🎥 Create cinematic commercial or short film using a repeatable system for story, reusable assets, shot direction, continuity, sound, and final delivery.

The core workflow: Write the story first. Build every recurring character, location, vehicle, and prop as a reusable Element. Then generate each scene with exact shot timing, camera logic, performance, physics, audio, and continuity locks.

🎬 What You Are Building

This is an end-to-end AI production system for turning a simple concept into a polished commercial or short film. The goal is not one impressive clip. The goal is a complete, consistent production.

🚀 What This System Can Create

Cinematic car and product commercials
Emotional narrative short films
Dialogue scenes with consistent eyelines
Sports sequences with believable action
Multi-location stories with recurring characters
VFX-style shots and video playing inside a generated scene The same production logic works across all of them:

Production Layer	What It Controls
Script	The story, emotion, scene order, required assets, dialogue, and payoff.
Elements	Identity and continuity for characters, locations, vehicles, wardrobe, and props.
Scene prompts	Shot order, timing, blocking, camera, acting, physics, lighting, and sound.
Edit	Pacing, audio, transitions, color consistency, and final delivery.

🧰 Tools You Need

Keep the entire production stack simple: Claude handles the thinking and prompt structure, while Higgsfield handles the reusable assets, visuals, motion, and delivery.

Higgsfield for Cinema Studio, Elements, image generation, Seedance, and final video generation
Claude with Fable 5 for the screenplay and production-ready Seedance prompts
Soul Cinema for cinematic characters and locations
GPT Image 2 for controlled character, wardrobe, and product-style prop sheets
Seedream Pro 5 or Nano Banana Pro when a different image model produces stronger depth, materials, geometry, or editing
Seedance 2.0 in 4K or the strongest available Seedance model for video
Higgsfield Supercomputer for optional clip analysis, voiceover, and audio tasks
CapCut, Premiere Pro, DaVinci Resolve, or another editor

No single image model wins every task. Choose the model by the pixels it produces. If a prop looks flat, a face drifts, or a location lacks depth, keep the instruction and test another model before moving to video.

🧱 The Golden Rule

Build the production tools before generating the production. Every scene is made from characters, locations, and props. Lock those first, save them as named Elements, and reuse them everywhere.

Longer prompts cannot rescue weak references. If the model must invent the face, room, car interior, outfit, and hand prop during video generation, each shot can become a different movie.

✍️ Step 1: Write the Story First

The screenplay is not just creative writing. It is your asset list, shot map, emotional blueprint, and production plan.

Start with: - One-sentence premise - Main character and goal - Emotional problem or tension - Beginning, escalation, payoff - Final length - Maximum length of each generated shot - Dialogue and sound requirements - Every character, location, vehicle, outfit, and prop that appears For a two-minute film, the source workflow asked Claude for a screenplay where each individual shot could be produced within the model's maximum duration. A practical starting rule is one generation per shot or scene block, usually 15 seconds or less.

Copy-Ready Script Prompt

You are the screenwriter and production planner for a cinematic AI short film.

CONCEPT:
[Describe the idea in one paragraph.]

TARGET LENGTH:
[Example: 60 to 120 seconds.]

FORMAT:
[16:9 commercial, 21:9 short film, or 9:16 social video.]

STORY REQUIREMENTS:
- Give the story a clear beginning, escalation, emotional turn, and payoff.
- Keep every individual shot or producible scene block within 15 seconds.
- Use visual storytelling instead of unnecessary exposition.
- Make every character action physically filmable.
- Keep the number of recurring characters and locations manageable.
- Include dialogue only where it adds tension, emotion, or a punchline.

RETURN:
1. Logline
2. Character list
3. Location list
4. Prop, vehicle, and wardrobe list
5. Scene-by-scene screenplay
6. Shot duration estimates
7. Dialogue and sound notes
8. Continuity risks to solve before production

📋 Step 2: Build the Asset Inventory

Read the screenplay and identify everything the model must recognize repeatedly before generating a single frame of motion.

