michailbul/burst-frame-cars

BURST FRAME — Automotive

The BURST FRAME technique generates consistent, full-resolution car reference frames from a single Seedance 2.0 generation. Instead of running 5+ separate image generations, all angles are encoded into one Rapid Fire video prompt. Seedance's consistency within a single generation ensures every frame shares the same car identity, paint color, and visual style. ffmpeg extracts the individual frames.

The result: 1 prompt → 1 Seedance video → usable reference frames ready to feed back into Seedance as start frames, or into Nano Banana as consistency anchors.

Default: 20 shots / 10 seconds. Use 10-shot mode when the user asks for fewer frames or when prompt length is tight.

---

What you always deliver

1. Shot list — automotive angles tailored to the car and environment
2. Seedance BURST FRAME prompt — compressed, copy-paste ready, ≤ 3500 characters
3. ffmpeg extraction command

---

⚠️ Character Limit — Non-Negotiable

Every prompt must be ≤ 3500 characters total before delivery.

This limit exists for Runway ML aggregator compatibility. Steps:

1. Write the full prompt
2. Count characters (run echo -n "..." | wc -c or equivalent)
3. If over 3500: compress shot descriptions to shorter phrases, trim the Global block, shorten the identity/setup header
4. Never output a prompt over 3500 characters regardless of how much detail is requested
5. When trimming, preserve identity lock and shot diversity over verbose descriptions

---

Step 1 — Lock the inputs

Extract from context or ask for:

| Input | Examples | |---|---| | Car | Ferrari 812 rosso corsa, Porsche 911 RWB dark plum, black Lamborghini Huracán | | Environment | Studio, alpine road, canyon night, forest, urban, desert | | Style | Cinematic hyperrealistic, luxury editorial, stylized 3D | | Color system | The car's color + the world's palette relationship | | Frame count | 20 (default) or 10 (compact mode) |

If the user provides a reference image, extract car model, paint color, and environment from it.

---

Step 2 — Build the shot list

Draw from this angle bank. Select shots that best serve the car and environment combination.

Exterior — Full Car

• Front 3/4 (driver's side, ~45°) — bonnet, one headlight, profile line
• Rear 3/4 — roofline, wing/spoiler, one taillight, exhaust
• Dead front (0°) — headlights centered, symmetrical, confrontational
• Dead rear (0°) — taillights, diffuser, exhaust layout
• Side profile (90°) — full car length, stance, ride height, wheel well
• Overhead top-down — roof geometry, full footprint, body symmetry
• Ground level extreme low — car looming, underside glimpse, heroic angle
• Low front 3/4 — fender flares prominent, aggressive low perspective

Detail Shots

• Headlight cluster — DRL signature, lens housing, projector detail
• Taillight cluster — LED pattern, lens depth, color
• Front grille and badge — brand emblem, chrome, intake mesh texture
• Wheel and rim — spoke pattern, brake caliper color, tyre sidewall
• Wheel arch / fender — panel gap, flare edge, body line
• Exhaust tips — tip count and layout, material finish
• Rear diffuser — fin detail, carbon or body-color
• Bonnet/hood surface — paint depth, reflections, power dome if present
• Side mirror — stalk, housing shape, reflection in glass
• Door shut line — panel alignment precision, handle recess

In-Motion (road environments)

• Tracking side pan — car sharp, environment motion-blurred behind
• FPV behind — tail of car leading into road ahead
• Retreating front — camera backing away, headlights filling frame
• Cockpit POV — through windshield, interior in shadow, road ahead
• Low panning — tarmac blur at wheel level, horizon low
• Overhead aerial — car small against landscape

Environment-Specific

• Studio — mirror floor reflection beneath car, backdrop gradient catch
• Night road — headlight beam reaching into dark, brake disc orange glow
• Alpine/forest day — wildflower foreground blur, god ray on roof
• Canyon — cliff edge in frame, stars overhead, puddle reflection
• Desert — dust trailing behind, heat shimmer off hood

---

Step 3 — Write the prompt

20-shot template (10 seconds)

same vehicle throughout all shots, same [car name + color] consistent identity every shot.

10 seconds, 20 shots (Rapid Fire). [Car identity — 1 sentence]. [Environment]. [Style + mood]. [Color palette]. [Film format].

[1] [one-line shot description].
...
[20] [one-line shot description].

Global: same [car] identity every shot, [environment constraint], [lighting rule], no other vehicles. Audio: [sound design].

10-shot template (5 seconds)

same vehicle throughout all shots, same [car name + color] consistent identity every shot.

5 seconds, 10 shots (Rapid Fire). [Car identity — 1 sentence]. [Environment]. [Style + mood]. [Color palette]. [Film format].

[1] [one-line shot description].
...
[10] [one-line shot description].

Global: same [car] identity every shot, [environment constraint], [lighting rule], no other vehicles. Audio: [sound design].

Physics — name these explicitly

| Element | Exact language | |---|---| | Tire smoke | tire smoke radial displacement | | Brake disc | brake disc igniting orange-red under compression, heat shimmer rising | | Exhaust (cold air) | exhaust pulse visible as white breath in cold air | | Road motion blur | road surface motion blur at speed | | Wildflower blur | radial motion blur on foreground wildflowers | | Wet tarmac | wet tarmac surface mirror reflections | | Studio floor | seamless studio floor mirror reflection | | Paint surface | paint depth visible, orange-peel texture catching raking light |

---

Step 4 — ffmpeg extraction

# 20-shot: 1 frame per 0.5 seconds
ffmpeg -i your_video.mp4 -vf fps=2 frame_%02d.png

# 10-shot: 1 frame per 0.5 seconds (same rate, 5s video)
ffmpeg -i your_video.mp4 -vf fps=2 frame_%02d.png

# Scene-change detection — better when Seedance cuts are crisp
ffmpeg -i your_video.mp4 -vf "select=gt(scene\,0.25)" -vsync 0 frame_%02d.png

Tell the user to drop the Seedance video here when ready — offer to run extraction directly.

---

Environment presets

Studio Seamless [color] backdrop and matching floor, soft diffused overhead-left light source, no hard shadows, mirror floor reflection. Medium format lens compression. Clean and minimal.

Alpine Mountain Road (Day) Warm golden daylight, dense pine forest on both sides, god rays through mist, pink wildflowers on road shoulder. 35mm grain, cinematic hyperrealistic.

Canyon Night Pitch-black, cold white DRL headlights as primary source, brake disc orange-red glow, stars overhead, crushed blacks. Anamorphic 2.39:1, predatory.

Forest Atmospheric Teal-amber grade, god rays through canopy, wet dark tarmac, mist in air. 70% deep shadow, cinematic.

Desert Burnt amber and ochre. Volumetric dust haze, harsh directional sun, isolated.

---

Quality rules

• ≤ 3500 characters — count before output, trim if over. Runway ML aggregator hard limit.
• Identity lock is non-negotiable — open every prompt with same vehicle throughout all shots, same [car] consistent identity every shot
• One line per shot — verbosity causes drift and pushes past character limit
• Name physics explicitly — unnamed physics renders flat and weightless
• Studio shots: car stationary, camera moves — never move the car in studio context
• Motion shots: state what moves — car, camera, or both must be explicit
• No STOP MOTION in BURST FRAME prompts — conflicts with rapid-fire pacing

---

Reference: full example outputs

See references/examples.md for complete worked examples:

• Ferrari 812 · Studio (charcoal grey backdrop)
• Ferrari 812 · Alpine Forest Road (daytime)
• Porsche 911 RWB · Studio (mauve backdrop)
• Ferrari 812 · Canyon Night