Higgsfield Marketing Studio

Higgsfield Marketing Studio

QUICK FACTS

Generated-checked block (scripts/build_index.py verifies anchors). Read the linked sections for full context — these lines are routing aids, not the rules themselves.

• Hard duration cap: 4–15s per clip; out-of-range values get clamped; longer narrative = multi-clip sequence edited externally →
• 9 presets: UGC, Tutorial, Unboxing (ugc_unboxing), Hyper Motion, Product Review, TV Spot, Wild Card, UGC Virtual Try On, Pro Virtual Try On →
• Pro Virtual Try On slug is virtual_try_on, NOT pro_virtual_try_on →
• Preset routing happens via show_marketing_studio.mode — generate_video has NO mode parameter at all →
• Hook + setting picklists on FIVE presets only: UGC, Tutorial, Unboxing, Product Review, UGC Virtual Try On (NOT Pro Virtual Try On) →
• 9 hooks (4 stunt / 5 subtle) as of 2026-05-18 — picklists drift; enumerate live for current UUIDs →
• 14 settings (8 realistic / 6 unrealistic) as of 2026-05-18, passed by UUID as setting_id →
• avatars array MUST contain exactly one entry; empty avatars: [] substitutes a random face per render — always pass one →
• Two-person scenes: primary in avatars, secondary as a reference image in medias →
• avatars and medias are top-level siblings of params — NOT nested under params; wrong nesting rejects →
• prompt is optional; aspect_ratio enum: auto/21:9/16:9/4:3/1:1/3:4/9:16; resolution: 480p/720p/1080p (default 720p) →
• No get_cost: true preflight for MS — verify spend post-hoc via transactions(limit=200) →
• Three MS models exist: marketing_studio_video (this skill), marketing_studio_image, ms_image — call ms_image "DTC Ads" with users →
• Three avatar types: preset (~40 in library), uploaded, text-generated →
• TV Spot has a default packshot beat — negate explicitly ("ABSOLUTELY NO PACKSHOT") when unwanted →
• Flowing AND sectioned prompt styles both render — "no section labels" is craft opinion, not an engine constraint; default flowing under 50 words →
• Cannot do: >15s clips, non-human lip-sync, multi-character dialogue, multi-setting single output, free-form hook/setting IDs, >1 avatar →
• Escape hatches when MS can't render it: Wan 2.7, Veo 3.1, Cinema Studio Video 3.0, Seedance 2.0, Kling 3.0 →
• Budget anchor: ~$0.06/credit ≈ ~$9 per video (the more credible of two non-canonical rate samples) →

Marketing Studio is Higgsfield's opinionated short-form ad-video product. Generations are bounded by a 4–15 second per-clip cap [Phase 0: Probe 0.3-b], routed through one of nine named presets that dictate camera language, register, and the available picklist of hooks and settings. It's distinct from Higgsfield's general video models (Soul, Seedance, Cinema Studio) — those let you write whatever you want; Marketing Studio enforces ad-format conventions per preset and ships with a built-in library of preset hooks, settings, and avatars.

This sub-skill covers Marketing Studio video. The two image-side Marketing Studio surfaces (marketing_studio_image and ms_image / "DTC Ads") are named in §2 but covered in the companion cross-surface-workflow.md.

1. What Marketing Studio is

Marketing Studio is a routing layer on top of generate_video for short-form ad creative. You provide a product image (required) plus optional avatar, location, and additional-asset references. You pick one of nine presets. You optionally write a prompt. Marketing Studio handles the rest — camera language, pacing, hook execution, setting context — per the preset's built-in conventions.

The opinionated thing about Marketing Studio is that the preset choice is load-bearing. Pick Hyper Motion and you get kinetic product hero shots with an auto pack-shot at the end [SRT-2:00:13:33]. Pick TV Spot and you get composed cinematic narrative — sometimes from a five-word prompt when the asset references carry the weight [SRT-3:00:18:30; PDF items 12, 13]. Pick Wild Card and you get permission to break ad-grammar conventions [director:L113–L117; PDF items 10, 11]. The presets aren't decoration; they choose the production rules.

