AI-Assisted Beat Sheets and Cold Opens: How to Test Multiple Hooks Before Committing to a Script
The cold open is the highest-stakes 15 seconds in vertical drama production. Everything downstream of it, episode completion rate, next-episode continuation, paywall conversion, depends on whether the first frame stops the scroll and creates a specific question in the viewer's mind.
Most productions commit to a single cold open before any testing occurs. The writer produces the cold open, the production company reviews it, they agree it is strong, and they commission the rest of the scripts. The cold open that goes into production is the first serious attempt at the opening, reviewed by people who already know the premise and cannot evaluate whether a stranger encountering the series for the first time would be stopped by it.
This is the wrong sequence. AI can be used to pressure-test the premise and generate options before any final creative decision is made. For vertical drama cold opens specifically, AI's ability to generate multiple alternative versions of the same structural requirement quickly makes a specific pre-production testing workflow possible that conventional script development cannot afford in time or cost.
The workflow: generate five to ten alternative cold open approaches for the same premise using AI scripting tools, evaluate each against the hook detonation standard, select the strongest two for human writer development, and produce the production-ready version from the developed alternative. This workflow does not replace the human writer. It compresses the discovery process that the human writer would otherwise conduct across multiple drafts into a structured parallel generation session that produces the options before any draft investment is made.
This is the complete guide to that workflow.
The Problem With First-Draft Cold Opens
Most vertical drama scripts fail before a single frame is shot. Not because the story is bad, but because the writer approached it like a TV pilot. The same failure mode applies to cold opens specifically: the writer approaches the cold open with television instincts, building toward conflict rather than opening in it.
The first-draft cold open problem has three specific manifestations:
Setup before conflict. The writer's television instinct is to orient the viewer before the conflict begins. The character walks into a room, looks around, sits down. Something happens that initiates conflict. By the time the conflict arrives, the viewer has been watching setup for 20 seconds and has already made the decision to continue or swipe. The swipe decision was made during the setup, before the conflict that the writer considers the episode's hook even started.
Stated emotion rather than physical behavior. The writer describes the character's emotional state through dialogue or internal monologue before the physical behavior that communicates that state is established. The character says she is angry before her physical behavior shows suppressed anger. In a 9:16 close-up, the physical behavior is the hook. The stated emotion is redundant.
Premise explanation before premise detonation. The writer includes context that establishes why the conflict matters before the conflict itself is visible. The viewer needs to know why this conflict matters before they have any reason to care about it, which is backwards: the conflict creates the caring, not the context.
These are craft failures, but they are craft failures that even experienced writers make on first drafts of vertical drama cold opens because the format's requirements are counterintuitive relative to conventional television instincts. The AI generation process, when correctly prompted, does not have these instincts. It follows the specified structural requirements without the television craft muscle memory that produces the setup-before-conflict failure.
The AI Tool Stack for Cold Open Generation
The cold open generation workflow uses two categories of AI tools in sequence: AI scripting and story development tools for beat sheet and cold open alternatives, and AI video generation tools for visual hook previsualization where the production budget justifies it.
AI Scripting Tools for Beat Sheet and Cold Open Alternatives
The AI scripting tools most applicable to vertical drama cold open generation:
Claude or ChatGPT for structured alternative generation. General AI assistants with strong instruction-following capability generate alternative approaches to a specified structural requirement when the prompt is precise enough to constrain the output correctly. For cold open generation, the prompt specifies the episode's structural requirements, the character configuration, and the cold open's specific commercial requirement: conflict in the first frame, genre readable without audio, specific question created in the viewer's mind.
Storyflow for beat sheet generation alongside character and world context. Storyflow earns the second slot for beat sheets that need to live beside characters, research, and the outline, with AI that reads the whole board. For vertical drama productions that maintain their character profiles, arc map, and world-building documentation in Storyflow's canvas environment, the AI reads all of that context when generating beat sheet alternatives. The generated alternatives are contextually informed rather than generated from the cold open prompt alone.
