[softmarshmallow]

Add the November 5 release of google/gemini-omni-1.2-flash as a video model with native Google, fal, OpenRouter and Vercel bindings. Preserve Omni 1.1 and every existing model; this is a data-only addition with no schema change.

The card records 3–15-second generation, native audio, fal text/image generation parameters and regular resolution-specific prices. First/last-frame interpolation remains limited to 10 seconds. Native token billing stays separate from each provider’s output-second prices. Vercel confirms the identity and video output, but its checked sources do not establish a video-output rate, so none is invented. No Grida binding is advertised before public availability.

Sources:

• Google launch, November 5: exact release and model identity.
• Google model reference and regular pricing: native durations, limits, audio and token meters.
• fal text-to-video and exact schema: regular per-second prices and parameter bounds.
• fal image-to-video and exact schema: matching output rates, included input images and conditional end-frame duration limit.
• OpenRouter video discovery and model page: identity, frame inputs, supported output settings and undiscounted video prices.
• Vercel discovery and model page: exact identity and video-output availability; video price remains unrecorded.

Validation: uv run scripts/build.py, uv run scripts/check.py, uv run pytest (302 passed), and npx -y oxfmt@0.71.0 --check. Complete generated diff review confirms exactly one added card and no changes to existing facts. No paid inference.