Multi-shot video

Produce a multi-shot reel (a single subject's story told across several cuts) in one video call, instead of one continuous take or hand-stitching clips in an editor. The lever is the model's own shot template written into the prompt, with the cuts, the per-shot timing, and the cross-cut continuity all carried by the text (and, on some models, an anchor frame).

Inputs to collect

• The story beats. Two to six shots: what each shot shows, in order (wide establishing, close on hands, end on the finished thing).
• Total length and roughly how many cuts. This sets the per-shot time budget.
• The subject's identifying details (wardrobe, color, key features) that must stay the same across every cut.
• Optional: an anchor frame (one or two stills) to lock the opening and/or closing look exactly. Ask only if brand assets or an exact character must survive frame-for-frame.
• Optional: aspect/orientation and whether you want the native ambient audio bed.

Models

• Default storyboard reel: Alibaba HappyHorse 1.1 (alibaba:happyhorse@1.1). Natural-language storyboard cues (Begin with… / Cut to… / Shift to… / End on…), up to 15 seconds, native audio crossing the cuts. Best when the brief reads like a script.
• Per-shot timing control: Kling 3.0 Turbo (klingai:kling-video@3.0-turbo). Inline shot N, seconds, description; template, up to 6 shots, per-shot seconds you budget yourself. Optional first-frame anchor via inputs.frameImages.
• Anchor-frame multi-clip: PixVerse V6 (AIR pixverse:1@8, slug pixverse-v6). One or two anchor frames pin the bracketing look, settings.multiClip: true plus a beat prompt invents the cuts between them. Best when you have stills that must match exactly at the open/close.
• Confirm each is live and inspect its schema via runware-models + runware-run before calling. Do not hardcode a stale choice. All three are videoInference with io:text-to-video and io:image-to-video.