Morphic for visual hook previsualization. Open on the conflict, not the setup. Describe the first beat as an action mid-motion, framed tight in 9:16. Tell the AI the cold open lands in the first three seconds so the cut keeps a scrolling viewer from leaving. Morphic generates rough visual previsualization of 9:16 cold opens from script descriptions, which allows the production team to see a rough rendering of each cold open alternative before selecting the one for human writer development.
Vertical Haus AI workflow tools for trope maps and cliffhanger scoring. AI-assisted scripting tools now offer trope maps, beat grids, confession variants, cliffhanger scoring and episode continuity checks. The cliffhanger scoring function is specifically applicable to cold open evaluation: it applies a structured assessment of whether each generated cold open variant meets the format's specific hook requirements.
The Cold Open Generation Prompt Framework
The quality of AI-generated cold open alternatives depends almost entirely on the prompt's specificity. A vague prompt generates generic cold opens that do not serve the specific premise's character configuration and power dynamic. A precise prompt generates alternatives that are specifically calibrated to the premise's commercial requirements.
The cold open generation prompt framework for vertical drama:
Layer 1: Series premise in two sentences. The premise description gives the AI the character configuration, power dynamic, and genre register from which the cold open must emerge. More than two sentences over-constrains the generation toward the premise description's framing and reduces the diversity of alternatives produced.
Layer 2: The cold open's structural requirements. Explicit specification of the vertical drama cold open's commercial requirements: the conflict is present in the first frame before any dialogue, the genre and power dynamic are readable without audio in the first three seconds, the cold open creates a specific question in the viewer's mind that the episode answers, and no setup or orientation precedes the conflict.
Layer 3: The cold open's position in the arc. Specify whether this is episode one, a mid-arc episode, or the paywall episode. The arc position affects what conflict is appropriate: episode one establishes the premise's central power dynamic for the first time, mid-arc episodes escalate within an established situation, and the paywall episode cuts at the arc's first genuine tension peak.
Layer 4: The constraint set. Explicit prohibitions that prevent the AI from producing the failure modes described above: no setup before conflict, no stated emotion without physical behavior, no exposition through dialogue, no establishing shots.
Layer 5: The output format. Request five to ten alternatives in a specific format: a three-to-five sentence description of what the viewer sees in the first 15 seconds, the specific question the cold open creates in the viewer's mind, and the physical behavior or action that establishes the conflict without audio.
The complete prompt for a CEO romance series episode one cold open generation session:
Premise: A new employee at a pharmaceutical corporation discovers that the controlling CEO whose company acquired her former employer is the man who broke her heart three years ago. She hid her identity to get the job. He does not know she is there.
Generate seven alternative cold open approaches for episode one of this series. Each cold open must: open in conflict in the first frame before any dialogue; be readable as a status power dynamic without audio in the first three seconds; create a specific unanswered question in the viewer's mind about what will happen next; use physical behavior rather than stated emotion to communicate the character's emotional state; and contain no setup, orientation, or exposition before the conflict.
Do not use: establishing shots, characters walking toward the scene, dialogue that explains the situation, internal monologue, flashback, or any framing device that delays the conflict.
For each alternative, provide: a three-to-five sentence description of what the viewer sees in the first 15 seconds; the specific question the cold open creates in the viewer's mind; and what physical behavior communicates the protagonist's emotional state without audio.
This prompt generates seven alternatives in a single AI session. The session takes approximately three minutes. A human writer generating seven serious alternative cold open approaches for the same premise would typically need two to four hours of deliberate creative work.
Evaluating the Generated Alternatives
The seven to ten AI-generated alternatives are not all usable. Some will violate the constraint set despite explicit instruction. Some will produce the correct structural requirement but with generic rather than specific physical behavior. Some will produce genuinely strong alternatives that a human writer working from the same prompt alone might not have generated.
The evaluation framework for AI-generated cold open alternatives applies three questions to each alternative:
Question 1: Is the conflict in the first frame?
