01
VO 1.6s / 3.0s
Tool 01 · Pre-production system
Lock the character. Time every beat. Ship a prompt the model can follow.
A browser-based control room for 9:16 AI animation: define a stable character identity, map shots to seconds, check English voiceover timing, and export one structured production brief.
Target specification
Vertical master format
Consistency engine
Character lock
Dynamic timeline
Timed visual beats & audio sync
02
VO 2.8s / 4.0s
03
VO 3.2s / 3.0s
Workflow guide
How to get reliable results with AI Video Character Lock & Storyboard Prompt Builder
Turn a loose vertical-video idea into a timed production brief with a repeatable character identity, shot-level direction, voiceover timing, and exportable text or JSON.
What this tool helps you control
- Lock a character name, silhouette, outfit, colors, line weight, and recurring props before writing shots.
- Break a 5, 10, or 15 second vertical video into editable beats with explicit start and end times.
- Pair camera, lighting, physical action, and English dialogue inside every beat.
- Catch voiceover that is too long for its visual window using a 2.5 words-per-second benchmark.
- Export a readable master prompt or a structured JSON specification without sending the idea to a server.
Quick workflow
- 01Choose the total duration and an art direction that matches the generator or animation workflow you plan to use.
- 02Define one character identity in concrete visual terms, then lock any prop that must survive every cut.
- 03Edit each timed beat so the camera, lighting, action, and spoken line describe the same moment.
- 04Resolve amber or red voiceover warnings by shortening dialogue or changing the beat timing.
- 05Review the compiled output, switch between text and JSON, then copy or download the production brief.
Why character consistency starts before generation
AI video systems do not remember a character in the same way a human animator does. A prompt may say “the same host,” yet the next shot can quietly change the face, jacket, body proportions, line thickness, or hand holding a prop. The reliable fix is not a larger paragraph of adjectives. It is a compact identity contract that separates permanent facts from temporary actions. 916Studio calls that contract the character lock. A useful lock names the character, states observable visual traits, and lists any object that cannot drift between shots.
Write traits so another artist could check them frame by frame. “Cool futuristic man” is subjective; “angular silver helmet, cyan horizontal visor, matte charcoal flight suit, orange shoulder patch” is testable. Include shape, color, material, outline treatment, and costume details that matter to recognition. Avoid adding ten decorative details simply because the field has room. Every extra instruction competes for attention inside the generation model, so keep only the identifiers that distinguish the character on a phone screen.
Build a vertical story as timed visual beats
Short-form video works when each second has a job. A ten-second clip usually needs an immediate visual interruption, a clearly readable action, and a payoff that can hold long enough to register. The default Hook, Main Action, and Punchline timeline is a starting point rather than a rigid formula. You can rename beats, change their boundaries, or add a fourth beat for a reaction, product reveal, or call to action. What matters is that the total sequence has no accidental dead time and no two events fighting for the same instant.
Describe action with visible verbs. Instead of writing “the character feels nervous,” write “the character grips the mug with both hands, shoulders rise, and eyes snap toward the warning light.” The second version can be staged. Keep one primary action per beat, especially below fifteen seconds. When a beat includes a camera move, a location change, three gestures, a text reveal, and a prop transformation, generation systems often blend the instructions or omit half of them. Split complex events into consecutive beats and preserve screen direction between them.
Synchronize voiceover with what the viewer can see
Dialogue length is a production constraint, not a cosmetic detail. English conversational delivery is often around 2.5 words per second, though the final pace changes with emotion, pauses, pronunciation, and the performer. The timing indicator uses that rate as an early warning. Green means the line has comfortable space, amber means it may feel tight, and red means the spoken words are likely to outlast the visual beat. Treat the signal as a planning benchmark and confirm the final cut with an actual voice recording.
Shorten a line by removing setup that the image already communicates. If the screen visibly shows a giant countdown, the voiceover does not need to say “a giant countdown appears on screen.” Let the image carry concrete information and use speech for the argument, surprise, or emotional turn. Put the most important noun and verb early because viewers may swipe before the sentence finishes. If a punchline needs a pause, leave time for it; a six-word line with dramatic silence can occupy more space than a ten-word informational line.