When Marketing Studio isn't the right tool:

• If your video needs to run longer than 15 seconds — Cinema Studio (higgsfield-cinema) covers narrative length, or use one of the escape-hatch models in §9
• If you need character-consistent multi-shot video with lip-sync — Wan 2.7 or Veo 3.1 are designed for sync (per §9 escape hatches)
• If you want free-form video without ad framing — Seedance (higgsfield-seedance), Soul (higgsfield-soul), or Cinema Studio give you the open canvas Marketing Studio's preset routing closes off

2. Two access surfaces — and three MS models

You reach Marketing Studio two ways:

Both surfaces share the same preset / hook / setting / avatar taxonomy. The web UI is what's demonstrated in SRT-2 and SRT-3; the MCP is what higgsfield-content-factory orchestrates programmatically.

Three Marketing Studio models exist — this sub-skill covers the video one

[Phase 0: Probe 0.3-a] surfaced that "Marketing Studio" is three models, not one — source-corpus reconciliation #11:

Naming note [from show_marketing_studio MCP tool description]: when discussing ms_image with the user, refer to it as "DTC Ads" — that's the user-facing brand name. The model ID ms_image is internal nomenclature.

This sub-skill scopes to marketing_studio_video. References to "Marketing Studio" below mean the video model unless the surrounding text says otherwise.

3. The 9 presets

The nine presets are the routing layer. Pick one, and Marketing Studio applies its built-in production rules.

Slug naming is canonical; routing mode lives on a different MCP call

The slug for Pro Virtual Try On is virtual_try_on, not pro_virtual_try_on — non-obvious mapping, verified at [Phase 0: Probe 0.1]. The other eight slugs match their display names predictably.

Preset routing happens via show_marketing_studio.mode, NOT via generate_video.mode. The generate_video schema for marketing_studio_video has no mode parameter at all [Phase 0: Probe 0.3-b]. content-factory L681–L692 documented a "slug mismatch table" claiming generate_video.mode accepts ugc_how_to for Tutorial and product_showcase for Hyper Motion — neither slug exists in the live API [Phase 0: supplementary probe]; the entire premise of that table is wrong. Source-corpus reconciliation #7. The actual routing happens in the show_marketing_studio flow:

1. show_marketing_studio(action='fetch', url='<product URL>', mode='hyper_motion')
   # OR action='create' with uploaded medias[] + mode='hyper_motion'
   # This registers the product AND sets the next-step preset.

2. generate_video(params={
     model: 'marketing_studio_video',
     prompt: '<optional>',
     duration: 8,
     aspect_ratio: '9:16',
     resolution: '720p',
     generate_audio: false
   }, medias=[...], avatars=[...])

The preset chosen in step 1 routes the generation in step 2. No mode parameter on the generate_video call.

4. Hook + setting picklists

Hooks and settings are open-ended picklists you select by UUID. They're available for five of the nine presets, not six.

Key rule [from generate_video MCP tool description — Phase 0 pre-probe finding]: hook + setting picklists are supported only for UGC, Tutorial, Unboxing, Product Review, and UGC Virtual Try On. Pro Virtual Try On is NOT in the family — contradicts content-factory L193–L194 which included it. Source-corpus reconciliation #10.

Picklist contents drift over time. The Phase 0 snapshot at the v3.7.13 Phase 0 probes notes (internal build notes) is correct as of 2026-05-18; for canonical current data, run the live enumeration calls below.

Hooks (9 entries as of 2026-05-18) — visual scene templates, not verbal copy

show_marketing_studio(action='list', type='hook', size=100)

[Phase 0: Probe 0.2a] returned 9 hooks, classified by type field as either stunt (4) or subtle (5):

• Stunt: Product Hit, Random Object Mic, Blizzard, Product Dodge
• Subtle: Spicy, Interview, Product Crash, Camera Bump, Epic Fail

Each hook ships with a descriptive prompt field that explains the visual concept. Example — "Interview": "Interviewer asks a second stranger a question based entirely on the first stranger's random answer; confusion builds until the second person naturally notices the product (Erewhon-style aspirational item) and pivots into a casual review." You don't have to use that exact prompt — but the hook's preset structure is what gets baked into the generation.

Pass the hook's UUID as params.hook_id. To skip the hook entirely (still on a hook-eligible preset), omit hook_id.