Asset Type	Best Reference Format	Why It Matters
Main character	Front full-body, rear full-body, tight face portrait	Locks identity, proportions, hair, and wardrobe.
Alternate outfit	Separate clean sheet or ghost-mannequin wardrobe sheet	Prevents clothes from changing between scenes.
Location	Empty 3/4 view showing two walls or strong depth	Gives the camera room to move and clarifies geography.
Prop	Front, side, rear, and useful detail views on gray	Prevents flat, changing, or floating objects.
Vehicle	Exterior turnaround plus separate interior views	Locks the body, wheels, badge, dashboard, seats, and cabin.
Composition guide	Annotated still, trajectory drawing, or approved frame	Controls exact placement, motion path, and camera angle.

Recommended Folder Structure

01 Script
02 Character Sheets
03 Wardrobe
04 Locations
05 Props and Vehicles
06 Approved Frames
07 Video Tests
08 Final Scenes
09 Audio
10 Final Edit

👤 Step 3: Build Character Sheets Seedance Can Read

A strong character sheet should be useful, clean, and unambiguous, not decorative. The less Seedance has to guess, the more consistent the performance becomes.

Real three-view character sheet generated inside the Higgsfield workflow 👆 Real Higgsfield example: a clean character sheet with consistent identity, full-body coverage, and a readable portrait.

Character Sheet Standard

Use a neutral gray seamless background.
Keep the lighting soft, even, neutral, and nearly shadowless.
Show one full-body front view from head to feet.
Show one full-body rear view from head to feet.
Show one large, readable face and shoulders portrait.
Keep identity, hair, wardrobe, proportions, and lighting identical.
Include shoes, hands, and important accessories clearly.
Avoid dramatic poses, colored lighting, hard shadows, and busy backgrounds. The videos found gray more reliable than white or black. A clear face image plus a full-body outfit view reduces what the video model must guess.

Character Sheet Template

Create a three-panel photorealistic character reference sheet of one identical person.

PANEL 1:
Full-body front view, head to toe, feet visible, neutral standing pose.

PANEL 2:
Full-body rear view, head to toe, same posture and scale.

PANEL 3:
Tight face and shoulders portrait, direct calm gaze, facial features clearly readable.

CHARACTER:
[Age, heritage, build, height, facial structure, eyes, hair, skin, distinctive details.]

WARDROBE:
[Describe every garment, fabric, fit, color, footwear, and accessory.]

LIGHTING AND BACKGROUND:
Neutral mid-gray seamless studio background. Soft, even, wraparound light at approximately 5500K to 5600K. Minimal shadows. Natural skin texture and eye catchlights.

LOCKS:
The same person, identity, facial structure, hairstyle, outfit, accessories, body proportions, scale, background, and lighting in all three panels. No extra people, no text, no cropped feet, no dramatic pose, no CGI appearance.

Wardrobe changes need their own locked asset. Do not ask the video model to redesign an outfit while also performing a complex scene. Create and approve the new look first.

🏙️ Step 4: Build Locations for Movement

A beautiful room can still be a bad video reference. Build locations around depth, geography, camera travel, and character movement.

Empty football stadium location generated in Higgsfield for the Santiago short film 🏟️ Real Higgsfield example: the stadium is created empty first so the architecture, scale, lighting, and action space remain clean.

Strong Location Checklist

Use a 3/4 angle so at least two planes of the space are visible.
Make depth and travel paths easy to understand.
Leave enough clear space for the planned action.
Remove people unless they are essential to the still.
Remove floating objects, impossible contact points, unreadable furniture, and broken architecture.
Define the main light source, white balance, time of day, and atmosphere.
Keep the image free of real logos and accidental text.
Decide whether the image is a style and geography reference or an exact composition lock. For large crowd scenes, the source workflow generated the stadium empty and asked Seedance to populate it later. This avoids messy still-image crowds becoming baked into the reference.

Location Template

Create an empty cinematic location reference for [SCENE].

GEOGRAPHY:
[Describe the room, street, stadium, or landscape from a 3/4 angle. State where important doors, windows, furniture, roads, and action areas are located.]

CAMERA:
[Height, angle, field of view, asymmetric composition, and whether the horizon is level or canted.]

LIGHT:
[Time of day, motivated light source, white balance, haze, shadow behavior, and highlight roll-off.]

