Gen‑4 models lock characters
People working with Runway and Gen‑4 workflows report new model releases that better ‘lock’ or preserve characters across clips, and Seedance 2.0 now accepts multiple image, video, and audio references for richer generative video. (x.com) (x.com)
A new crop of artificial intelligence video tools is moving from “make a clip” to “keep the same person on screen,” with Runway’s Gen-4 line and ByteDance’s Seedance 2.0 both pushing harder on consistency. (runwayml.com) (seed.bytedance.com) Runway said Gen-4 can generate “consistent characters” from a single reference image, keeping the same subject recognizable across different lighting, locations, and shot treatments. Runway later positioned Gen-4.5 as a higher-fidelity follow-on model focused on more control over video generation. (runwayml.com 1) (runwayml.com 2) ByteDance launched Seedance 2.0 on February 12, 2026, and said the model accepts four input types: text, image, audio, and video. Its product page says creators can combine those references to steer performance, lighting, shadows, and camera movement in a generated clip. (seed.bytedance.com 1) (seed.bytedance.com 2) In plain terms, these systems work like a director using reference boards: one image defines who a character is, while other clips or sounds define how a scene should move or feel. The technical problem is that older video models often let faces, clothes, or proportions drift from shot to shot. (runwayml.com) (seed.bytedance.com) That drift has been one of the biggest obstacles to using generated video for ads, short films, and branded characters, because every new shot can look like a recast. Runway’s March 9, 2026 launch of Runway Characters showed the company was already building products around persistent on-screen identities, not just one-off generations. (runwayml.com) Runway’s own changelog shows how quickly the market is converging around this workflow. On April 7, 2026, Runway added Seedance 2.0 to paid plans outside the United States, letting users generate or edit videos from text, image, video, or audio inside the same platform. (runwayml.com) The competitive pressure is visible in the language both companies use. Runway emphasizes “infinite character consistency” from one image, while ByteDance says Seedance 2.0 offers the industry’s most comprehensive set of multimodal references and editing controls. (runwayml.com) (seed.bytedance.com) Neither claim is an independent benchmark for every production use case, but both point to the same buyer demand: fewer broken identities between shots and more direct control over motion and style. For creators, the practical test is simple — whether a character still looks like the same character after a sequence of edits, camera changes, and new scenes. (runwayml.com) (seed.bytedance.com)