Read the first sentence of the alternative's description. Is there a conflict present, or is the character arriving at, approaching, or preparing for a conflict? The distinction is precise. A character who is already in a confrontation when the episode starts has conflict in the first frame. A character who is entering a room where a confrontation is about to happen has not yet arrived at the conflict. Alternatives that describe arrival rather than presence fail this question and are eliminated.
Question 2: What is the specific question the cold open creates?
The alternative provides this directly in the second component of the output format. Evaluate whether the question is specific enough to motivate continuation. A question like "will she be discovered?" is specific. "What will happen next?" is not. The question that motivates the viewer to watch the next episode is the question that no other information source can answer. The viewer cannot Google the answer. They can only continue watching.
Question 3: Is the physical behavior specific or generic?
Generic physical behavior: she looks nervous. Specific physical behavior: her hand grips the edge of the desk hard enough that her knuckle goes white. The first describes an emotional state. The second shows a physical expression of that state that is readable in a 9:16 close-up at arm's length. Alternatives that provide generic emotional description rather than specific physical behavior fail this question.
From seven generated alternatives, a strong evaluation session typically identifies two to three that pass all three questions. These are the alternatives that proceed to human writer development.
The Human Writer Development Stage
The AI-generated alternatives that pass the evaluation framework are not scripts. They are structured descriptions of what the viewer sees in the first 15 seconds. The human writer's job in the development stage is to take the strongest alternative and develop it into the production-ready cold open.
The human writer's development adds three elements that the AI generation cannot produce reliably:
Character voice in dialogue. The cold open description identifies the conflict and the physical behavior. The human writer writes the specific dialogue that serves the conflict without explaining it. The dialogue has to be in the character's specific voice, with the specific register and vocabulary that the character profile establishes. AI-generated dialogue often becomes too direct, stating the conflict rather than embodying it. The human writer revises for subtext, rhythm, and character voice.
Specific physical action sequencing. The cold open description identifies the physical behavior. The human writer sequences the specific actions within the 15-second window: which action happens at second zero, which at second five, which at second twelve. The sequencing determines whether the cold open's escalation builds correctly within the 15-second frame.
The button-to-cold-open bridge. The cold open of episode one does not precede a prior episode. Episodes two through seventy do. The human writer ensures that the cold open connects correctly to the button cut of the preceding episode. A cold open that does not logically follow from the prior episode's button creates a continuity failure that the AI generation does not account for.
The human writer development stage takes two to three hours per alternative rather than the two to four hours it would take the same writer to generate the cold open from scratch. The time saving comes from the AI having already established the structural concept and the physical behavior framework. The writer is developing and refining rather than discovering.
The Beat Sheet Testing Workflow
The cold open testing workflow scales to the full episode beat sheet. The same principle applies: generate multiple alternative beat sheet approaches before committing any writer time to a specific beat structure.
The beat sheet generation workflow for vertical drama:
Specify the arc position. The beat sheet for episode twenty-five in the dark middle serves different structural requirements from the beat sheet for episode ten at the paywall or episode forty at the midpoint reversal. The AI needs to know the episode's structural position in the arc to generate relevant beat sheet alternatives.
Specify the tension axes active at this position. Which of the series' three to four tension axes is primary in this episode? Which are secondary? An episode where the love interest axis is primary and the antagonist axis is secondary has a different beat structure from an episode where the antagonist axis is primary and the love interest axis provides secondary tension.
Specify the one forward move. The vertical drama escalation section contains exactly one forward move. Specify what that forward move is before generating the beat sheet. The beat sheet alternatives explore different ways to achieve that forward move rather than generating the forward move itself. The forward move is specified in the arc map. The beat sheet is the implementation of that specified move.
Generate five alternative beat sheet structures. The five alternatives explore different ways to structure the episode's hook, escalation, spike, and button around the specified forward move. Some alternatives front-load the spike. Some build slowly through the escalation before the spike. Some use the hook itself as a mini-spike. The range of alternatives tests which structural approach serves the specific forward move most effectively.