MATERIALS AND PALETTE:
[Dominant materials and a 60:30:10 color structure.]

ACTION SPACE:
Leave clear physical room for [character movement, vehicle path, camera travel, or interaction].

CONSTRAINTS:
No people, no broken architecture, no floating objects, no real logos, no readable text, no CGI appearance. Photorealistic and physically coherent.

🚗 Step 5: Make Props and Vehicles Feel Physical

Props must have visible thickness, contact shadows, believable materials, real scale, and multiple perspectives. If it looks flat in the still, it may animate like a cardboard cutout.

Multi-angle car reference sheet created for the Higgsfield cinematic commercial workflow 🚘 Real Higgsfield example: one approved multi-angle car sheet becomes the design source for every exterior shot.

Prop Sheet Standard

Neutral gray background
Front, side, rear, and any important top or underside view
Same design and scale in every panel
Soft product lighting
Visible surface texture and construction
Subtle contact shadow
No hands or unrelated objects
Crisp text only when the exact text is essential If a generated prop looks like a flat cardboard cutout, switch models. The remote-control example improved when the same instruction was moved from GPT Image 2 to Nano Banana Pro.

Vehicle Production Pack

For a recurring car, create: - Exterior front, rear, side, and 3/4 views - Wheel and badge close-ups - Key fob - Front cabin - Rear cabin - Steering wheel and dashboard - Driver-seat view - Trunk or cargo area if used - Clean parked-position frame for exact placement - Optional drawn trajectory for difficult turns or stops

Exterior consistency does not guarantee interior consistency. Treat the cabin, steering wheel, seats, dashboard, badge, and controls as separate references when they appear clearly.

🏷️ Step 6: Save Everything as Named Elements

Elements are your continuity system. Give every recurring asset a short, exact, reusable name and use that same name everywhere.

Inside Higgsfield

Approve the final asset.
Select Create Element.
Give it a short, exact name.
Repeat for every recurring asset.
Use the exact same names in Claude.

Naming Example

@hero          main character in casual wardrobe
@hero-kit      hero in sports uniform
@consultant    supporting character
@car           exterior vehicle reference
@car-front     front cabin and dashboard
@car-rear      rear cabin
@key           vehicle key fob
@showroom      dealership location
@street        downtown driving location
@stadium       night arena
@ball          tournament ball
@position      exact composition or parked-position lock

Matching names allow Claude's prompts to call the correct Elements consistently.

🧠 Step 7: Turn Claude Into the Prompt Department

Claude should see the screenplay, approved assets, Element map, timing rules, and creative direction before writing any final scene prompt.

The linked tutorials provide downloadable Claude skills in their descriptions or companion breakdowns. Add the relevant Seedance prompt-writing skill through: Claude → Customize → Skills → Upload skill Then open a fresh chat and attach: - The final screenplay - Every approved asset or sheet - The Element name map - Format, duration, and delivery requirements - Any approved composition frames - Your creative and continuity rules

Describe the result you want. The source workflow warns that negative phrasing can backfire. Instead of “he is not crying,” write “his face stays dry, his eyes are anxious, and his lips are tightly pursed.”

🎞️ Step 8: Direct Every Shot Like a Filmmaker

Use camera language to communicate story, emotion, geography, and power, not just style. Every lens, angle, movement, and cut should have a reason.

Technique	When to Use It	What to State
Wide establishing shot	Reveal geography, scale, or danger	Subject placement, environment, travel path, and focal priority
Static anchor shot	Lock who sits or stands where before complex coverage	Exact left and right positions, eyelines, and furniture contact
Over-the-shoulder pair	Dialogue and emotional exchange	Who is foreground, who is sharp, screen direction, and eyeline
180-degree rule	Keep dialogue geography coherent	Character A always screen-left, Character B always screen-right
Dutch angle	Instability, pressure, or loss of control	Degree of cant and whether every shot shares it
Continuous handheld take	Immersion, urgency, and physical presence	Operator path, reframes, focus movement, and no-cut lock
Whip pan or match cut	Connect time periods, locations, or actions	Exact outgoing motion, incoming motion, and cut point
Speed ramp	Hold a decisive action, then restore impact	When slow motion begins, duration, and snap back to real time
Shadow-side camera	Natural depth and cinematic faces	Motivated light source, backlight, face shadow, and catchlights

Composition Rules That Improve Control

Place subjects on thirds instead of centering everything.
State headroom and lead room.
State who looks screen-left or screen-right.
Use one static shot to establish the layout before fast coverage.
Give the first frame active blocking, not a frozen empty setup.
Define exact hard-cut times.
Specify whether the camera is locked, handheld, tracking, craning, panning, or using optical zoom.
State whether a reference controls identity, style, geography, composition, or all four.

