danishi/veo

Veo - AI Video Generation Skill

Use the Python script in scripts/ to generate videos via Google Gemini Veo 3.1.

Default model is the cheapest one. The more expensive models are used only when explicitly requested.

| Flag | Model | When used | |---|---|---| | (default) | veo-3.1-lite-generate-preview | Cost-effective; used unless a flag says otherwise | | --pro | veo-3.1-generate-preview | Premium / cinematic, 4k, native audio — explicit only | | --fast | veo-3.1-fast-generate-preview | Faster generation — explicit only |

--pro and --fast are mutually exclusive. The VEO_MODEL env var overrides everything.

Prerequisites

1. Install dependencies

pip install google-genai

2. Configure API credentials (one of the following)

Option A: Gemini Developer API (recommended for personal use)

Set the GEMINI_API_KEY environment variable. Get a key at https://aistudio.google.com/apikey

export GEMINI_API_KEY="your-api-key"

Option B: Vertex AI API (for Google Cloud users)

Set GOOGLE_CLOUD_PROJECT and optionally GOOGLE_CLOUD_LOCATION. Requires a GCP project with the Vertex AI API enabled and Application Default Credentials configured (gcloud auth application-default login).

export GOOGLE_CLOUD_PROJECT="your-project-id"
export GOOGLE_CLOUD_LOCATION="us-central1"   # optional, defaults to us-central1

Priority: If both GOOGLE_CLOUD_PROJECT and GEMINI_API_KEY are set, Vertex AI is used.

3. Optional environment variables

| Variable | Default | Description | |---|---|---| | VEO_MODEL | veo-3.1-lite-generate-preview | Force a specific model (overrides flags) | | VIDEO_OUTPUT_DIR | ./veo-videos | Default output directory | | VEO_NO_SSL_VERIFY | (unset) | Set to 1 / true / yes to disable SSL certificate verification |

---

Script

scripts/generate.py - Video generation

Video generation is a long-running operation; the script submits the request and polls until the video is ready, then downloads and saves the .mp4.

Text-to-video (default Lite model)

python scripts/generate.py "a cat surfing a wave, cinematic lighting" -o cat.mp4

Portrait, 6-second clip

python scripts/generate.py "slow zoom over a city at dawn" --ratio 9:16 --duration 6 -o city.mp4

Image-to-video (animate a starting frame)

python scripts/generate.py "gentle breeze, leaves drifting" -i start.jpg -o anim.mp4

First + last frame (constrain the ending)

python scripts/generate.py "morph from sunrise to sunset" -i sunrise.jpg --last-frame sunset.jpg -o morph.mp4

Premium model with 1080p (explicit opt-in)

python scripts/generate.py "epic drone shot over mountains" --pro --resolution 1080p -o drone.mp4

Multiple variations

python scripts/generate.py "abstract liquid motion" -n 4 -o liquid.mp4
# -> liquid_0.mp4, liquid_1.mp4, liquid_2.mp4, liquid_3.mp4

Disable SSL verification (for corporate proxies or self-signed certs)

python scripts/generate.py "ocean waves" --no-ssl-verify -o waves.mp4

JSON output (for programmatic use)

python scripts/generate.py "ocean waves" --json -o waves.mp4

Full options

usage: generate.py [-h] [-o OUTPUT] [-i IMAGE] [--last-frame PATH]
                   [-r {16:9,9:16}] [--resolution {720p,1080p,4k}]
                   [-d DURATION] [-n COUNT]
                   [--person-generation {auto,dont_allow,allow_adult,allow_all}]
                   [--negative-prompt TEXT] [--seed SEED] [--pro] [--fast]
                   [--poll-interval SECS] [--timeout SECS] [-v] [--json]
                   [--no-ssl-verify] prompt

Arguments:
  prompt                  Text prompt for video generation

Options:
  -o, --output PATH       Output .mp4 path (index suffix added if -n > 1)
  -i, --image PATH        Starting frame image (image-to-video)
  --last-frame PATH       Ending frame image (constraint)
  -r, --ratio RATIO       Aspect ratio: 16:9 or 9:16 (default: 16:9)
  --resolution RES        720p, 1080p, or 4k (4k requires --pro)
  -d, --duration SECS     Clip duration in seconds (default: 8)
  -n, --count N           Number of videos, 1-4 (default: 1)
  --person-generation P   auto / dont_allow / allow_adult / allow_all
                          (default: auto = allow_all on Gemini Developer API,
                          allow_adult on Vertex AI)
  --negative-prompt TEXT  Things to avoid in the video
  --seed SEED             Seed for improved consistency
  --pro                   Use the premium Veo 3.1 model (explicit)
  --fast                  Use the Veo 3.1 Fast model (explicit)
  --poll-interval SECS    Seconds between status checks (default: 10)
  --timeout SECS          Max seconds to wait (default: no limit)
  -v, --verbose           Show detailed output
  --json                  Output result as JSON
  --no-ssl-verify         Disable SSL certificate verification

---

Capabilities & Limits

| Capability | Lite (default) | Pro (--pro) / Fast (--fast) | |---|---|---| | Text-to-video | ✅ | ✅ | | Image-to-video (first / last frame) | ✅ | ✅ | | Aspect ratios | 16:9, 9:16 | 16:9, 9:16 | | Resolution | 720p, 1080p (8s) | + 4k | | Duration | 4 / 6 / 8 s | 4 / 6 / 8 s | | Reference images, video extension | ❌ | ✅ (Pro / Fast) |

Notes:

• 1080p typically requires an 8-second duration.
• 4k is not available on Lite — use --pro.

Person generation policy

The --person-generation policy controls whether people appear in the output:

| Platform | Supported values | Default (auto) | |---|---|---| | Gemini Developer API (GEMINI_API_KEY) | allow_all only — dont_allow / allow_adult are currently rejected | allow_all | | Vertex AI (GOOGLE_CLOUD_PROJECT) | dont_allow, allow_adult, allow_all | allow_adult |

• The default is auto, which picks a platform-safe value automatically.
• As a safety net, if the API rejects the chosen value with a personGeneration ... not supported error, the script automatically retries once with allow_all — so a single run succeeds without manual intervention.

---

Prompting Tips

A strong video prompt describes motion and cinematography, not just a static scene:

1. Subject & action — what happens (a fox leaps across a stream)
2. Camera — shot type / movement (slow dolly-in, aerial drone shot, handheld)
3. Setting & lighting — (misty forest at golden hour)
4. Style — (cinematic, anime, documentary, 35mm film)
5. Pace — (slow motion, time-lapse)

Use --negative-prompt to steer away from unwanted elements, and --seed to keep results consistent across runs.

See references/prompts.md for category-specific templates.

---

Error Handling

| Error | Solution | |---|---| | google-genai package not installed | Run pip install google-genai | | No API credentials found | Set GEMINI_API_KEY or GOOGLE_CLOUD_PROJECT | | Veo 3.1 Lite does not support 4k | Use --pro for 4k, or pick 720p/1080p | | No videos were generated | Rephrase the prompt; it may have been blocked | | personGeneration is currently not supported | The Gemini Developer API only supports allow_all. The script auto-retries with allow_all; or pass --person-generation allow_all explicitly | | Content blocked by safety filters | Rephrase; check --person-generation policy | | Timed out ... waiting | Increase --timeout or omit it (no limit) | | API rate limit reached | Wait and retry | | SSL: CERTIFICATE_VERIFY_FAILED | Use --no-ssl-verify or set VEO_NO_SSL_VERIFY=1 |