Evaluate against the episode's commercial requirement. The paywall episode's beat sheet must produce a button at maximum unresolved tension. A mid-arc episode's beat sheet must produce a button that creates discomfort strong enough to motivate return after a 60-minute rewarded ad cooldown. Different episode positions have different button requirements. The beat sheet evaluation confirms that the generated alternative's button serves the specific commercial requirement of the episode's position.
The Trope Map as a Pre-Generation Tool
Before any cold open or beat sheet generation, the trope map is the pre-generation tool that ensures the generated alternatives are drawing from the genre's commercially validated narrative inventory rather than from generic dramatic convention.
AI-assisted scripting now offers trope maps as a pre-production tool. The trope map for a CEO romance series catalogues the specific narrative elements that the genre's highest-performing titles use: the forced proximity situation, the concealed identity, the controlled exterior that cracks involuntarily, the public humiliation that sets up the vindication arc, the antagonist's specific injustice type.
The trope map's commercial function is not to encourage copying. It is to ensure that the AI-generated alternatives are drawing from the emotional vocabulary the target audience has already demonstrated willingness to pay for, rather than generating emotionally novel alternatives that are less commercially tested.
The trope map feeds the generation prompt as context: after the premise description, include the three to five trope elements that are active in this episode. The AI generates cold open alternatives that incorporate those trope elements rather than generating cold opens that replace them with unfamiliar narrative devices.
A viewer who encounters a CEO romance cold open that uses the familiar vocabulary of the concealed identity genre has her investment primed by prior engagement with the genre's established emotional architecture. A cold open that uses unfamiliar narrative devices requires more viewer investment to establish the same emotional engagement. For a format where the scroll stop decision is made in under three seconds, the familiar vocabulary has a measurable advantage.
Cold Open Scoring: The Systematic Evaluation Tool
After generating and evaluating alternatives against the three-question framework, a systematic scoring system produces a rank-ordered selection for human writer development.
The cold open scoring system assigns points across five criteria, each weighted by its commercial importance:
Conflict in the first frame (0 or 3 points). Binary pass/fail. Either the conflict is present in the first frame or it is not. Three points for presence, zero for absence. No partial credit.
Specific physical behavior communicating emotional state (0 to 2 points). One point for physical behavior that is present but generic. Two points for physical behavior that is specific, cinematically precise, and readable in a 9:16 close-up without audio.
Specific question created in the viewer's mind (0 to 2 points). One point for a question that is general. Two points for a question that is specific, unanswerable without continuing, and directly generated by the cold open's action rather than by background premise knowledge.
Audio independence (0 or 2 points). Binary pass/fail. Does the conflict and emotional register communicate without audio? Two points for audio independence, zero if the cold open requires audio to be understood.
Genre signal in the first frame (0 or 1 point). One point if the visual register of the cold open communicates the genre immediately without audio or prior context.
Maximum score: 10 points. The cold open that scores 9 or 10 is a strong candidate for human writer development. The cold open that scores 7 or 8 is a conditional candidate. Below 7, the cold open fails at least two commercial requirements and does not proceed.
The Iteration Cycle: When to Generate Again
The first generation session produces five to ten alternatives. From those, two to three typically score above 7 and proceed to human writer development. One of those two to three becomes the production-ready cold open after development.
When none of the first generation session's alternatives score above 7, a second generation session runs with a revised prompt rather than with the same prompt repeated. Repeating the same prompt produces similar alternatives with similar failure modes. The revised prompt changes one of the input parameters to shift the generation's output:
Parameter change 1: Premise reframing. If the first session produced alternatives that failed the conflict-in-first-frame criterion consistently, the premise description may be framing the story as setup rather than as conflict. Reframe the premise as a conflict statement rather than as a story description.