🧩 Step 9: Build the Complete Scene Prompt

A strong scene prompt is a production document. It tells the model what exists, where everything is, what happens, how it is filmed, and what must never drift.

Master Scene Prompt Template

SCENE PURPOSE:
[What changes emotionally or narratively in this scene?]

ACTIVE REFERENCES:
@character — identity, wardrobe, and role
@location — style and geography reference, or exact composition lock
@prop — exact design and scale
@vehicle — exterior or interior design lock

FORMAT:
[One continuous shot or timed multi-shot.]
[Total duration, aspect ratio, frame rate, exact cut times.]

LOCATION MAP:
Foreground:
Midground:
Background:
Screen-left:
Screen-right:
Character and camera travel paths:

FIRST FRAME:
Describe an active, non-empty opening frame with exact blocking.

SHOT PLAN:
SHOT 1 [time range]:
- Framing and lens
- Camera position and movement
- Subject placement and eyeline
- Character action
- Dialogue or sound
- Transition or cut

SHOT 2 [time range]:
[Repeat as needed.]

PERFORMANCE:
Describe micro-pauses, eye movement, breath, posture, tension, and restrained reactions.

PHYSICS:
State body weight, inertia, contact shadows, object weight, cloth behavior, vehicle suspension, debris, water, snow, or impact behavior.

LIGHTING:
Motivated key source, camera on light or shadow side, white balance, haze, reflections, highlight roll-off, and practical lights.

COLOR:
Use a 60:30:10 palette tied to materials and real light sources.

AUDIO:
Environmental sound, dialogue, effects, and whether music is present. State no subtitles if required.

STYLE:
Photorealistic live-action, physical cine-lens character, natural grain, realistic motion blur, no game-engine appearance.

POSITIVE LOCKS:
Repeat the exact identities, wardrobe, prop geometry, screen positions, action direction, camera rules, cut count, and continuity conditions that must remain unchanged.

⚙️ Step 10: Generate in the Right Order

Approve the idea, then the assets, then the movement, then the resolution. Do not spend premium credits polishing a scene whose motion or continuity is still broken.

Generate and approve all still assets.
Create and name Elements.
Start with a short test or the hardest technical beat.
Generate a simple anchor shot before a complex multi-angle sequence.
Review identity, spatial continuity, action direction, and object geometry.
Fix the prompt or reference that caused the problem.
Regenerate only the failed shot when possible.
Lock approved clips and move forward.
Upscale or use 4K after the motion and continuity are approved.

Credit-saving rule: A questionable still becomes an expensive video failure. Regenerate the face, location, vehicle, or prop before animation if anything is unclear.

🏎️ Step 11: Apply the Car Commercial Workflow

The car tutorial uses the same story-first system, with extra attention to vehicle geography, exterior continuity, cabin consistency, road physics, and character reactions.

Car showroom location created inside the Higgsfield commercial workflow 🏢 Real Higgsfield example: the showroom is locked as a reusable location before the hero, consultant, car, and action are animated inside it.

Car Commercial Production Order

Lock the car exterior sheet.
Isolate the badge or logo when it needs close-up consistency.
Create the hero and alternate disguise or wardrobe.
Create the supporting characters.
Generate the key, framed photo, newspaper, and other story props.
Build each location: showroom, pickup point, residential street, downtown, highway, tunnel, checkpoint, and final location.
Generate separate cabin references.
Capture approved steering-wheel and interior frames.
Create exact parked-position and trajectory references.
Generate the story scene by scene.

