Google¶

Google provides Veo 3 video generation and Imagen 3 / Gemini image generation via the google-genai SDK. Supports both the Gemini Developer API (API key) and Google Cloud Vertex AI (service account / ADC).

Installation¶

pip install tarash-gateway[google]

Image Generation¶

Quick Example¶

from tarash.tarash_gateway import generate_image
from tarash.tarash_gateway.models import ImageGenerationConfig, ImageGenerationRequest

config = ImageGenerationConfig(
    provider="google",
    model="imagen-3.0-generate-002",
    api_key="AIza...",
)

request = ImageGenerationRequest(
    prompt="A serene Japanese garden with cherry blossoms, photorealistic",
    aspect_ratio="16:9",
)

response = generate_image(config, request)
print(response.images[0])

Parameters¶

Parameter	Required	Supported	Notes
`prompt`	✅	✅	Text description of the image
`aspect_ratio`	—	✅	Supported by Imagen 3 and Gemini image models
`n`	—	✅	Number of images to generate (see image count rules)
`seed`	—	—	Not supported
`negative_prompt`	—	—	Not supported

Supported Models¶

Model ID	Notes
`imagen-3.0-generate-001`	Imagen 3
`imagen-3.0-generate-002`	Imagen 3 latest
`gemini-2.5-flash-image`	Nano Banana
`gemini-3-pro-image`	Gemini pro image

Imagen models use the dedicated generate_images API. Gemini image models (gemini-*) use the generate_content API. Tarash detects automatically based on model name prefix.

Video Generation¶

Quick Example¶

from tarash.tarash_gateway import generate_video
from tarash.tarash_gateway.models import VideoGenerationConfig, VideoGenerationRequest

config = VideoGenerationConfig(
    provider="google",
    model="veo-3.0-generate-preview",
    api_key="AIza...",
)

request = VideoGenerationRequest(
    prompt="A drone shot over a coastal cliff at sunset",
    aspect_ratio="16:9",
    resolution="1080p",
)

response = generate_video(config, request)
print(response.video)

Parameters¶

Parameter	Required	Supported	Notes
`prompt`	✅	✅	Text description of the video
`aspect_ratio`	—	✅	Passed to Veo3
`resolution`	—	✅	`720p`, `1080p`, `4k`
`image_list` (reference/asset/style)	—	✅	Up to 3 reference images
`image_list` (first_frame)	—	✅	Interpolation mode start
`image_list` (last_frame)	—	✅	Interpolation mode end
`negative_prompt`	—	✅
`seed`	—	✅
`enhance_prompt`	—	✅
`generate_audio`	—	✅

Supported Models¶

Model ID	Notes
`veo-3.0-generate-preview`	Veo 3 — standard
`veo-3.1-generate-preview`	Veo 3.1 — longer sequences
`veo-3.1-fast-generate-preview`	Veo 3.1 Fast — faster generation

Supported resolutions: 720p, 1080p, 4k

Image-to-Video¶

Veo3 supports using reference images as start and/or end frames for interpolation. Pass images via image_list with type="first_frame" and/or type="last_frame".

from tarash.tarash_gateway.models import ImageType

request = VideoGenerationRequest(
    prompt="The sun rises above the mountain",
    image_list=[
        ImageType(image="https://example.com/dawn.jpg", type="first_frame"),
        ImageType(image="https://example.com/sunrise.jpg", type="last_frame"),
    ],
)

Provider-Specific Notes¶

Authentication:

Gemini Developer API: Set api_key to your AIza... key. No provider_config needed.
Vertex AI: Set api_key=None and supply a provider_config dict with at minimum gcp_project. Optional keys: location (default us-central1) and credentials_path (path to a service account JSON; omit to use Application Default Credentials).

# Vertex AI config
config = VideoGenerationConfig(
    provider="google",
    model="veo-3.0-generate-preview",
    api_key=None,
    provider_config={
        "gcp_project": "my-gcp-project",
        "location": "us-central1",               # Optional, default us-central1
        "credentials_path": "/path/to/key.json", # Optional, uses ADC if omitted
    },
    timeout=600,
)

Image count rules (Veo3):

Scenario	Allowed
Reference images only (asset/style)	Up to 3
First frame only	1
First + last frame (interpolation)	1 + 1
Reference images + first_frame	❌ Not allowed
Any + last_frame without first_frame	❌ Requires first_frame

No SDK caching: Google clients are created fresh per request. The google-genai SDK manages its own connection pooling internally.

Person generation policy: Veo3 has strict content policies for generating people. If your request triggers a policy violation, a ContentModerationError is raised with details in raw_response.