Settings (14 entries as of 2026-05-18) — environment templates

show_marketing_studio(action='list', type='setting', size=100)

[Phase 0: Probe 0.2b] returned 14 settings, classified by type field:

• Realistic (8): Bedroom, Bathroom, Kitchen, Office, In Car, Street, Gym, Nature
• Unrealistic (6): Airplane Wing, Roofing, Volcano Rim, Tiny Reviewer, Car Roof, Train Surf

Each setting ships with a descriptive prompt field describing the scene. Pass the setting's UUID as params.setting_id.

Live-enumeration discipline

The picklist names listed above are a 2026-05-18 snapshot. New hooks and settings get added; existing ones can be revised. For production work — especially when documenting a campaign or building a multi-shot plan — call the live enumeration first and pull the current UUIDs. Hardcoding UUIDs from a stale snapshot is the failure mode this discipline prevents.

When the user describes a scene that matches a known hook or setting name, you can name the hook / setting in your routing without an upfront live enumeration — Marketing Studio's web UI lets users pick by name from the dropdown [SRT-3:00:09:00, 00:09:40]. The UUIDs only matter when you're hitting the MCP directly.

5. Avatar handling

Marketing Studio takes one avatar per generation. There are three avatar types, and one hard API limit you can't work around without a workaround.

Three avatar types

content-factory L630 documented preset-only avatars (avatars: [{ id, type: "preset" }]); Phase 0 confirms the API accepts type: 'custom' | 'preset' and SRT-2 / SRT-3 demonstrate all three creation paths. Source-corpus reconciliation #3.

Field-name calibration [source-corpus reconciliation #9]: content-factory L612 said filter the avatar list by source: "preset"; the live API returns the field as type: "preset". Adjust the filter recommendation accordingly.

Avatar field constraints — consolidated

Three constraints, all enforced at the API or UI level:

1. The avatars array MUST contain exactly one entry. Marketing Studio rejects two simultaneous avatars at the UI level [SRT-3:00:19:00, verbatim: "technically you can't use two avatars"]. show_marketing_studio.avatars parameter is explicitly documented as max 1 item [Phase 0: pre-probe finding]. Source-corpus reconciliation #5.

2. Empty avatars: [] substitutes a random face per render. content-factory L632–L636 documents this as a silent fallback that kills campaign visual consistency — the same prompt run twice with empty avatars produces two different faces. This claim is source-corpus only; not directly verified in Phase 0 (a multi-generation comparison would be needed to prove randomness). Practical guidance: always pass an avatar.

3. Two-person scenes use the reference-image workaround. When you need a primary character plus a second person, pass the primary in avatars and the secondary as a reference image in medias [SRT-3:00:19:00 demonstrated workaround]. Consistent with director L143's ≤ 2 humans tracked rule.

What this sub-skill does NOT do

This sub-skill names the three avatar types and the field constraints. It does NOT enumerate individual preset avatars by name. Reasons:

• The preset library drifts over time (avatars added / removed by Higgsfield)
• The live enumeration call is the canonical source — show_marketing_studio(action='list', type='avatar', size=100) returns the current list with names, genders, and preview URLs
• Picking by name from the live list is the right pattern for any production work

For reference: the 2026-05-18 snapshot in the Phase 0 log shows 40 preset avatars with a ~24F / ~16M gender ratio. Pick by name from the live list, not from any hardcoded table in this sub-skill.

6. Reference media — 4 slots

Marketing Studio accepts up to four kinds of reference media. The web UI exposes them as four upload slots; the MCP exposes them across two top-level arrays.

avatars is a SEPARATE top-level media slot — not a nested parameter

Critical correction [source-corpus reconciliation #8, Phase 0: Probe 0.3-b]: content-factory L630 documented avatars as params: { avatars: [...] } — nested inside params. The schema shows avatars and medias are TWO DISTINCT top-level media arrays declared under the model's medias schema entry.

