osidemedia / higgsfield-motion

Higgsfield Named Motion Presets

Higgsfield has 100+ named motion presets. Reference them by exact name in your prompt
and the platform applies the preset's signature effect. Think of these as VFX macros —
one name triggers a complete pre-built visual effect.

Scope caveat `[FIELD — 13-project community harvest + live
preset-catalog pull, both 2026-07-18]`: motion presets are the
viral-effects / social product, not the film grammar. None of the 13
harvested community film/ad productions used a single motion preset —
serious film work free-prompts its camera through Cinema Studio /
Seedance block briefs (../higgsfield-camera/SKILL.md,
../higgsfield-seedance/SKILL.md). Separately, the live catalog pull
showed ~100 unique preset names expanded into ~1,900 per-model variants
(kling / higgsfield / wan / minimax / seedance families with baked
params); category weights at pull time skewed to effects/viral/vfx/ugc. Route preset requests here as ever — but don't
steer a filmmaking request into presets when free-prompted camera language
is the production-proven path.

How to use in a prompt:

[Your scene description.] Apply the [Preset Name] preset.

Transformation & Body Effects

Elemental Effects

Transitions

Action & Motion Effects

Surreal / Artistic

Nature / Environment

Horror / Dark

Social / Fun

How to Request a Preset in a Prompt

A woman stands at the edge of a cliff at sunset, hair flowing in the wind.
She raises her hands slowly as energy builds around her.
Camera: Crane Up revealing the ocean below.
Style: Cinematic, golden hour, 16:9.
Apply the Plasma Explosion preset at the moment her hands reach full extension.

Two fighters face each other in a rain-soaked alley.
Camera: Bullet Time as the first punch lands.
Style: Anamorphic, desaturated, high contrast.
Apply the Flame Transition preset to cut to the next scene.

Negative constraints: For motion/VFX artifacts (preset not visible, VFX looks cartoonish,
jittery motion) and their prevention phrases, see ../shared/negative-constraints.md —
Body/Motion Artifacts and Temporal/Consistency Artifacts sections.

Intent-First Choreography (Cinema Studio 3.0 / Seedance 2.0 Best Practices)

These principles apply to Cinema Studio 3.0's generation engine (Business/Team plan only) and complement the motion preset library above.

Intent over Timestamps

Describe the INTENT and CONSEQUENCE of action, not precise frame-by-frame choreography. Let the AI director handle interpolation between key moments.

Wrong: At 0s: character raises right arm. At 1s: arm reaches 45 degrees. At 2s: arm fully extended. At 3s: energy ball forms.

Right: The character raises her arm slowly, energy gathering in her palm until it erupts in a blinding flash. Sparks scatter across the floor.

@Video Reference as Primary Method

For all action and fight scenes, @Video reference is the most reliable approach:

1. Upload a reference clip showing the movement you want
2. Describe the high-level action in 1–3 sentences
3. Add degree adverbs (violently, gracefully, explosively) and physics consequences (dust erupts, sparks fly)
4. Tell the model to reference the clip

Reference @Video1 for movement choreography.
A swordsman draws his blade explosively, cutting through three bamboo stalks in rapid succession.
Stalks topple, leaves scatter, blade gleams in the sunlight.

Capability boundary: For what @Video reference reads reliably vs. less reliably
across world / texture / camera character / production feel — and what it can't do at
all — see ../higgsfield-camera/SKILL.md § Video Reference — What It Reads, and What
It Can't. The action-specific guidance above remains the primary method for fight
and chase scenes.

One Action Per Shot Rule

For text-only action (no video reference), do not chain multiple distinct actions into one sentence. Break them down per shot:

Wrong: She jumps over the wall, rolls on landing, draws her weapon, and fires three shots while running.

Right (as separate Cinema Studio 3.0 shots):

• Shot 1: She leaps over the wall, coat flaring. Landing hard, she rolls forward.
• Shot 2: Rising from the roll, she draws her weapon in one fluid motion.
• Shot 3: Running at full sprint, she fires three shots. Muzzle flash lights up the alley.

Beat Density Diagnostic

If output is blurry, jittery, or morphing: too many actions packed into too short a duration.

The model handles 1–2 distinct action beats every 5 seconds. If your prompt has 4+ actions in a 5-second window, the model can't resolve them all and produces artifacts.

Fix: Split dense action across multiple shots, or extend the duration to give each beat more time.

