XNews

Video model release and storyboard-to-video workflow

Video model release and storyboard-to-video workflow

KeLing releases version 2.6 video model with integrated audio-visual output. - KeLing 2.6 supports joint audio and video generation. - Audio capabilities reported comparable to Veo 3 for dialogue, lip-sync, singing, rap, and basic ambient sounds. - Storyboard-to-video workflow: convert a storyboard grid into a continuous video with smooth transitions. - Storyboard guidance: 3x3 grid (or 4x3 if more); sequence left-to-right, top-to-bottom maps to shot order. - Storyboard rules: analyze subjects, keep strict consistency of characters/costume/environment/lighting. - Storyboard must not add unseen elements or text; maintain depth changes across shots. - Video prompts recommended to be concise: each cell is one shot, preserve cinematic tone and pure frames.

Key insights

KeLing 2.6 adds synchronized audio and video generation

KeLing announced a 2.6 video model that generates audio and video together. The release states audio capabilities have been brought close to Veo 3 levels, including normal dialogue, lip-synchronized speech, singing, rap, and basic environmental sounds. The announcement frames the model as capable of handling both visual and audio aspects of video output, expanding multimodal generation beyond visuals alone.

Storyboard-to-video workflow specified as ordered-shot conversion

Authors provide a concise recipe to convert a storyboard grid into a continuous video: treat each storyboard cell as an individual shot, order cells left-to-right top-to-bottom, and ensure smooth transitions between shots. Prompts should preserve cinematic consistency (same characters, costumes, environment, and lighting) and avoid introducing elements not present in the reference. The guidance emphasizes maintaining pure frames without text and achieving a filmic tone.

Detailed storyboard role prompt enforces strict cinematic consistency

A role-based storyboard prompt instructs a model-as-film-storyboarder to expand a reference image into a 3x3 grid of sequential cinematic shots. Rules require analysis of all subjects, strict consistency of character/attire/environment/light across shots, prohibition on inventing new elements, and depth-of-field variation per shot type. The output requirement is a clean 3x3 storyboard grid image with no text or labels, each panel a pure film frame adhering to a unified color grade.

Source posts