The actual call shape is:

generate_video(
  params={
    model: 'marketing_studio_video',
    prompt: '<optional>',
    duration: 8,
    aspect_ratio: '9:16',
    resolution: '720p',
    generate_audio: false,
    hook_id: '<uuid>',         # UGC family only
    setting_id: '<uuid>',      # UGC family only
  },
  avatars=[{id: '<uuid>', type: 'preset'}],   # max 1 — see §5
  medias=[
    {value: '<product uuid or url>', role: 'image'},
    {value: '<location image>', role: 'image'},    # optional
    {value: '<packaging asset>', role: 'image'},   # optional
  ]
)

avatars and medias are sibling parameters at the top level of generate_video, NOT nested under params. Get this wrong and the call shape rejects.

Product registration — two paths

Both register the product as a reusable entity you can pass by UUID:

• From URL scan: show_marketing_studio(action='fetch', url='<product page URL>') — server scans the URL, extracts images, registers the product. Demonstrated live at [SRT-2:00:08:50, verbatim: "paste in the link and create the product automatically"]
• From uploaded media: show_marketing_studio(action='create', type='product', medias=[...]) — pass uploaded image media; server registers the product with auto-derived title

After registration, the product UUID is what you pass in subsequent generate_video calls. The user's library is enumerable via show_marketing_studio(action='list', type='product').

7. Generation parameters

Full schema for marketing_studio_video from [Phase 0: Probe 0.3-b]:

Top-level media arrays (separate from params — see §6):

What is NOT in the schema

Three notable absences:

• No mode parameter. Preset routing happens via show_marketing_studio.mode (§3), not via generate_video.mode.
• No get_cost: true support. Marketing Studio models do not support preflight cost estimation [from generate_video.params.get_cost MCP tool description — Phase 0 pre-probe finding]. Consistent with the existing higgsfield-stack documentation. Verify spend post-hoc via transactions(limit=200) (§12).
• No multi-character avatar slot. One avatar max — see §5 for the reference-image workaround.

8. Output prompt style — flowing OR sectioned (NOT a hard rule)

Marketing Studio accepts both prompt styles. Pick the one that fits your concept; don't enforce either as a universal rule.

Source-corpus reconciliation #6: director L213 declared "One continuous flowing paragraph — no section labels, no tags, no headers" as a hard output rule. PDF item 10 (Wild Card — Levitation in Clouds), authored by the same person, uses an explicitly sectioned structure with Style & Mood:, Dynamic Description:, Static Description: headers plus quality suffixes and negative inclusions appended. The "no section labels" rule is craft opinion about elegant output, NOT an MS engine constraint. Both styles render correctly.

When flowing paragraph fits

• Short, concept-driven prompts where the asset references carry the weight. PDF item 8's Hyper Motion logo treatment is literally five words: "logo creating with comet tail." No sections needed.
• Five-word TV Spot prompts when the avatar and location assets supply the rest [SRT-3:00:18:30].
• Vanilla UGC where the hook + setting picklist values already structure the generation and the prompt just adds tone.

Flowing paragraph reads like production prose — easier to skim, harder to over-engineer.

When sectioned structure fits

• Narrative-arc prompts with multiple beats. PDF item 6 (Pro Virtual Try-On — Liquid Scan Transition) uses time-coded segments: 0–3s: FULL BODY SHOT…, 4–7s: The glasses reach maximum size…, 8–10s: THE MOMENT the glasses touch her face — LIQUID SCAN TRANSITION…, 11–15s: MEDIUM CLOSE-UP…. The time-coding makes the beat structure legible to the model.
• Hyper-detailed Wild Card concepts where you want explicit Style / Dynamic / Static separation. PDF item 10's Levitation in Clouds is the canonical example: Style & Mood: (dreamlike high-fashion film, Hollywood wire-work VFX, anamorphic 28mm), Dynamic Description: (multi-paragraph action narrative), Static Description: (single-frame pose), then quality suffixes ("4K, Ultra HD, Rich details, Sharp clarity, Cinematic texture") and negative inclusions ("No ghosting, No flickering, No text overlays").
• Multi-clip campaigns where one prompt has to render coherently across 5 separate generations. PDF item 7 (UGC Try-On — 5 Clips Indoor Room) uses ## CLIP 1 — 4s, ## CLIP 2 — 4s — JACKET, etc.

