Runway: Aleph 2.0

Runway: Aleph 2.0 is a video model from Runway, listed on api14 under the model id runway/aleph-2. It sits in the video category and is described by the provider as an in-context video editing model. Rather than generating footage from nothing, it works on existing material, applying edits to footage you already have. The provider release date is 2026-07-29.

The model accepts text, image and video as inputs, and its only output modality is video. That combination is what makes in-context editing possible: you supply the source footage, optionally alongside images, and direct the change with text instructions. According to the provider description, Aleph 2.0 applies text instructions and keyframe-guided edits across existing footage while preserving details that are not meant to change. The model supports a wide range of aspect ratios: 16:9, 4:3, 3:2, 1:1, 2:3, 3:4, 9:16 and 21:9. Audio generation is not supported, so any audio in your pipeline has to be handled separately. No context length is listed for this model.

The practical uses follow from editing rather than generating. If you have shot footage and want to alter specific elements while leaving the rest untouched, this is the kind of model you would reach for. Keyframe-guided edits suit work where a change has to be anchored at particular points in a clip and carried through consistently. Because the model takes images as input as well as video, it can be driven by reference stills alongside the text instruction. The broad aspect ratio support means the same model can handle landscape, square, portrait and wide formats, which helps when one edit has to be delivered to several placements.

You call it through api14's OpenAI-compatible API using the model id runway/aleph-2. Because the interface follows the OpenAI format, existing clients and SDKs that already point at api14 should work without special handling for this model. The supported parameters list for Aleph 2.0 is empty, so there are no documented request parameters to set beyond the model id and your inputs. In practice that means you send your text instruction together with the video, and any images, and rely on the model's defaults rather than tuning generation settings per request.

There are a few things to keep in mind when deciding. The model outputs video only, so it is not a fit if you need text or image output from the same call. Audio generation is off, so narration, dialogue or sound design must come from elsewhere. With no supported parameters listed, you cannot adjust behaviour through request options, which may matter if your workflow depends on fine control. And since it is an editing model, it assumes you already have footage; for work that starts from a blank slate, a text-to-video model would be the more natural choice.