Runway Aleph 2.0

Aleph 2.0 is Runway’s flagship in-context video editing model. Give it a video and a prompt, and it makes the edit you describe while preserving everything you didn’t ask to change. Give it an edited keyframe alongside the prompt and the model carries that exact look through the rest of the clip.

Compared to the original runwayml/gen4-aleph, Aleph 2.0:

• Works with up to 30 seconds of input video
• Applies edits across multiple shots and cuts in the same clip
• Supports up to 5 keyframes with positions (first, last, or an exact timestamp) for image-level precision over the edit

When attributing to Runway, please use “Powered by Runway” and link to runwayml.com from the user interface.

• Runway’s announcement
• Usage policy
• Terms of use
• Privacy policy

What it can do

• Targeted edits. Change a single product, color, hairstyle, or piece of clothing without touching the background, lighting, or other details.
• Environmental changes. Swap locations, weather, seasons, or time of day. Lighting, shadows, and reflections update naturally.
• Object addition and removal. Add or remove props, people, or scene elements with correct perspective and occlusion.
• Style transfer. Apply a new visual style across the whole clip, including across multiple shots.
• Multi-shot consistency. When a clip has cuts, the edit propagates across the relevant shots so you don’t have to edit shot-by-shot.

Keyframes

Aleph 2.0 accepts up to 5 keyframe images. Each keyframe is paired with a position that tells the model where in the video it applies:

• first — start of the video
• last — end of the video
• A timestamp in seconds, like 1.67, for any other point in the clip

Keyframes give you image-level control over the edit: you decide exactly what the change should look like in a specific frame, and Aleph 2.0 carries that look through the rest of the clip.

The keyframe_images and keyframe_positions inputs are parallel lists — one position per image.

Inputs

• prompt: Text description of the edit.
• video: Input video. Must be between 2 and 30 seconds and under 16 MB.
• keyframe_images: Up to 5 optional keyframe images.
• keyframe_positions: Position for each keyframe (first, last, or a timestamp in seconds).
• seed: Optional seed for reproducible generation.

Limitations

• Input video must be between 2 and 30 seconds.
• Maximum input file size is 16 MB.
• Output video matches the input video’s aspect ratio.
• This is an early-access model from Runway. Behavior and pricing may change before general availability.