Picking between them

Default to flowing paragraph for prompts under 50 words. Use sectioned structure when the prompt has discrete beats (time-coded narrative, multi-clip campaign, explicit Style / Dynamic / Static separation). Don't impose section labels on a prompt that doesn't need them — but don't refuse to use them on a prompt where the structure actually helps.

9. What Marketing Studio cannot do

Six things Marketing Studio is not designed for. Trying to force these typically produces broken output:

1. Clips longer than 15 seconds. Hard cap at the API level [Phase 0: Probe 0.3-b]. For longer narrative, design as a 2-clip or 5-clip sequence (each clip ≤15s) and edit externally — PDF item 7 demonstrates the 5-clip pattern.
2. Lip-synced dialogue from non-human characters. Talking fruit, talking products, etc. — the model doesn't reliably sync audio to non-face geometry.
3. Multi-character coordinated dialogue with consistent identities across cuts. The one-avatar API limit (§5) prevents this; even with the reference-image workaround, identity drift across multiple generations is high.
4. Single output with multiple settings, split-screen, or day-X-vs-day-Y diary structure. One setting per generation. For multi-setting narrative, split into separate clips per the multi-clip pattern. PDF item 11 (Wild Card — Jacket Portal Cuts) is the workaround when six locations must appear in one 15s clip — uses portal-cut transitions instead of a multi-setting parameter.
5. Free-form hook_id or setting_id not in the picklist. Picklists are closed — you can only pass UUIDs returned by show_marketing_studio(action='list', type='hook'|'setting'). To skip the hook, omit hook_id.
6. More than one avatar per generation. One avatar slot, enforced. Workaround: primary in avatars, secondary as reference image in medias (§5).

Escape-hatch models for what MS can't render

When an idea genuinely needs lip-sync, longer narrative, or multi-shot continuity, route to one of these instead of forcing MS [content-factory L260–L264]:

• Wan 2.7 — synchronized audio + character-consistent video
• Veo 3.1 — ultra-realistic cinematic, audio-friendly
• Cinema Studio Video 3.0 — most advanced cinema-grade
• Seedance 2.0 — reference-driven, multi-SKU, consistent identity
• Kling 3.0 — multi-shot, audio sync, motion transfer

Use the escape hatch sparingly — Marketing Studio's opinionated structure is the value, and reaching for an escape model means losing the preset routing.

10. Worked examples — production reference library

Eight production-grade Marketing Studio prompts, sourced from PRODUCT GENERATION PROMPTS EXAMPLES.pdf items 6–13. These are the raw prompts Adil used to generate the HIGGS fictional clothing brand demonstrated in the SRT-2 video. Maximum credibility — paired prompt + video evidence for each.

Each example is cited by PDF item number plus the SRT-2 timestamp where the resulting video appears on camera.

10.1 — Pro Virtual Try-On — Liquid Scan Transition

Source: PDF item 6, paired SRT-2 demo at ~00:11:30. Sectioned time-coded structure across 4 beats within a single 15s clip.

all HIGGS items including: white jersey t-shirt, neon yellow puffer jacket, black
track pants with neon green stripe, grey sneakers with neon yellow laces, and
black-to-neon-yellow gradient sunglasses. the female model — use her exact face,
hair, and body throughout the entire video without any changes.

CAMERA: COMPLETELY STATIC throughout the entire video. Fixed position, fixed
angle, zero movement. Only subjects and objects move within the frame.

Location part 1: Urban pedestrian overpass bridge. Glass-panel ceiling above
letting in diffused grey daylight. Metal railings on both sides, concrete floor,
city street with traffic visible far below. Overcast natural lighting.

Location part 2: Dark underground parking garage. Single harsh overhead lamp,
tight pool of light on concrete floor, deep shadows, concrete pillars in background.

0–3s: FULL BODY SHOT — static camera, low angle pointing upward. Female
model stands centered on the overpass in her own clothes, relaxed stance. She
raises the HIGGS sunglasses pinched between two fingers, cocks her wrist back,
then THROWS the glasses forward toward the camera with a sharp snap. The
sunglasses fly directly at the static lens in full 3D — tumbling and spinning in slow
motion, growing larger frame by frame until they nearly fill the screen. Glasses
stay fully intact. Neon yellow frame catches daylight. Model stands still in the
blurred background. CAMERA DOES NOT MOVE.