Use camera and lighting as continuity controls
Camera choices should clarify the story rather than decorate the prompt. A close-up is useful for recognition, reaction, or a small prop. A wide full-body shot is better for readable movement. A low angle changes perceived power, while a push-in focuses attention at the payoff. Jumping between unrelated lenses and angles can make a very short clip feel like disconnected samples, so create a deliberate progression—for example, close-up hook, wide demonstration, push-in climax. Repeat the character lock in the compiled brief so identity remains the priority during every change.
Lighting descriptions work best when they name direction and contrast as well as mood. “Warm cinematic” gives a broad target; “warm key light from camera left with a cool rim” gives a production relationship. For flat graphic animation, simpler is usually better because complex reflections can produce visual noise and identity drift. The included presets are practical vocabulary shortcuts, not promises of a particular model style. Review them against the generator you use and replace any term that the model interprets unpredictably.
Text prompt or JSON: choose the right handoff
The structured text view is designed for direct reading, editing, and pasting into tools that accept a single prompt. It keeps the format, character lock, continuity rules, and timeline in a predictable order. The JSON view carries the same intent as fields that can be stored in a production document, passed into an automation you control, or compared across revisions. JSON does not make a generation model inherently more accurate; it simply reduces ambiguity for systems or collaborators that understand the field structure.
Save a master version before experimenting. When a new generation fails, change one variable at a time—first the action, then the camera, then style wording—rather than rewriting the entire brief. This gives you evidence about what improved the result. The tool runs in your browser and the download is created locally, so your character notes and dialogue are not submitted to a 916Studio server. The AI generator you paste into may have its own data policy, which you should review separately.
A practical quality-control pass before rendering
Read the compiled prompt from top to bottom as if you were a new animator joining the project. Confirm the duration, then circle the three or four character traits that truly define identity. Check that every prop stays in the same hand unless a transfer is explicitly animated. Confirm that beat times do not overlap unintentionally, that the main action fits the framing, and that the final pose has enough time to hold. Remove instructions that contradict each other, such as “static locked frame” and “fast orbiting camera.”
Finally, preview the concept at phone size. Important gestures should remain readable inside a 9:16 crop, and captions or faces should avoid platform interface zones. Use the Safe Zone Checker for placement and the Script Timer for a second, pace-specific voiceover estimate. Those tools solve different parts of the same workflow: this builder organizes the generation brief, the safe-zone tool protects composition, and the timer protects spoken duration. Together they turn an idea into a testable production plan rather than a single fragile prompt.
| Planning method | Character continuity | Timing control | Best use |
|---|---|---|---|
| One-paragraph prompt | Implicit and easy to lose | Rough duration only | Fast concept tests |
| 916Studio structured text | Explicit identity and prop lock | Beat-by-beat plus VO warning | Direct generator handoff |
| 916Studio JSON | Machine-readable fields | Numeric start/end values | Automation and versioning |
Frequently asked questions
Does 916Studio generate the video?
No. It creates a production-ready prompt and timing specification that you can use with the AI video or animation generator of your choice.
Will a character lock guarantee identical frames?
No prompt can guarantee model output. A character lock reduces ambiguity and makes revisions more consistent, but the result still depends on the generator, model version, reference images, seeds, and settings.
Why is voiceover estimated at 2.5 words per second?
That is a practical conversational benchmark of about 150 words per minute. Actual delivery varies, so record or read the final line aloud before publishing.
Can I use a duration other than 5, 10, or 15 seconds?
The current target presets focus on common short AI video lengths. You can customize beat boundaries within the selected duration and combine exported segments in an editor for longer work.
Is my prompt uploaded to 916Studio?
No. The builder operates in your browser. Copying and downloading are performed by browser APIs on your device.
When should I choose JSON instead of text?
Choose text for reading or direct pasting. Choose JSON when another program, template, or controlled workflow needs named fields and numeric timing values.