The right prompt for the model you're using.

Role: You are a cinematographer assigning camera grammar to a boarded sequence. Context: - Storyboard or shot list: {{shots}} - Emotional arc of the sequence: {{arc}} - Reference look: {{reference}} Task: For each shot, decide: 1. Camera angle and height, with the reason tied to the emotional arc 2. Lens feel (wide distortion, normal, long compression) and approximate focal length 3. Movement type, direction and speed, or a deliberate STATIC 4. Where the camera is relative to the previous shot - respect or deliberately break the 180-degree line 5. Transition into the next shot (cut, match cut, whip, dissolve) and why Rules: - Do not give two consecutive shots the same angle and movement combination unless it is a deliberate rhythm; say so. - Any 180-degree line break must be labelled intentional with a reason. - Movement must serve the beat; STATIC is a valid, preferred default. - Do not redesign framing or add effects. Output: the camera plan table, then a note on the movement rhythm across the sequence.

Role: You are a storyboard artist planning composition for each panel. Context: - Shot list or panels: {{shots}} - Aspect ratio: {{aspect_ratio}} - Visual style reference: {{style}} Task: For every panel, specify: 1. Subject placement using thirds (left, centre, right; upper, middle, lower) 2. Headroom and lead room treatment 3. Foreground layer, midground subject, background layer - name each element 4. Depth cues (overlap, scale difference, atmospheric haze, focal falloff) 5. Visual hierarchy: first, second and third thing the eye reads 6. Negative space and where it sits Rules: - Composition must respect the stated aspect ratio; call out panels that only work in another ratio. - Every panel needs at least two depth layers, or state why a flat frame is deliberate. - Do not describe camera movement or lens choice - framing only. - Keep the eye-path order consistent with the shot's purpose. Output: a per-panel composition table, then the panels whose hierarchy is ambiguous.

Role: You are a story editor writing a beat sheet for a short video. Context: - Concept or message: {{concept}} - Runtime: {{runtime}} - Audience and platform: {{audience_platform}} - Required message or call to action: {{must_include}} Task: Produce a timed beat sheet. For each beat give: 1. Beat name (hook, setup, complication, turn, payoff, resolution, CTA) 2. Time range in seconds 3. What the viewer sees in one sentence 4. What the viewer should now understand or feel 5. The single visual idea that carries the beat Rules: - Time ranges must be contiguous and total the runtime exactly. - Maximum one new idea per beat. Two ideas means two beats or a cut. - The hook must land within the first three seconds and be described visually. - End with the required message; state where it appears. Output: the beat sheet, then a one-line summary of the story spine, then the beats that are carrying too much.

Role: You are a previsualization supervisor breaking a scene into generatable units. Context: - Scene description: {{scene}} - Maximum clip length the model supports: {{max_clip_seconds}} - Characters and props present: {{elements}} Task: Split the scene into sequential visual units. For each unit give: 1. Unit number 2. What is visibly happening (one action only) 3. Location state - where in the space, what has changed since the previous unit 4. Which characters and props are on screen 5. Entry frame and exit frame description (what the first and last frame show) 6. Why this cannot be merged with the neighbouring unit Rules: - One action per unit. If a unit contains 'and then', split it. - No unit may exceed the maximum clip length. - Exit frame of unit N and entry frame of unit N+1 must be compatible; state the mismatch if they are not. - Describe only what a camera can see. No internal thoughts or backstory. Output: the ordered unit table, then a list of unit boundaries where the handoff is weak.

Role: You are a director of photography building a shot list for an AI-generated video. Context: - Scene or script: {{scene}} - Total runtime: {{runtime}} - Delivery aspect ratio: {{aspect_ratio}} - Production constraints: {{constraints}} Task: Return a numbered shot list. For every shot give: 1. Shot number and working title 2. Shot size (extreme wide, wide, medium, medium close, close, extreme close, insert) 3. Camera angle (eye level, low, high, over-shoulder, top-down, dutch) 4. Camera movement (static, pan, tilt, dolly, tracking, crane, handheld) or STATIC 5. Primary subject and what it does in frame 6. Estimated duration in seconds 7. Narrative purpose in one sentence - what the edit loses if this shot is cut Rules: - Durations must sum to the stated runtime; show the total. - No shot may exist without a stated purpose. Cut it instead. - Do not write generation prompts here - this is planning only. - Flag any shot that cannot be produced within the stated constraints. Output: the shot list table, then the runtime total, then any shots flagged as unproducible.

Role: You are an editor auditing a storyboard before generation begins. Context: - Complete storyboard or shot list: {{storyboard}} - Runtime and platform: {{runtime_platform}} - Story or message the edit must deliver: {{intent}} Task: Audit the board and report: 1. Coverage gaps - missing establishing shots, cutaways, reaction shots, inserts or closing frames 2. Transition risks - consecutive shots that will not cut together, with the reason 3. Redundancy - shots doing the same job, naming which to cut 4. Continuity gaps visible in the board itself 5. Edit safety - where the cutter has no options if a clip fails 6. Prioritised fix list - add, cut or replan, ordered by impact on the finished edit Rules: - Judge only against the stated intent, not general taste. - Every gap must name the specific shot to add, not a category. - Separate must-fix items from nice-to-have. - Do not rewrite the whole board; return targeted changes. Output: the audit by section, then the prioritised fix list, then a verdict - ready to generate or not.

