Define a common source scene and fidelity rubric

Choose a short authorized scene with a clear turn, dialogue or narration, character state, setting, and at least one continuity dependency. Freeze its version and annotate the facts, chronology, motivation, emotional shift, required action, and material that may be compressed or omitted. Build a shared rubric for source traceability, meaning preservation, visual control, audio planning, continuity, review effort, revision isolation, accessibility, and delivery fit. Keep product availability outside the first comparison. The goal is to test how each editorial pipeline represents decisions, not to rank model output or assume that Seedance 2.5 exposes any particular workflow.

• Authorized source scene
• Meaning and continuity rubric
• Revision and review burden
• Product-neutral first pass

Walk the scene through three distinct methods

In a storyboard-first path, visualize the sequence early and inspect composition and transitions, while watching for missing source links. In a structured shot-packet path, map beats to contracts before imagery, gaining traceability at the cost of more setup. In an edit-first path, plan narration, placeholders, timing, and evidence in a rough cut before committing to final shots, which can expose pacing but defer visual detail. Use the same team roles and record every ambiguity, revision, duplicated field, continuity miss, and approval. Do not introduce generated results unless a later authorized, dated test is separately documented.

Choose by failure recovery, then verify any product branch

Select the method that lets the team identify and repair its most expensive likely failure: changed meaning, unclear evidence, continuity drift, unusable duration, or late stakeholder review. Define the canonical artifacts, owners, handoffs, and point where a branch can safely change. If a named product will be used, verify the visible model label, accepted inputs, controls, region, entitlement, terms, and output handling at that date; record a fallback when unavailable. The comparison supports a production decision with disclosed trade-offs. It does not establish script parsing, generation, integration, access, quality, customer outcomes, or model superiority.