4–7s: The glasses reach maximum size, SLOW DOWN, STOP mid-air, then
REVERSE on their own — flying back toward the model spinning. As they travel
back the model shifts her pose dynamically — she leans her head slightly forward
to meet the glasses, lifts her chin, squares her shoulders with attitude. The
glasses arc back and land perfectly onto her face. CAMERA STATIC.

8–10s: THE MOMENT the glasses touch her face — LIQUID SCAN TRANSITION: a
wave of liquid mercury ripples outward from her face across the entire frame, like
a 3D scan pulse sweeping through space. The overpass environment MELTS and
DISSOLVES in 3D depth — the bridge, railings, ceiling peel away in layers like
liquid geometry dissolving into the scan wave. The dark parking garage
MATERIALIZES through the liquid scan, environment rebuilding itself in 3D layers
from background to foreground as the wave passes. Simultaneously the full
HIGGS outfit liquifies onto her body: white jersey t-shirt, neon yellow puffer jacket,
black track pants with neon green stripe, grey sneakers with neon yellow laces —
all five items flow onto her in one fluid wave synced with the environmental scan.
The model strikes a new powerful pose as the outfit locks in — weight shifted to
one leg, one hand at side, chin up. CAMERA STATIC.

11–15s: MEDIUM CLOSE-UP — static camera at chest-to-head height. She stands
fully dressed in all HIGGS items in the parking garage. Hard overhead lamp light,
sharp shadows. She adjusts sunglasses with one finger, holds the pose, stares
directly into camera with cold confident expression. No glow, no outlines on body.
Cinematic grade — deep blacks, punchy contrast, neon yellow only as real material
color on clothing. 9:16 vertical.

What makes this example work: the time-coded segments give the model discrete beats to render. The Camera Contract opens with a hard rule (COMPLETELY STATIC, CAMERA DOES NOT MOVE) and reinforces it at every segment boundary. The avatar-fidelity directive ("use her exact face, hair, and body throughout the entire video without any changes") is the explicit identity lock content-factory describes.

10.2 — UGC Try-On — 5 Clips Indoor Room

Source: PDF item 7. Multi-clip campaign pattern — five separate generations of 3–4s each, sharing one location + camera contract.

# HIGGS — 5 CLIPS

LOCATION: Indoor room. Large white-framed floor-to-ceiling glass door centered
in background. Dark night outside. Grey linen curtains floor-length on both sides.
Warm hardwood parquet floor. Cool blue backlight from behind the door, soft
frontal key light on face. Cold blue-teal cinematic color grade.
CAMERA: STATIC, FULL BODY SHOT, eye level. Same angle all 5 clips, never moves.
IMPORTANT: Female model has NO bag, NO backpack, NO purse on her at any
point in any clip. Hands are empty unless specified.

---

## CLIP 1 — 4s
Female model stands centered in her own clothes. No bag, no backpack. She
reaches off-frame, grabs the white HIGGS jersey t-shirt, pulls it over her head and
on — adjusts it, smooths it down. Looks into camera. No effects. 9:16 vertical.

---

## CLIP 2 — 4s — JACKET
Female model stands centered, already wearing the white HIGGS t-shirt. No bag,
no backpack. She reaches off-frame, grabs the HIGGS puffer jacket, shrugs it on
— one arm then the other — adjusts collar with both hands. Looks into camera.
No effects. 9:16 vertical.

---

## CLIP 3 — 4s — PANTS + SNEAKERS

Female model stands centered, already wearing HIGGS t-shirt and jacket. No bag,
no backpack. Already wearing HIGGS track pants with neon green side stripe and
grey HIGGS sneakers — changed off-screen. She runs one hand down the side
stripe, lifts one foot and stomps on the parquet once. Looks into camera. No
effects. 9:16 vertical.

---

## CLIP 4 — 3s — SUNGLASSES
Female model stands centered, fully dressed in HIGGS outfit. No bag, no
backpack. She holds the HIGGS sunglasses between two fingers, slides them onto
her face in one smooth motion. Looks into camera — cold confident expression.
No effects. 9:16 vertical.