Parameter change 2: Physical behavior constraint tightening. If the first session produced alternatives with generic emotional description rather than specific physical behavior, add an explicit example of the distinction in the constraint set: do not write she looks nervous, write what she physically does with her hands, her eyes, her posture, or her breath.
Parameter change 3: Genre vocabulary specification. If the first session produced alternatives that do not draw from the genre's trope vocabulary, add the trope map specification to the prompt and re-run.
The second generation session with a revised prompt typically produces at least two alternatives that score above 7. If neither generation session produces a strong alternative, the premise itself may need reconsideration before the cold open problem can be solved. A premise that cannot generate a strong cold open alternative in two generation sessions is a premise that does not have a natural conflict-in-first-frame entry point.
Axis AI Studios Perspective
The AI-assisted cold open testing workflow is the pre-production investment that most directly improves the format's most commercially critical element at the lowest possible cost. A production company that commissions a writer to produce multiple cold open alternatives and then selects the strongest for development is doing the right thing. The AI generation session is the cheaper way to produce those alternatives, which allows the writer's time to be spent on the development and refinement stage rather than on the discovery stage.
The cold open is too commercially important to commit to on the basis of a single writer's first serious attempt. The hook rate that the cold open determines is the algorithmic foundation of the series' entire distribution trajectory. A cold open that scores below 40% hook rate produces an algorithmically suppressed series regardless of the quality of episodes two through seventy.
The generation workflow described in this post takes approximately three hours total: twenty minutes of prompt development, three minutes of generation, two hours of evaluation and scoring. It produces five to ten tested alternatives against which the human writer develops the production-ready version. The three hours is the cheapest possible risk management investment for the series' highest-consequence production decision.
For production companies who want to commission vertical drama from a partner that applies systematic cold open testing before any script is committed to production, reach out at business@axisaistudios.com.
FAQ
Does AI-Assisted Cold Open Generation Reduce the Writer's Creative Contribution?
No. The AI generation produces structured descriptions of what the viewer sees in 15 seconds. The writer's contribution is everything that makes those 15 seconds cinematically specific, emotionally precise, and in the specific voice of the characters. The AI identifies the structural concept. The writer produces the art. The generation workflow redistributes the writer's time from discovering structural concepts, which AI can do adequately at speed, toward developing and refining the strongest structural concept into a production-ready cold open, which only the writer can do well.
What If All Generated Alternatives Are Structurally Similar to Each Other?
This indicates that the prompt's constraint set is too narrow or that the premise description is framing the story in a way that limits the generative space. Two interventions: first, add an explicit instruction to generate alternatives that approach the opening from different emotional registers, different characters' perspectives, or different moments within the first hour of the story's events. Second, revise the premise description to describe the story's central tension rather than its situation. A tension description generates more diverse alternative approaches than a situation description because tension has inherent directionality that situation does not.
How Many Generation Sessions Should a Production Company Run Before Committing to a Cold Open?
One well-prompted generation session producing five to ten alternatives is sufficient if two or three alternatives score above 7 on the scoring system. A second session is warranted when the first session produces fewer than two alternatives above 7. Beyond two generation sessions without a strong result, the problem is more likely the premise's cold open entry point than the prompt quality, and the production company should evaluate whether the premise requires restructuring before the cold open problem can be solved.
Further Reading
For the full writers' room structure that the cold open testing workflow described in this post feeds into as its first quality gate, the guide to how to build a vertical drama writers' room covers the cold open specialist role, the batch outline production process, and the quality control system that catches structural problems before scripting time is invested.
For the writer brief that communicates the cold open detonation standard to human writers before any script is commissioned, the guide to how to brief a writer for vertical drama covers the hook detonation requirement, the timestamp skeleton format, and the dialogue constraints that the cold open must serve.
For the audience testing framework that validates the production-ready cold open against real viewer behavior before full production is committed, the guide to audience testing before you commit covers in-app survey mechanics, cliffhanger variant testing, and the go/stop decision framework.
Let's set
the new standard together.
If you're working on something, we'd like to hear about it.