Related skills

• higgsfield-camera — Camera controls (combine with motion presets)
• higgsfield-style — Visual styles (pair with motion effects)
• higgsfield-mixed-media — Artistic preset overlays (different from motion presets)
• higgsfield-recipes — Genre templates referencing specific presets
• templates/ — Annotated templates showing preset usage in context

Kling 3.0 Motion Control — When and How to Run It

Important: This is NOT a named motion preset. Everything above in this
sub-skill is the VFX preset library — Explosion, Werewolf, Air Bending, and so
on — where the preset name triggers a pre-built effect. Kling 3.0 Motion Control
is a different system entirely: a reference-video motion transfer workflow. You
upload a character image plus a short clip of the motion you want, and the model
transfers that motion onto the character. The prompt describes the world around
the motion — not the motion itself.

When to Choose Motion Control

Reach for Motion Control when the generation needs directed, repeatable motion
rather than a loose interpretation. Typical cases: dance choreography, sports or
athletic beats, acting or reaction beats, precise gestures for a talking head,
and transferring the same motion pattern across different characters (e.g., one
captured performance applied to multiple hero images). It's also useful as a
scene prototyping step before a heavier Kling 3.0 assembly — you can lock the
motion first, then build the surrounding scene around it.

The 8-Step Higgsfield Workflow

Run Motion Control through the Higgsfield UI in this order:

1. Open the Video tab in Higgsfield.
2. Choose Kling Motion Control 3.0 from the model selector.

3. Upload the motion reference video — the short clip carrying the
   performance you want transferred.

4. Upload the character image — face and body should both be readable; this
   is what the motion gets mapped onto.

5. Pick the output resolution — 720p for faster iteration, 1080p for final
   delivery.

6. Set the Scene source — decide whether the environment should come from
   the motion video's background or from the character image's background.

7. Open Advanced Settings — describe the lighting, atmosphere, and
   background treatment in the prompt field here, and select the orientation
   mode (Image Orientation for camera-driven shots with a mostly static body;
   Video Orientation for full-body movement like dance or action).

8. Generate — review the result and iterate on the reference clip first if
   the motion reads wrong, and on the prompt second if the scene reads wrong.

Motion Reference Input Checklist

Input quality determines output quality. If the reference clip is noisy, the
transfer will be noisy — no amount of prompt refinement fixes a bad source.
Before clicking Generate, confirm the reference clip against this pre-flight
checklist:

• [ ] One clear subject in frame. Multi-person clips confuse the motion
  extraction; crop or trim to a single performer.

• [ ] Head and body both visible. If either is cut off, the motion transfer
  loses the connection between facial performance and body motion.

• [ ] Real human motion. Stylized animation, rotoscoped footage, and heavily
  filtered clips transfer unreliably. Natural performance reads best.

• [ ] No cuts, dissolves, or hard camera transitions. The reference must be
  one continuous shot — any cut inside the clip will show up as a jump or
  snap in the generated output.

• [ ] Avoid very fast actions. Movement that reads as a blur in the source
  will fail to transfer cleanly. Slow-moderate speed performs best.

• [ ] Clip duration 3–30 seconds. Below 3s there isn't enough motion data;
  above 30s is outside Motion Control's supported range.

Rule of thumb: if you'd struggle to describe what's happening frame-by-frame
in the reference clip, the model will too. Trim, re-shoot, or re-select the
reference before burning credits on a generation that was always going to fail.

Binding modes, motion source, and the shortfall diagnostic

A few controls and one diagnostic round out the Motion Control workflow:

• Motion source — uploaded clip OR motion library. The motion can come
  from your own uploaded reference clip or from Higgsfield's built-in motion
  library. The library is the faster path when you want a known-good
  performance (a dance, a sports beat) without sourcing your own footage.

• Matches Video vs Matches Image. The transfer binds the result either to
  the motion video or to the character image. Matches Video tracks the
  reference clip's full-body movement and timing; Matches Image keeps the
  result anchored to the character image (better for camera-driven shots where
  the body stays mostly static). This is the same axis as the
  Image-Orientation / Video-Orientation choice in step 7 of the workflow —
  pick the one that matches whether the body or the camera is doing the moving.

• Element binding (Matches Video only). Binding individual elements of the
  performance is available only in Matches Video mode. If you need element-
  level binding, you must be in Matches Video.

• Close-up face input + emotional transitions. A close-up face input
  carries emotional transitions well — reach for it when the beat is an acting
  or reaction moment rather than full-body motion.

• The shortfall diagnostic. If the **output is suddenly shorter than the
  source clip, the motion is usually too fast or too complex** for a clean
  continuous transfer. Slow the reference, simplify the action, or re-trim
  before regenerating — a shorter-than-source output is the model telling you
  the input exceeded what it can transfer cleanly.