Trial task · Higgsfield AI Generation Specialist · Addx Studio
45 seconds, in a day and a half.
Your brief, delivered: a spot for Arbi. No script, no new concept — I took your reference video as the template and condensed it to 45 seconds.
Full film · with sound · youtu.be/l6006TY5rHU
Watch on YouTube ↗- Role
- AI Generation Specialist · Higgsfield
- Pipeline
- Seedream → Seedance / Kling → Resolve
- Sound
- Voice and music from ElevenLabs
- Duration
- A day and a half, solo
01 · Approach
Your own reference, condensed.
01
Reference, not concept
Your reference video was the template. I read it beat by beat and condensed it to 45 seconds.
02
Middle act: your sequence
The motion-graphics passage comes from your video — re-cut, placed on my timing and integrated into the film.
03
Cold, then warm
Corporate cold and centred framing while the problem is on screen. Daylight and air as soon as the product answers.
02 · Prompts
Built with a custom Claude skill.
From one beat the skill always writes the same fields: subject, text in the scene, camera, light, physics, diegetic sound, exclusions. A failed generation then means: change one field, leave the rest. Model choice, selection and editing stay with me.
Opening seconds · muted loop
Centered slow macro push-in toward the center of a circular wall clock, accompanied by the mechanical whir of fast-spinning gears and an accelerated, relentless metallic clock-ticking sound. The clock hands (minutes and hours) spin rapidly clockwise in a continuous motion blur. Bold text pops sequentially word by word onto the clock face in a fast, rhythmic cadence: "EVERY" "RECRUITER" "HAS" "THE" "SAME" "8" "HOURS", each word accompanied by a sharp, tactile typewriter keystroke and a muted mechanical shutter snap.
A sharp, violent whip-pan transition tears across the frame with an aggressive, low-end air whoosh and tape-hiss swipe, cutting directly to a locked-off static wide shot facing a wall covered with dozens of identical clocks. The background erupts into an overwhelming, out-of-sync cacophony of overlapping mechanical ticks and spinning clockwork mechanisms.
The sound is written into the prompt, event by event — that's why the edit lands on the beat.
03 · Pipeline
One chain, no handoff losses.
01
Seedream
Every scene starts as a still: composition, palette, text placement. Fixing an image takes minutes, fixing a clip takes a regeneration.
02
Seedance
Beats from cut to cut: text word by word, whip-pans, hard cuts. Keeps type stable on a surface.
03
Kling
Everything that keeps moving without a cut: continuous camera moves, flicker, screen behaviour.
04
ElevenLabs
Voice-over and score, laid against the edit — not the other way round.
05
DaVinci Resolve
Edit, sound design, mix. One grade and one grain pass bind every generation into a single camera.
Everywhere
Higgsfield Cinema
Everything is generated and iterated inside Cinema. Look, optics and lighting language stay consistent across models.
04 · Beats
Every line of text sits on a real surface.
Clock face, word by word
Macro push-in · Seedance
CRT in the clock wall
Locked-off wide · Seedance
Hourglass against the number
Continuous move · Kling
The source, selected on screen
Macro, cursor drags the line · Kling
Short answers
What you'll probably ask first.
- Script or concept?
- Neither. The template was your own reference video, condensed to 45 seconds.
- Where does the middle act come from?
- The motion-graphics sequence from your video, re-cut and integrated into my film.
- How are the prompts written?
- With a custom Claude skill: six fields per shot, diegetic sound included. Judgement and editing stay with me.
- Which model for what?
- Seedream for stills, Seedance for cuts and text events, Kling for continuous motion. Decided per shot.
Available now · remote, CET
Send the next reference and a deadline.
This is how I'd work on your projects from day one. Showreel and six full projects at hervetailliez.de.