OhhO Train — Higgsfield Video Prompt

  • Product: OhhO Train — Turn demonstrations into policies.
  • Category: Intelligence · Primary accent: Cyan #00D4FF
  • Platform: Higgsfield.ai · storyboard / long-video mode · 6 shots × 15 s ≈ 90 s · 16:9 · 1080p+
  • Model pick: Seedance 2.0 (default — 15 s shots, multi-shot continuity, native audio, unlimited-friendly) · Kling 3.0 (per-shot camera control) · Veo 3.1 (scene-extension chains)
  • What it does: The training engine — fine-tune VLA / imitation / RL policies from OhhO Data or in sim, track every run, export to Serve + ONNX for Fleet.

Full film — paste into Higgsfield storyboard mode (6 shots × 15 s ≈ 90 s)

Each SHOT below is one 15-second generation. Keep the STYLE and ACCENT paragraphs at the top of every shot's prompt. In storyboard / long-video mode, let Higgsfield carry continuity between shots; when generating shots one at a time, feed each shot's last frame as the next shot's first frame.

STYLE — OhhO (identical for every film, do not change):
Photorealistic live-action cinematography — a premium robotics brand film shot like an Apple product documentary. Real physical robots and real hardware in a modern robotics lab / light-industrial space: polished concrete floor, dark matte walls, aluminum-extrusion workbenches, neat cable runs, tool walls, server racks and large monitors. All light comes from practical sources inside the space: one cool white key light, cyan LED strip accents (hex #00D4FF) and deep violet ambient fill (hex #7C3AED), with thin atmospheric haze catching the beams. Cinema-camera look: full-frame sensor, 35mm and 50mm prime lenses, shallow depth of field, filmic highlight roll-off, subtle natural film grain. The camera moves only on a motorized slider, gimbal or jib — one slow deliberate move per shot, no shake. Materials must read as physically real: brushed aluminum, anodized black metal, rubber wheels, braided cables, glass screens with true reflections and fingerprint-level detail. Every monitor or tablet in frame shows only a soft, out-of-focus dark dashboard — never readable text. Mood: calm, confident, precise, quietly futuristic. Color grade: cool and high-contrast, near-black shadows with neon cyan and violet practicals, never oversaturated.
NEGATIVE: no CGI look, no 3-D render or motion-graphics aesthetic, no floating holograms, no wireframes, no cartoon or anime, no recognizable human faces (hands, silhouettes and out-of-focus figures are allowed), no readable on-screen or printed text, no company logos or watermarks, no warm sunny tones, no oversaturation, no cuts inside a shot, no shaky handheld camera, no heavy lens flares.

PRIMARY ACCENT for this film: cyan (hex #00D4FF) — the dominant practical lighting in every shot is cyan (bench strips, status rings, screen spill); violet (hex #7C3AED) appears only as a faint ambient fill deep in the background.

SHOT 1/6 — 15 s — Ignition: A GPU workstation corner at night: fans spin up, intake glow brightens, cyan practicals rim the tower. A slow push-in as the machine leans into load, haze curling through the light.

SHOT 2/6 — 15 s — Before: In the lab, static frame: the robot attempts a pick and fumbles it — misses the grasp, nudges the object, tries again clumsily. Honest, unglamorous, real.

SHOT 3/6 — 15 s — The run: A defocused monitor traces a slowly descending curve; GPU LEDs pulse underneath. In the bokeh behind, a wall clock's hands sweep visibly fast — hours compressing inside one slow slider move.

SHOT 4/6 — 15 s — After: Same table, same object, identical framing to the earlier attempt: the robot now executes the pick smoothly, first try — reach, grasp, lift, place. The difference is the whole story.

SHOT 5/6 — 15 s — Export: Macro. Hands pull a small storage module from the workstation and click it into the robot's port. The robot's LED ripples cyan head-to-tail as the policy lands.

SHOT 6/6 — 15 s — Hero: The robot repeats the perfected pick in a calm loop, the workstation glowing soft behind. A slow fifteen-degree orbit. Upper third clean dark negative space for a title. Hold.

Single-clip fallback — one 15 s generation

For a quick single-clip version (or models capped below 15 s), paste the STYLE and ACCENT paragraphs above, then this condensed scene:

SCENE — OhhO Train (one continuous 15 s shot): A robot fumbles a pick; a GPU tower spins up through the night, a defocused loss curve falling as a clock's hands sweep fast in the bokeh; then — same table, same framing — the robot nails the pick first try, a hand clicks the trained module into its port, and a slow orbit holds on the perfected loop with clean space up top.

Going longer (unlimited plan)

  • 15 s is the per-shot ceiling on Higgsfield's top video models — length comes from sequencing shots, not from one long generation. The 6-shot film above is the baseline ≈90 s cut.
  • Every shot ends on a deliberate hold, so Higgsfield's AI Video Extender can stretch any shot past 15 s (or loop the hero shot indefinitely) without a visible seam.
  • To add beats, duplicate a mid-film shot, change only the action sentence, and chain it via last-frame → first-frame. The STYLE block keeps it in the family.
  • On Veo 3.1, use scene extension from the hero shot for arbitrarily long ambient loops of the end frame.

Text to add in post (don't let the model render it)

  • Wordmark: OhhO (first & last O cyan #00D4FF, middle hh white)
  • Product name: OhhO Train · Tagline: Turn demonstrations into policies.
  • CTA: Open training console → · end on The Open Robotics Platform

Platform cheat

Higgsfield → Create → Video → storyboard / long-video mode. Model: Seedance 2.0 (or your plan's unlimited model). 16:9 · 1080p · 15 s per shot. Reuse the OhhO photoreal keyframe still as the first frame of SHOT 1; keep one fixed seed across all films where the model exposes it.

All docs