Vertex AI Integration

Vertex AI provides higher rate limits than AI Studio for Veo video generation and Gemini image generation.

Config

• GCP Project: gen-lang-client-0013158607
• Location: us-central1
• Auth: GOOGLE_APPLICATION_CREDENTIALS → ~/.config/gcloud/vertex-service-account.json

AI Studio vs Vertex AI

Vertex Model IDs

Connectivity Tests

# Test Veo/text access (no video quota spent)
uv run ~/.openclaw/skills/vertex-ai/scripts/test_vertex_veo.py

# Test image gen via litellm proxy
uv run ~/.openclaw/skills/vertex-ai/scripts/test_vertex_imagen.py

Explicit Vertex Backends

Veo Video Generation

uv run ~/.openclaw/skills/veo-video-gen/scripts/generate_video.py \
  --prompt "A mountain lake at dawn, golden light on water" \
  --filename "/Users/novarussell/.openclaw/media/test-vertex.mp4" \
  --backend vertex-veo-fast \
  --no-upload

Image Generation (direct Vertex model)

uv run ~/.openclaw/skills/nano-banana-pro/scripts/generate_image.py \
  --prompt "a golden sunset, minimalist" \
  --filename "/Users/novarussell/.openclaw/media/test-vertex-img.png" \
  --backend vertex-gemini

Image via litellm proxy (curl)

curl -s -X POST http://localhost:4000/v1/images/generations \
  -H "Authorization: Bearer $LITELLM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"nano-banana-vertex","prompt":"a golden sunset, minimalist","n":1,"size":"1024x1024"}'

Auto-Fallback (Veo)

The veo-video-gen skill automatically falls back to Vertex AI when AI Studio returns 429/ResourceExhausted. No manual intervention needed. Logs: AI Studio unavailable (...). Falling back to Vertex AI...

Auto-Fallback (Images)

The litellm router in config.yaml routes:

• nano-banana → fallback to nano-banana-pro → nano-banana-vertex
• nano-banana-pro → fallback to nano-banana → nano-banana-pro-vertex

Known Limitations

• Reference images NOT forwarded on Vertex image fallback: LiteLLM's vertex_ai/ image path does not expose reference image params. Vertex fallback generates fresh from prompt only.
• Veo style images not supported: Only Veo 2.0-generate-exp supported style reference; Veo 3.1 does not.

REST API Reference

# Text (verify auth)
curl -X POST \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  -d '{"contents":[{"role":"user","parts":[{"text":"hello"}]}]}' \
  "https://us-central1-aiplatform.googleapis.com/v1/projects/gen-lang-client-0013158607/locations/us-central1/publishers/google/models/gemini-2.5-flash:generateContent"