Veo 3.1 on Google Flow and Firefly: Resolution, Credits, Modes

Why model names are not technical specifications

In the rapidly expanding ecosystem of commercial AI video infrastructure, technical directors and pipeline architects frequently assume that a model name represents an absolute, portable specification. If a creative brief specifies that a shot should be rendered using Google Veo 3.1 Fast, engineers often assume that the model's capabilities, input modalities, duration parameters, and pricing structures will remain identical regardless of which cloud hosting provider executes the job. In production reality, this assumption is completely false.

Cloud aggregators and enterprise hosting platforms implement proprietary middleware, custom conditioning wrappers, and divergent billing abstractions around base foundation models. In traditional software engineering, an API client interacting with an identical model endpoint expects deterministic parameter validation and uniform billing logic across cloud providers. In commercial foundation model hosting, however, platforms frequently inject proprietary prompt rewriters, frame sampling schedulers, temporal latent governors, and arbitrary duration clamps. The exact same underlying model—Veo 3.1 Fast—operates under radically contradictory constraints depending on whether it is accessed through Google Flow or Adobe Firefly. A prompt specification engineered for Google Flow will fail immediately or produce unexpected visual compromises if dispatched to Adobe Firefly without translation. Understanding these platform-specific idiosyncrasies is essential for maintaining production timelines and preventing catastrophic budget overruns.

Google Flow: fixed 8 s durations and start-plus-end frames

When operating within Google Flow, Veo 3.1 Fast occupies a highly specialized operational niche. Under Flow's architecture, Veo 3.1 Fast does not permit variable duration selection: every generation is locked to an immutable 8 s fixed duration, billed at a flat rate of 20 credits. More importantly, Veo 3.1 Fast functions as Google Flow's exclusive rail for start-and-end frame interpolation (FLF2V). If a director requires a video clip that begins precisely on one photographic keyframe and concludes seamlessly on a second keyframe, Veo 3.1 Fast on Flow is the only model that accepts both inputs.