Module 07 — TikTok Symphony: B-Roll Generation
Category: Production | Duration: ~90 min | Prerequisites: Module 06
Learning Objective
By the end of this module, you will be able to generate usable, consistent visual assets through TikTok Symphony by writing one precise prompt per shot, defining all required visual parameters, stating continuity constraints, reviewing results against the storyboard, and naming/organizing every accepted asset. You will treat Symphony as the visual-production stage, not the complete storytelling solution.
1. Principle: Precise Prompts Make Reliable Assets
TikTok Symphony generates AI-assisted B-roll. It is a powerful tool, but it is not intelligent — it does exactly what you ask. A vague prompt produces decorative but unusable footage. A precise prompt produces intentional, story-serving B-roll.
The Prompt Structure
Every prompt must follow this structure:
Subject + action + setting + composition + camera movement + lighting + mood + visual style + continuity constraints + unwanted elements
Breaking Down the Elements
| Element | What to specify | Example |
|---|---|---|
| Subject | The primary focus of the shot | "a lone paramedic in navy scrubs" |
| Action | What the subject is doing | "checking a patient's pulse" |
| Setting | Where it happens | "inside an ambulance, night" |
| Composition | Framing and 9:16 vertical | "close-up, subject in center safe area" |
| Camera movement | How the camera moves | "slow push-in" |
| Lighting | Quality and direction | "sickly fluorescent overhead, cold tone" |
| Mood | Emotional atmosphere | "clinical, tense" |
| Visual style | Aesthetic treatment | "documentary realism, shallow depth of field" |
| Continuity constraints | What must match other scenes | "same paramedic, same scrubs, same ambulance interior" |
| Unwanted elements | What to exclude | "no logos, no text, no distorted hands" |
Sample Prompt (Instructional)
Vertical 9:16 cinematic B-roll of an empty coastal railway platform at dusk, a lone figure in a dark coat standing beneath a flickering station light, wind moving loose paper across the platform, distant fog obscuring the tracks, slow forward camera movement, restrained navy and amber palette, realistic atmospheric detail, strong depth, subject positioned in the central safe area, no visible text, no logos, no watermark, no fabricated book cover, no extra limbs or distorted hands. The scene must convey uncertainty without revealing a specific plot event.
This is an illustrative visual prompt — it conveys a mood and atmosphere for a mystery novel, not a claim about any real book.
1b. Visual Demonstration — Professional vs. Weak B-Roll
Professional Example — Approved B-Roll Set
Prompt used: “Vertical 9:16 cinematic B-roll of a hospital corridor at night, a paramedic’s gloved hands checking a pulse monitor, blue emergency lighting casting sharp shadows on the wall, shallow focus on the monitor display, 35mm lens, slow push-in camera movement, desaturated blue-green palette, realistic detail, consistent with scene S02 character and lighting, no text, no logos, no watermark, no distorted hands. The scene conveys urgency without showing the patient’s face.”
Result: The monitor shows a steady, realistic waveform. The paramedic’s hands match S02 continuity (same glove style, same ring). Blue emergency lighting is consistent. No artifacts. 12-second clip fits the beat.
Weak Example — Rejected B-Roll
Prompt used: “Paramedic checking patient, medical room, cinematic, lots of action.”
Result:
- Shows three different people (continuity broken)
- “Cinematic” without specifying lighting or palette (generic look)
- Generated text overlay on the monitor (unwanted)
- Extra finger on the paramedic’s hand (AI artifact)
- 20 seconds long (too long for the beat)
- No negative constraints (logo from a real hospital appears in background)
- Character appearance doesn't match other scenes
Reviewer note: “Rejected — fails composition (busy), continuity (character mismatch), visual quality (extra finger), and usefulness in edit (wrong duration). Regenerate with the structured prompt above.”
2. Asset Review Criteria
Every generated asset must pass this review before it's approved for the edit:
| Criterion | Good | Poor |
|---|---|---|
| Composition | Clear focal point, depth, safe area | Busy, no clear subject |
| Relevance | Serves the storyboard's narrative function | Decorative, generic |
| Movement | Purposeful, not jittery or chaotic | Erratic, no rhythm |
| Continuity | Matches character/lighting/props | Inconsistent with other scenes |
| Visual quality | Clean, no artifacts | Distorted hands, glitches |
| Text-safe space | Clear area for captions | Text or objects in safe zone |
| Usefulness in edit | Fits the beat and duration | Too long/short, wrong framing |
3. Common Prompt Mistakes (Avoid All of These)
| Mistake | Correction |
|---|---|
| Combining too many actions in one short clip | One primary action per prompt |
| "Make it cinematic" (vague) | Specify lighting, palette, lens, movement |
| Conflicting lighting/mood instructions | Choose one lighting approach and stick |
| Unnecessary dialogue or generated text | No words unless the character is silent |
| Unintended visible logos | Add "no logos" to every prompt |
| Inconsistent character appearances | Specify continuity details explicitly |
| AI artifacts (extra limbs, distorted hands) | Add "no extra limbs, no distorted hands" |
| Overly literal imagery that reveals everything | Convey mood, not plot resolution |
| Unapproved use of real person's likeness | Use illustrative, composite characters |
| Generated cover art replacing real cover | Never generate book covers; use the real one |
4. Step-by-Step Workflow
Generate B-Roll
- Read the approved storyboard — each scene has defined parameters
- Identify which shots need AI B-roll — which scenes can't use stock/photo?