What Makes the Commercial Feel Directed

The car is introduced through story, not random beauty shots.
A disguise and reveal create a narrative question.
Static exterior and interior shots establish the product.
Driving shots alternate exterior speed with interior reactions.
Character positions are locked before six-angle cabin coverage.
Dialogue includes micro-pauses and restrained reactions.
A tollbooth punchline uses four fast shots and one longer final hold.
The last held reaction gives the joke time to land.

Car Motion Prompt Priorities

Exact lane and direction of travel
Which line the car never crosses
Speed and overtaking side
Body roll and suspension load
Tire contact and road grip
Camera rig position
Reflections moving across glass and paint
Driver and passenger screen positions
Seatbelts, hands, steering wheel, and assist-handle contact
Engine, wind, tire, and cabin sound

⚽ Step 12: Apply the Emotional Short-Film Workflow

The football film demonstrates how camera grammar, restrained acting, sound, and transitions can carry emotion without overexplaining the story.

Santiago character reference from the Higgsfield football short-film workflow 🎭 Real Higgsfield example: a locked identity reference gives every emotional scene the same recognizable lead character.

Story Construction

Begin with a present-day emotional problem.
Reveal the pressure through restrained dialogue.
Use a flashback or decisive memory as the visual centerpiece.
Connect past and present with a motivated transition.
End on a new human interaction that changes the hero's direction.

Directing Lessons

A Dutch angle can make a calm room feel psychologically unstable.
The 180-degree rule keeps emotional dialogue readable.
A wide shot establishes distance between characters.
Over-the-shoulder shots intensify the personal exchange.
A speed ramp lets the decisive kick hold tension before the save.
A whip pan or match cut can move between memory and the present.
The ending can keep the adult hero in silhouette so the scene belongs to the child.
A low camera rising with the child makes recognition and hope feel larger.

Performance Language

Avoid generic directions such as “sad” or “shocked.” Use: - Eyes held down before slowly lifting - Fingers working against each other - A planted heel with a nervous forefoot bounce - A micro-pause before a difficult line - Wet catchlights while tears remain contained - A slow blink and quiet breath from the listener - A smile that builds gradually instead of appearing instantly - Restrained surprise with a small eyebrow lift and short open-mouth beat

📺 Step 13: Place Video Inside Generated Scenes

Seedance can treat an approved clip as media playing on a physical screen inside another scene, creating an advanced VFX-style composite without a traditional manual workflow.

Workflow

Export the exact portion of the source video that should play on the screen.
Add it as a video reference.
Define the television as a physical object inside the room.
State that the source must match 1:1 in content, timing, framing, motion, and color.
Keep the screen fully visible and unobstructed.
Add filmed-screen realism: glass sheen, mild glare, room reflections, subtle pixel structure, moiré, bloom, and off-axis perspective.
Keep the surrounding room, viewer, and television scale locked. This lets an existing generated clip become story content inside another AI-generated shot without a traditional manual composite.

🔊 Step 14: Build Audio Into the Scene

Sound is part of the direction. Define the voice source, room tone, dialogue, physical effects, music rules, and silence inside every prompt.

Include These Audio Instructions

Environmental room tone
Dialogue with exact wording
Vehicle, impact, weather, or object sound effects
Whether music is present
Voice location, including on-screen, off-screen, or from a television
No subtitles when subtitles are unwanted For documentary-style footage, Higgsfield Supercomputer can analyze an approved clip and draft a matching voiceover. The tutorial used a simple request:

Analyze this video and create a concise voiceover in the style of a premium wildlife documentary. Match the pacing and visible action. Keep the facts grounded in what is actually shown.

In the final edit, replace or realign audio when generation introduces drift, repeated words, or inconsistent sound.

🛠️ Step 15: Fix the Common Failures

Do not keep regenerating blindly. Identify whether the failure came from the reference, blocking, action density, camera instruction, or continuity lock, then repair that specific layer.

Character identity changes

Strengthen the face sheet, attach a larger portrait, reduce simultaneous changes, and repeat exact identity anchors in the positive locks. Regenerate the failed shot instead of the entire sequence.

Prop looks flat or floats

Use multiple views, visible thickness, material response, a contact shadow, exact scale, and real hand contact. Test a different image model before generating video again.

