Generate AI videos from text prompts using Google Veo

This promptfoo example demonstrates Google's Veo video-generation models, creating AI-generated video from text prompts (and from a starting image). It runs against either Google AI Studio/Gemini API or Google Cloud Vertex AI, and covers three feature areas: text-to-video (promptfooconfig.yaml), image-to-video (promptfooconfig-image.yaml), and Veo 3.1 video extension (promptfooconfig-extension.yaml).

Six model provider IDs are available, all supporting 4, 6, or 8 second generation. On Google AI Studio/Gemini: google:video:veo-3.1-generate-preview (with video extension support), google:video:veo-3.1-fast-generate-preview (faster generation), and google:video:veo-3.1-lite-generate-preview (lower cost). On Vertex AI: vertex:video:veo-3.1-generate-001 (GA), vertex:video:veo-3.1-fast-generate-001 (faster GA), and vertex:video:veo-3.1-lite-generate-001 (lite preview). Configuration options include aspectRatio (16:9 default or 9:16), resolution (720p default or 1080p), durationSeconds (4, 6, or 8; extension requires 8), personGeneration (allow_adult or dont_allow), negativePrompt (concepts to avoid), image (source image for image-to-video), lastImage (an end frame for interpolation), sourceVideo (a generated video URI for AI Studio, or base64/file:///gs:// for Vertex AI), and referenceImages (up to 3 asset reference images).

Video extension on Veo 3.1 works by copying response.metadata.videoUri from an exported evaluation into sourceVideo and setting durationSeconds: 8; Veo adds 7 seconds to the source video. On AI Studio, the original generated file must finish processing in Google first - downloaded files and base64 input are not accepted for extension there, and Vertex operation IDs are not accepted either. On Vertex AI, extension instead accepts a gs:// URI, base64 data, or a file:// path (not an operation ID), and extension requests omit durationSeconds entirely - any configured duration is ignored with a warning.

Use it to generate or extend AI video from text or image prompts with Veo 3.1, comparing standard, faster, and lower-cost model variants, or to test image-to-video and video-extension workflows before building on them. Generated videos are stored in promptfoo's blob storage with content-addressable hashing for automatic deduplication, and are served via the local server for viewing in the UI; pass --no-cache to force regeneration instead of reusing a cached result. Veo models run as long-running operations with polling for completion, so expect generation to take real time rather than returning instantly.

gcloud services enable aiplatform.googleapis.com
gcloud auth application-default login
export GOOGLE_PROJECT_ID=your-project-id

That is the Vertex AI setup path; for AI Studio, set GOOGLE_API_KEY instead. Run the example with npx promptfoo@latest eval. google:video:* provider IDs route to Google AI Studio by default; use the vertex:video:* IDs above for explicit Vertex AI routing.

Uses promptfoo's Google provider against either the Google AI Studio/Gemini API or Google Cloud's Vertex AI API, authenticated via an API key or gcloud application-default credentials respectively.

Developers evaluating or building on Google's Veo 3.1 video-generation models - text-to-video, image-to-video, or extending an already-generated video - who want working promptfoo configurations covering both AI Studio and Vertex AI deployment paths.