Runway¶

Runway provides video generation (Gen-4, Gen-3, Aleph) via the runwayml Python SDK. It supports text-to-video, image-to-video, and video-to-video editing.

Installation¶

pip install tarash-gateway[runway]

Quick Example¶

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

config = VideoGenerationConfig(
    provider="runway",
    model="gen4_turbo",
    api_key="YOUR_RUNWAY_KEY",
)

request = VideoGenerationRequest(
    prompt="A lone wolf running across a snow-covered tundra at dawn",
    aspect_ratio="16:9",
)

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

Parameters¶

Parameter	Required	Supported	Notes
`prompt`	✅	✅	Text description of the video
`aspect_ratio`	—	✅	Converted to Runway pixel dimensions
`image_list`	—	✅	Used as input image for image-to-video
`video`	—	✅	Required for Aleph (video-to-video)
`duration_seconds`	—	—	Duration is model-dependent, not user-configurable
`seed`	—	—	Not supported
`negative_prompt`	—	—	Not supported
`generate_audio`	—	—	Not supported

Content moderation¶

Pass Runway's content moderation options via extra_params:

request = VideoGenerationRequest(
    prompt="...",
    extra_params={
        "content_moderation": {"publicFigures": "disallow"},
    },
)

Supported Models¶

Tarash automatically selects the correct Runway endpoint (text_to_video, image_to_video, or video_to_video) based on the model name and the inputs provided.

Model	Notes
`gen4_turbo`	Gen-4 Turbo; auto-selected based on input (text or image)
`gen4_aleph`	Gen-4 Aleph; requires video input; editing/extending
`veo3.1`	Veo 3.1 via Runway; auto-selected based on input
`veo3.1_fast`	Veo 3.1 Fast via Runway; auto-selected based on input
`gen3a_turbo`	Gen-3 Turbo; requires image input
`gen-3-alpha`	Gen-3 Alpha; auto-selected based on input (text or image)
`aleph`	Gen-3 Aleph; requires video input; editing/extending

Aspect ratio support: Runway uses pixel dimensions instead of ratio strings. Tarash converts automatically.

`aspect_ratio`	Runway Size	Modes
`16:9`	`1280:720`	Text, Image
`9:16`	`720:1280`	Text, Image
`16:9-wide`	`1080:1920`	Text only
`9:16-wide`	`1920:1080`	Text only
`4:3`	`1104:832`	Image only
`3:4`	`832:1104`	Image only
`1:1`	`960:960`	Image only
`21:9`	`1584:672`	Image only

Image-to-Video¶

Use gen3a_turbo with a reference image to generate video from a starting frame.

from tarash.tarash_gateway.models import ImageType

config = VideoGenerationConfig(
    provider="runway",
    model="gen3a_turbo",
    api_key="YOUR_RUNWAY_KEY",
)

request = VideoGenerationRequest(
    prompt="The astronaut floats gently away from the ship",
    aspect_ratio="16:9",
    image_list=[
        ImageType(image="https://example.com/astronaut.jpg", type="reference"),
    ],
)

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

Video-to-Video¶

Use the aleph model to edit or transform an existing video clip.

from tarash.tarash_gateway.models import MediaContent

config = VideoGenerationConfig(provider="runway", model="aleph", api_key="YOUR_RUNWAY_KEY")

request = VideoGenerationRequest(
    prompt="Make the scene look like it's at night",
    video="https://example.com/original-clip.mp4",   # or MediaContent bytes
)

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

Provider-Specific Notes¶

Endpoint auto-selection: Aleph models → video_to_video. All others → image_to_video if an image is provided, otherwise text_to_video. gen3a_turbo always requires an image.
Task polling: Tarash polls until the task reaches SUCCEEDED, FAILED, or CANCELLED.
Video download: The response includes video bytes directly, not just a URL.