame rate yourself.

AI video models produce inconsistent stop-motion frame rates because they generate continuous, interpolated motion and only approximate the stop-motion aesthetic — there is no mechanism in the model that enforces discrete held frames at a fixed cadence. Real stop motion is defined by its timing: each pose is held for an exact number of frames before snapping to the next. A diffusion video model has no concept of that cadence; it renders smooth motion and then stylizes it to look choppy, so the "choppiness" wanders instead of holding steady.

This shows up as measurable oscillation in output. In one round of testing that covered 30+ generated outputs on Google Omni Flash, stop-motion clips oscillated between 12 FPS and 8 FPS — the apparent frame rate shifted mid-clip and between generations, because the model was re-deciding the hold length shot by shot rather than obeying a locked one. Recent research on measuring physical frame rate from visual dynamics confirms the underlying issue: video models don't encode a stable temporal sampling rate, so the perceived FPS of their output drifts (The Pulse of Motion).

This is a cross-model limitation, not an Omni-specific bug. Every current video model — Veo, Kling, Seedance 2.0, Runway — is trained on continuous footage and optimizes for smooth temporal consistency, which is exactly the property stop motion violates on purpose. Frame interpolation, the same machinery that makes normal AI video look fluid, actively fights intentional holds — animators debate this trade-off constantly, and the consensus is that interpolation and hand-timed animation pull in opposite directions.

The working fix is to take frame timing out of the model's hands. Generate your poses as still images, then assemble them at a locked frame rate in your edit, holding each still for a fixed number of frames at your delivery frame rate. Because the cadence is set in post, it cannot drift. The invideo agent supports this workflow directly: generate the pose stills with image models like Nano Banana or GPT-Image-2 for character consistency across poses, then sequence them at your chosen cadence — you're using the AI for the frames, not the timing.

On where the models stand today, invideo's testing was blunt: "the current state of the textures that Google is offering, I'm not so sure if they're ready for prime time cinema yet." Frame-rate cadence sits in the same category — a current ceiling of the models, with a workflow-level workaround that already works.

Watch some of these to see what works for you:

the current state of the textures that Google is offering, I'm not so sure if they're ready for prime time cinema yet.

— invideo's creative team, from testing 30+ Google Omni Flash outputs