---

## CLIP 5 — 3s — FINAL
Female model from stands centered, complete HIGGS outfit, sunglasses on. No
bag, no backpack. Arms relaxed at sides. She squares her shoulders, looks
directly into camera. Says: "That's it." Holds the pose cold and still. 9:16 vertical.

What makes this example work: the shared LOCATION and CAMERA blocks at the top apply to all 5 clips — the model gets a consistent rendering context. Each clip is named (CLIP N — Xs — DESCRIPTOR) so the editor can sequence them. The "no bag, no backpack" line is a negative constraint repeated in every clip — common pattern when the model has been adding incidental props.

10.3 — Logo — Hyper Motion

Source: PDF item 8. Verbatim Adil prompt from SRT-2:00:15:10. The five-word minimum.

logo creating with comet tail.

What makes this example work: Hyper Motion runs on extremely short prompts when an asset reference carries the load — the preset supplies the kinetic structure, so the prompt only has to name the concept. Output (per SRT-2:00:15:35): outer space scene, comet trail, logo materializes at the end with the brand name. No section labels, no quality suffixes, no negative inclusions — and it works.

10.4 — Pro Virtual Try-On — Skateboard Outfit Change

Source: PDF item 9, paired SRT-2 demo at 00:18:00–00:19:00. Long continuous-take prompt with items appearing in air.

The character — dark hair, grey tank top, oversized dark red baggy jeans,
crossbody tactical bag, red bandana on wrist, red paint mark on nose — rides a
skateboard continuously from left to right across the frame. Smooth push, steady
cruise, never stopping. Camera stays strictly side-on the entire time, drifting
smoothly alongside him at constant speed and height, never ahead, never behind,
never rotating. One continuous uncut shot.

As he skates, individual clothing items from the HIGGS collection levitate in the air
ahead of him — floating still, at body height, positioned exactly where they would
sit on a person — waiting. He skates straight through each one without slowing.
The moment he passes through a levitating item it is instantly on his body,
perfectly fitted, as if it was always there. No animation, no effect, no flash — it
simply appears on him as he exits the other side.

Item 1 — white and black HIGGS football jersey: floating at chest height in open air,
he skates through it, it is on him.
Item 2 — black cargo pants with neon green stripe: floating at waist height, legs
hanging naturally, he skates through it, it is on him.
Item 3 — neon yellow-black puffer jacket: floating with sleeves open wide, he
skates through it, it is on him.
Item 4 — grey and neon yellow chunky sneakers: floating at board level, pair side
by side, he skates through them, they are on his feet.
Item 5 — black wraparound sunglasses with neon yellow frame: floating at face
height, he skates through them, they are on his face.

After the last item he is fully dressed in the complete HIGGS collection. He does
not slow down. He pushes once more and keeps skating. Camera drifts to a halt.
He disappears into the distance still rolling. Frame holds on the empty space
where he just was. Silence.

Background: a single continuous urban environment scrolling — raw concrete wall,
cracked asphalt ground with skate marks, hard midday sun casting a sharp
shadow alongside him the entire time. Neon yellow-green appears on wall
markings and signage as he passes.

Sound: continuous skateboard wheel roll on asphalt throughout, occasional push
foot scrape. Each item landing produces no sound. No music. City ambience, wind,
distant traffic. Style: photorealistic cinematic, 2.39:1 Cinemascope, desaturated
warm daylight, neon yellow-green as the only saturated color, streetwear editorial,
identical grain throughout, strict side-on camera drift, one unbroken take.

What makes this example work: the Camera Contract is precise — "strictly side-on… drifting smoothly alongside him at constant speed and height, never ahead, never behind, never rotating." The five items are enumerated separately with location specifications. The Style block at the end ties it together (2.39:1 Cinemascope, desaturated warm daylight). Continuous-take discipline runs through every sentence.

10.5 — Wild Card — Levitation in Clouds

Source: PDF item 10. Canonical example of sectioned-prompt structure (Style & Mood / Dynamic Description / Static Description).

# LEVITATION IN CLOUDS — ONE SHOT HANDHELD

Style & Mood: Dreamlike high-fashion film. Practical Hollywood VFX — wire-work
levitation 

…(truncated)