fal.ai appears in the catalog as a media inference provider with both image and video endpoints. Media pricing cannot be reduced to one provider-wide rate. Each model has an exact endpoint, supported output profile, and billing rule, so the calculator represents every model as a separate offer.
Image generation is modeled from monthly image count and output dimensions. Video generation is modeled from clip count, duration, resolution, and audio mode. These categories remain independent from language, search, and scraping services, allowing a product to combine media generation with other AI features.
The catalog currently includes multiple fal.ai offers. That is why each model card has its own calculator action. A generic provider button would hide the material difference between an image endpoint and a video endpoint.
Supported image rules include a flat amount per image and a rate based on megapixels. A megapixel rule can include a billing step, so the charge may round the output area before applying the rate. Video rules can charge per fixed-profile clip or per output second with minimum and step durations.
The engine matches the requested width, height, duration, resolution, and audio setting against the offer profile before pricing it. A model is unavailable when the requested profile is not supported; the calculator does not substitute a nearby format or silently estimate GPU runtime.
The model cards above provide the authoritative endpoint ID, output profile, billing rule, source, and verification date from the current snapshot. Prices stay in structured catalog data instead of Markdown.
A lower price does not establish better or worse visual quality. Output style, prompt adherence, generation time, and reliability require comparable evidence. Fastest and Balanced only use recent evidence with matching media profiles and methodology. Missing evidence produces an unavailable recommendation rather than a marketing-based ranking.
The estimate also excludes storage, delivery, moderation, retries, taxes, and application infrastructure. A failed generation policy can materially change real spend, so measured retry volume should be included in exact mode when available.
Choose Estimate with… on an image or video offer. The calculator enables the corresponding scenario and locks that model. Current catalog profiles match the default 1024×1024 image or five-second 720p silent-video examples where applicable, and every field remains editable.
Review the exact endpoint and verification date before production use. Provider-level pricing pages are helpful orientation, but the endpoint source linked on each card is the relevant pricing authority.
