Sora is one route for text-led and multimodal visual creation. This lab answers a platform-neutral question: for *this* story, does written scene direction produce a better first draft than image/reference-led generation? Use the current product surface available to you at lab time; access, interface, and capabilities change.

Prerequisites

• One written scene beat (3–5 sentences): place, subject, action, mood, one camera move.
• Optionally one approved image/reference where the current product supports it.
• A 59.A card with the shared brief.

Workshop path

1. Turn the beat into concise visual direction. Motion anatomy: subject + action + environment + camera + duration/pace + constraints. One move, one action, no compound acrobatics. 2. Generate short alternatives. Save each direction verbatim with settings, time, and cost. Vary one variable at a time. 3. Select by narrative clarity and visual coherence. Does a stranger grasp the beat? Does the world hold together for the clip's full length? 4. Use approved reference material where supported and note where it is not — that gap is itself a finding. 5. Export the winner and cut it into the shared 15–30 second story (hook → proof → CTA per 59.5). Hand-correct captions; phone-check muted and with sound. 6. Write the comparison. Same beat, one reference-led environment (Higgsfield, Seedance, Kling, Flow, or Runway): which gave the better first draft for a mood scene? A product demo? A creator skit? A source-backed explainer? The answer may differ per job — record it per job.

Verify fast

Read the direction file without watching the clip, then watch with sound off. Did the words predict the picture? Log what written direction did well (atmosphere, composition ideas) and where references would have saved retries (identity, product, continuity). Note recheck date; Sora access and features change.

Worked mini-example + per-job verdict table

Take a mood beat first — "dawn light across empty bakery shelves, slow push-in, flour dust in sunbeam, calm, 5 seconds" — where written direction often shines: atmosphere and camera intent carry the shot with no identity to preserve. Then take a proof beat — the scored loaf with its exact label — where reference-led generation usually wins on fidelity. Fill this table from your own runs:

| Beat type | Written-direction draft (score 1–5, evidence) | Reference-led draft (score 1–5, evidence) | Winner for this beat + why |

|---|---|---|---|

| Mood / atmosphere | | | |

| Product demo | | | |

| Creator skit | | | |

| Source explainer | | | |

Save the direction file, all alternatives with settings/cost, the winner, the edited story placement, and this table. The lab succeeds when you can route the next four beats without guessing — not when one tool wins everything.

Writing direction that actually directs

Strong written direction is specific about the visible: one subject with its distinguishing features, one action with its pace, one environment with its light, one camera with its move and duration — then constraints that forbid the usual failures ("no readable text," "no extra people," "locked camera, no drift"). Weak direction narrates feelings ("an inspiring cinematic journey of bread") and leaves every visual decision to the system. After each alternative, annotate which sentence in your direction produced which visible result; sentences with no visible effect get cut or rewritten. That annotation loop — direction → frame → revision — is the transferable skill. The clip expires; the ability to write a direction that predicts its picture does not. Recheck Sora's supported inputs at lab time and record what the surface accepted.