Location changes between shots

Define one location Element, state the geography, lock furniture and architecture, keep white balance consistent, and use an approved establishing frame.

Characters switch sides during dialogue

Write the 180-degree rule explicitly. Lock one character screen-left and the other screen-right in every shot. Define who appears in each over-the-shoulder foreground.

Vehicle exterior or interior changes

Use separate exterior, front-cabin, rear-cabin, steering-wheel, badge, and position references. State which reference controls each shot.

Complex movement breaks

Simplify the number of actions, define the action by time segment, add a trajectory image, establish the blocking with a static shot, or split the beat into separate generations.

The model performs the opposite of the instruction

Remove negative phrasing and describe the visible positive result. Replace “not crying” with “dry cheeks, contained emotion, wet eyes, tightly controlled mouth.”

Emotion feels exaggerated

Replace broad labels with physical micro-performance: breath, blink, eye movement, posture, hand tension, and a short held pause.

Camera invents extra cuts or motion

State the exact number of shots and cuts, exact cut times, and a positive lock such as “one continuous locked-off shot, the camera stays in one physical position.”

✅ Step 16: Final Edit and Quality Control

The final edit is where separate generations become one believable film. Match the rhythm, exposure, sound, color, grain, and emotional progression across every scene.

Edit

Assemble approved scenes in screenplay order.
Preserve hard cuts when the story calls for them.
Use match cuts and whip transitions only when visually motivated.
Align dialogue and sound effects precisely.
Match exposure, white balance, contrast, and grain across scenes.
Add music only after the scene rhythm works without it.
Keep the strongest opening image and remove slow setup.

Final Review Checklist

Story

[ ] The opening creates a clear question.
[ ] Every scene changes the situation.
[ ] The ending delivers an emotional or commercial payoff.
[ ] Unnecessary shots are removed.

Identity and Continuity

[ ] Faces, hair, wardrobe, and proportions stay consistent.
[ ] Props and vehicles retain the same design.
[ ] Character screen positions and eyelines match.
[ ] Locations retain the same geography and lighting logic.

Camera and Motion

[ ] Every movement has a narrative purpose.
[ ] Cuts occur at the intended times.
[ ] Physics, weight, contact, and inertia look believable.
[ ] No unintended zooms, pans, cuts, or speed changes appear.

Image

[ ] Skin and materials retain texture.
[ ] Architecture and object geometry stay stable.
[ ] Reflections and shadows match the scene.
[ ] The final resolution holds up at normal viewing size.

Audio

[ ] Dialogue is correct and synchronized.
[ ] Environmental sound supports the location.
[ ] Music does not cover the important performance.
[ ] No repeated words, unwanted subtitles, or AI artifacts remain.

⚡ Summary

1. Write one strong premise. Define the hero, problem, escalation, and payoff.
2. Ask Claude for the screenplay. Keep every producible shot within the model's duration limit.
3. Extract the asset list. Characters, outfits, locations, props, vehicles, interiors, and composition guides.
4. Build gray-background character sheets. Front, rear, and face portrait.
5. Build locations in a 3/4 view. Prioritize clean geography and action space.
6. Build multi-view props and vehicles. Include thickness, materials, contact shadows, and interiors.
7. Approve every still before motion. Fix weak references now to save credits later.
8. Save assets as Higgsfield Elements. Use short exact names.
9. Give Claude the script, assets, and Element map. Use the downloadable Seedance skill from the linked tutorials.
10. Design each scene like a director. Specify shots, lenses, screen direction, blocking, timing, acting, physics, light, color, and audio.
11. Write positive locks. Repeat exactly what must remain consistent.
12. Test the hardest beat first. Use short generations before committing to full scenes.
13. Generate scene by scene. Lock good takes and repair only failed shots.
14. Assemble the edit. Match pacing, color, grain, dialogue, sound effects, and music.
15. Run the full quality checklist. Story, identity, continuity, camera, physics, image, and audio.
16. Export and publish. Use the aspect ratio and resolution required by the final platform.

🧠 The Entire Method in One Sentence

The script decides what must exist, the Elements decide what everything looks like, the prompt decides how the scene is directed, and the edit decides whether the audience feels it.