What is Storyboard script? Definition, key points and related tools
Per-shot text a video model executes: framing and camera, who is in frame, action written as cause not result, dialogue with speaker, lighting as source direction and colour temperature, seconds. In AI drama the storyboard text is the prompt.
Why the storyboard is where changes are still free
On a film set a storyboard is a guide that a director, a camera operator and actors interpret. In AI drama nobody interprets it: each shot line reaches the model as written, mistakes included. Two things follow. The storyboard is the last stage where a change costs nothing, since editing a line is free while regenerating 15 seconds at MiniMax H3's list price costs $1.20–$1.95 (MiniMax). And the storyboard is where continuity lives: models keep no memory between calls, so each segment's opening state has to restate the previous segment's end state in words.
The field-by-field template is in the storyboard template guide, and the wording rules (causes, not results) are in writing prompts for video models. This page covers what those leave out: formatting one storyboard for several models, and checking it before paying for video.
One storyboard, several model dialects
Vendors document different input formats, so a storyboard is written once and rendered per model at submission:
| Model | How shots are expressed | Dialogue and sound | Length and rewriting |
|---|---|---|---|
| Kling VIDEO 3.0 | Multi-Shot plans cuts itself; Custom Multi-Shot takes up to 6 shots, each with its own duration (Kling) | Native audio in Chinese, English, Japanese, Korean and Spanish (Kling) | 3–15 s per generation |
| Sora 2 | Separate shot blocks, each with one camera move and one subject action (OpenAI) | A dialogue block below the prose; a 4-second shot fits one or two short exchanges | 4, 8 or 12 s per clip (API) |
| Veo 3.1 | Timestamped shots inside one clip, e.g. four 2-second shots in 8 seconds (Google prompting guide); longer scenes by 7-second extensions (docs) | Speech in quotes; sounds described explicitly | 4, 6 or 8 s per clip |
| MiniMax H3 | Multi-shot text in one prompt (model page) | Up to 3 reference audio clips, 15 s in total | 7,000 characters per text item (MiniMax) |
| Wan 3.0 | Multi-shot storyboard with timestamps, 4–6 s per shot (Alibaba guide) | Native audio | 20,000 characters, excess truncated; prompt rewriting on by default (Alibaba Cloud) |
Keep the master in a model-agnostic structure (segment header, numbered shots, speaker-tagged lines, seconds) and convert it per engine. Turn prompt rewriting off where it exists: a rewriter merges shots and changes seconds (prompt extend).
Seven checks before you pay for video
- Shot seconds add up to the segment, and the segment length is one the model accepts (segments and shots).
- Everyone named in a shot is visible in it. A character named as leaving or off-screen tends to be drawn.
- Every line has a speaker and fits its seconds, about 2.5 English words per second (dialogue timing).
- Each action ends in a stated position that the next shot starts from.
- Light is written as a source (where it is, what colour temperature), not as a spot on a face.
- The text fits the model's limit and will reach it unrewritten.
- The episode's last shot holds the cliffhanger and ends on a closing beat rather than spare seconds.
Common mistakes, and where storyboard tools differ
Mistakes. One paragraph for fifteen seconds leaves the model to choose its own cuts, or none. Mood words ("tense", "cinematic") take the place of instructions the model can execute. Novel prose copied into shots ("she remembers her father") gives the model nothing visible. And a storyboard written in one model's dialect, sent unchanged to another, loses its shot structure.
Tools. SceneMixer drafts a storyboard per episode, editable per shot and versioned, so you review its choices rather than compose from scratch. LTX Studio breaks a script into scenes and shots with a shot-level editor inside one production. Sora's Storyboard tool and Flow's scene builder arrange clips in sequence; Morph Studio's Open Canvas and PixVerse's Canvas lay shots out by hand across several models; Kling's Custom Multi-Shot sets per-shot durations inside one generation. Runway and Pika generate from single prompts without a storyboard layer. The choice is between storyboards drafted for you across many episodes and storyboards composed by hand, shot by shot, for one production.
Key figures and sources
| Figure | Value | Source | Date |
|---|---|---|---|
| Kling VIDEO 3.0 multi-shot | Up to 6 shots per generation; 3–15 s total | kling.ai | 2026-09-12 |
| MiniMax H3 prompt limit | 7,000 characters per text item | platform.minimax.io | 2026-09-12 |
| Wan 3.0 prompt handling | Up to 20,000 characters (excess truncated); prompt_extend on by default | www.alibabacloud.com | 2026-09-12 |
| Sora 2 prompting guide | One camera move and one subject action per shot; a 4-second shot fits one or two short exchanges | developers.openai.com | 2026-09-12 |
| Veo 3.1 clip lengths | 4, 6 or 8 s; 8 s required with reference images, extension, 1080p or 4K | ai.google.dev | 2026-09-12 |
FAQ
Is a storyboard script the same as a prompt?
In AI drama, yes, with structure added: each shot line is what the model executes. The difference from a one-off prompt is that a storyboard carries continuity (who stands where, what just happened) from segment to segment, which a single prompt never has to do.
How detailed should each shot be?
Detailed about causes and state, brief about everything else: framing, who is visible, the action and its end position, the line with its speaker, and the light source. Mood adjectives add length without adding anything the model can act on, and long text risks truncation (Wan 3.0 truncates past 20,000 characters; MiniMax H3 allows 7,000 per text item).
Can an LLM write the storyboard?
Yes, and pipeline tools do that per episode. Review is still needed: an LLM drafts plausible shots, and a pass against the seven checks above is what catches a named off-screen character or a line with too few seconds.
Sources
- https://platform.minimax.io/docs/guides/pricing-paygo
- https://platform.minimax.io/docs/api-reference/video-generation-v2-create
- https://kling.ai/blog/kling-video-3-multi-shot-guide
- https://kling.ai/quickstart/klingai-video-3-model-user-guide
- https://developers.openai.com/cookbook/examples/sora/sora2_prompting_guide
- https://developers.openai.com/api/docs/api-reference/videos/create
- https://ai.google.dev/gemini-api/docs/veo
- https://www.alibabacloud.com/help/en/model-studio/wan3-video-generation-api-reference
- https://cloud.google.com/blog/products/ai-machine-learning/ultimate-prompting-guide-for-veo-3-1
- https://www.alibabacloud.com/help/en/model-studio/wan3-video-generation-guide