- Write one precise prompt per shot using the exact structure above
- Specify 9:16 vertical framing explicitly
- State continuity requirements for recurring elements
- Exclude unwanted elements in the negative
- Generate 2–3 candidate versions per shot (parallel takes)
- Review against storyboard using the review criteria
- Select the best usable version — document why
- Name and organize every asset
Naming Convention
All B-roll assets follow this pattern:
BOOKSLUG_CONCEPT01_SCENE03_TAKE02.mp4
- BOOKSLUG — the book's slug (e.g.,
paramedic) - CONCEPT01 — which creative concept (from the storyboard)
- SCENE03 — the scene ID (from the storyboard)
- TAKE02 — the take number
Organization
05_Symphony-Prompts/ paramedic_concept01_scene01_v1.txt ← prompt text paramedic_concept01_scene01_v2.txt paramedic_concept01_scene01_v3.txt 06_Broll-Generated/ paramedic_concept01_scene01_take01.mp4 paramedic_concept01_scene01_take02.mp4 paramedic_concept01_scene01_take03.mp4 07_Broll-Approved/ paramedic_concept01_scene01_take02.mp4 ← selected take (copy) paramedic_continuity-note_scene01.txt ← short continuity note
Quality Gate — B-Roll Review
- [ ] Every shot has at least 2 candidate versions generated
- [ ] Each shot has a selected take documented
- [ ] Continuity notes are written for each scene
- [ ] Rejected takes are preserved (for potential compositing)
- [ ] All assets are correctly named and organized
- [ ] Prompt log is complete (one prompt per shot)
5. Hands-On Exercise
Using your storyboard from Module 06 (the "paramedic The Overdose" concept), generate three candidate prompts for Scene S03 (the naloxone injection shot).
Write all three prompts following the exact structure, then:
- Select the strongest take and explain why (using the review criteria)
- Write a continuity note for the scene
- Name the selected file using the naming convention
- Document rejected takes and why they were rejected
If you cannot access TikTok Symphony directly, write the three prompts and evaluate them against the storyboard using the review criteria. The skill being tested is prompt precision and self-evaluation.
Scene S03 Reference (from your storyboard)
- Purpose: Action escalation (the intervention)
- Subject: Paramedic's gloved hands
- Action: Administer naloxone injection
- Setting: Ambulance interior, night
- Lighting: Sickly fluorescent, cold tone
- Camera: Close-up, overhead
- Mood: Clinical, tense
- Continuity: Same paramedic, same gloves, same patient
6. Assessment
Scored assessment (45 points total):
| Criterion | Points |
|---|---|
| Three prompts follow the exact structure | 15 (5 per prompt) |
| Prompts specify all 10 required elements | 10 (per prompt quality) |
| Prompts avoid all common mistakes | 5 |
| Selection rationale uses review criteria | 5 |
| Continuity note is specific and actionable | 5 |
| Naming convention is correct | 5 |
| Rejected takes documented with reasons | 5 |
| Total | 45 |
Pass threshold: 38/45
Submission Requirement: Submit the three prompts, the selected take with rationale, the continuity note, and the filename. Use the Prompt Log Template.
7. 15-Video Batch B-Roll Workflow
When producing the standard 15-video batch, generate all vertical B-roll from the approved master brief, not from ad hoc ideas.
Batch Rules
- The standard production unit is one batch of 15 short videos.
- Each video has one preferred 10-second scene; two scenes are the absolute maximum.
- Each prompt must be grounded in the relevant video brief and describe:
- Setting, composition, and atmosphere.
- Intended narrative beat.
- How the visual supports the accompanying narration.
- Relevant cover or character references when available.
- Prohibited text, logos, unrelated imagery, and misleading details.
- Generic footage that does not support the narration is not acceptable merely because it looks attractive.
- The book cover is the only permitted branding within the generated video.
TikTok Symphony Prompt Relevance Check
Before approving any take, answer all of these in one sentence:
- What narration does this shot support?
- What exact beat does it carry?
- Why is this footage better than a generic attractive clip?
If any answer is vague, reject or regenerate the take.
Deliberately Faulty Example (training)
Faulty prompt/take: A 10-second vertical clip of a generic luxury yacht at sunset for a memoir about emergency medical work. The voiceover says, "The night shift taught me to trust my hands." The footage is attractive but has no subject, no action, no setting from the brief, and no relevance to the narration.
Detection and rejection:
- The visual does not depict a night shift, hands, or medical context.
- The footage is decorative, not narrative.
- Mark the take as rejected in the prompt log.
- Regenerate using the structured prompt format with the correct subject, action, setting, and negative constraints.
- If the correction reserve is limited, fix wording and assembly in the editor first; regenerate only if the visual is unusable.