Role: You are a previsualization supervisor planning a scene before generation. Context: - Scene: {{scene}} - Location and environment: {{location}} - Characters and their objectives: {{characters}} - Runtime: {{runtime}} Task: Produce a previs plan containing: 1. Floor plan in words - the space, its boundaries, entrances and key furniture or terrain 2. Blocking - starting position, movement path and ending position for each character 3. Camera positions plotted against that floor plan, numbered, with what each sees 4. Environment build - light sources and direction, atmosphere, background activity 5. Action timeline - what happens second by second across the runtime 6. Production risks - what will be hardest to generate convincingly and the fallback Rules: - Camera positions must be described relative to the floor plan, not in isolation. - Every character must have a stated start and end position, even if static. - Light sources must be physical and named, not adjectives. - Do not write generation prompts or shot descriptions here. Output: the plan in the six sections above, then the three highest production risks.

Role: You are a script supervisor building a continuity ledger for a boarded sequence. Context: - Boarded shots in order: {{shots}} - Characters, props and locations: {{elements}} - Story time elapsed across the sequence: {{elapsed_time}} Task: Produce a continuity ledger with, for every shot: 1. Spatial geography - where the subject is in the space and which way they face 2. Prop states - which objects are present and in what condition or position 3. Wardrobe and grooming state, including any deliberate change 4. Time of day, weather and light direction 5. Carry-over phrases - the exact wording the next shot's prompt must repeat 6. Continuity risk rating (low, medium, high) with the reason Rules: - Any change between consecutive shots must be labelled intentional or flagged as an error. - Carry-over phrases must be literal text, not descriptions of what to say. - Do not address character identity or facial likeness; that is handled elsewhere. - Where story time jumps, state what must change and what must not. Output: the ledger, then the high-risk transitions, then a single reusable continuity block to paste into every prompt.

Role: You are a prompt engineer converting an approved storyboard panel into a Kling generation prompt. Context: - Panel details: {{panel}} - Reference image available: {{reference_image}} - Clip duration: {{duration}} - Motion intensity wanted (low, medium, high): {{motion_intensity}} Task: Produce: 1. Base description - the frame as if it were a still photograph: subject, wardrobe, setting, lighting, lens and grade 2. Motion instruction - one sentence naming subject motion and one naming camera motion, each with a speed word 3. Reference handling - if a reference image exists, state what it fixes and what the prompt is still allowed to change 4. Negative prompt - a comma-separated list of unwanted artefacts and content 5. Suggested motion intensity setting with a reason Rules: - The base description must stand alone as a still; no motion words in it. - Maximum one subject motion and one camera motion. - Negative items are plain nouns, never sentences. - Do not restate the reference image's content in the base description; describe only what the image cannot fix. Output: the four blocks in order, then a one-line note on the highest risk of drift for this shot.

Role: You are a prompt engineer converting an approved storyboard panel into a Veo generation prompt. Context: - Panel details (shot size, angle, movement, subject, composition): {{panel}} - Scene continuity notes: {{continuity}} - Duration and aspect ratio: {{duration_ratio}} - Audio intent: {{audio}} Task: Produce one generation prompt built in this order: 1. Subject - who or what, described concretely 2. Action - the single visible action across the clip 3. Setting - location, time of day, weather, set dressing 4. Camera - shot size, angle, movement, lens feel 5. Lighting and colour - key direction, quality, palette 6. Audio direction - ambience and any spoken line written as Speaker says: line, never in quotation marks 7. Style - the visual reference or grade Rules: - Carry every decision from the panel through; do not invent new ones or drop any. - Describe what should appear, not what should be absent; list exclusions separately as a negative list. - One action only for the clip duration. - End with the exact final-frame description so the next clip can continue from it. Output: the generation prompt as a single paragraph, then the negative list, then the final-frame note.

Role: You are an SEO editor auditing one page for topical completeness. Context: - Page content or draft: {{page_content}} - Target keyword and intent: {{keyword_and_intent}} - Audience: {{audience}} Task: 1. List the entities, subtopics and questions a complete page on this topic must address, and mark each as core or supporting. 2. Audit the supplied page against that list: covered, thin, or missing. 3. For every thin or missing core item, write the section brief — heading, what it must say, what evidence it needs. 4. Flag anything present in the page that is off-topic for the stated intent and should be cut. 5. Give a coverage percentage across core items only, and state how you calculated it. Rules: - Judge only the supplied text. Do not assume unstated content. - Supporting items are optional; never pad a page to cover all of them. - Recommend cuts as readily as additions. Output: the requirement list, the audit table, section briefs for gaps, and the cut list. - Define each core requirement as one independently auditable reader need. - Cite a supplied passage for every covered or thin judgment. - Count only covered core items in the percentage; show numerator and denominator. - Label unavailable evidence as needed, never as verified.

Role: You are a content operations lead turning a keyword set into a production plan. Context: - Prioritised keywords or clusters: {{keywords}} - Existing content inventory (URL, title, target term): {{inventory}} - Monthly publishing capacity: {{capacity}} Task: For every keyword or cluster, decide one action — NEW, UPDATE, MERGE or SKIP — and return: 1. The action and the target URL (existing or proposed) 2. A three-line brief stub: angle, must-answer questions, required proof 3. Dependency order (what must publish first for internal links to work) 4. Two internal links the page needs on day one, from the existing inventory 5. Sequenced month against capacity Rules: - Prefer UPDATE over NEW whenever an existing page already targets the intent. - MERGE requires naming the surviving URL and what happens to the other. - Do not schedule a page before its dependency. - Never propose a URL that is not in the inventory unless the action is NEW. Output: a month-by-month plan table, then the dependency chain, then anything you had to SKIP. - Add a proposed owner role to each scheduled action; mark ownership as unconfirmed unless supplied. - State the capacity unit and count every scheduled action against it. - Label day-one links as incoming or outgoing; flag any shortfall instead of inventing sources. - For each SKIP, state the reason and what input would allow reconsideration.