Ray 3.2 Controls & Workflows

Core Philosophy: How to Approach the Controls

Modify Video controls tell the model how closely to track different parts of your source footage across three independent control families:

• Motion: How movement carries over across the scene.

• Structure: How tightly shapes, edges, and forms are held.

• Characters: How facial expressions, bodies, and poses are captured.

The Golden Rule

Start simple with the lightest setup possible. See what is missing, then enable or raise a control only to fix that specific gap. Unnecessary controls pull the model's focus away from your target.

Key Takeaway: When using multiple keyframes, your control settings often matter more than your prompt. Prompts set the target look, but control settings hold the model to your footage.

1. Motion Controls (Scale: Off, 1 to 9)

Motion captures movement density across the entire scene independently of shapes or edges.

Reading the Slider:

• Low (1): Broad Strokes. Captures large, macro movements. In practice, this mostly locks onto camera passes because fewer moving elements are tracked. Use this when you want the camera path locked while leaving subjects free to move.

• High (9): Dense Capture. Tracks subtle, fast, or fleeting movements across the frame (such as individual limb movements or passing objects).

Key Behavior & Pro Tips:

• Overlapping action drops out: When two moving subjects cross paths, fine details in the overlapping area can drop out. The main direction carries through, but subtle crossing details may be lost.

• Scale preserves detail: Large subjects filling the frame hold motion detail much better than small, fast objects near the edges.

• Handles blur and lighting well: Works reliably through motion blur. Motion responds best to color and light gradients rather than flat shapes (shading or 3D lighting gives much stronger control than flat, single color blobs).

2. Structure Controls (Scale: Off, 1 to 9)

Structure locks the physical shapes and geometry of your original footage.

Important Direction Change: Higher numbers mean stronger adherence to your source footage.

Reading the Slider:

• Low (1 to 3): Soft & Loose. The model works from broad, soft placement cues with minimal camera tracking. Use this when heavily transforming shapes while keeping general placement.

• Middle (5): Balanced. Preserves real subject geometry while offering enough flexibility to transform the aesthetic style.

• High (8): Subject Lock ("Green Screen" Effect). Firmly locks the foreground subject's shapes while letting background elements more than a few feet back drop off and transform freely. Great for close ups.

• Top (9): Whole Frame Lock. Locks every shape, angle, and edge across the entire frame regardless of distance.

Pro Tips:

• Override with keyframes: Even at a setting of 9, adding custom edited keyframes in specific areas will override the structural hold in those regions.

• Facial detail nuance: For close up human faces, try 8 first. Setting 9 can occasionally suppress fine facial details compared to setting 8.

3. Character Controls (Faces, Bodies & Poses)

Character controls capture human and performance motion independently of scene Structure or Motion.

Face (Expression Tracking)

Reads expressions as proportional movements (such as mouth lift or eye squints) and maps them onto the output character, even if the new character looks entirely different.

• Best use: Dialogue, emotional reactions, and close ups.

• Blind spots: Less sensitive to subtle forehead or upper cheek wrinkles. Broad expressions transfer most effectively.

• Creature transformations: For close ups where a human becomes a creature, use low or no Structure + Face Onso the expression carries without locking the original human head shape.

Bodies & Poses (Skeletal Performance)

• Bodies + Poses (On): Captures complete body and limb movement. Superb on its own when Motion and Structure are Off, as it isolates the person without dragging in background clutter or furniture.

• Poses (Detailed Tracking): Carries every precise movement and limb position. The model automatically smooths over small jitters.

• Blocking (Pre-vis Tracking): A looser pose mode that captures general body posture rather than every exact limb placement. Ideal when your source footage contains rough stand in puppeteering.

Combining Controls: Common Recipes

Legacy Mode Translations (Adhere / Flex / Reimagine)

If you are accustomed to older Luma controls, use these reference points:

• Structure replaces the old Adherence scale, but the direction is reversed (higher numbers now mean tighter hold).

• Old Reimagine 2 is equivalent to Structure Off, Motion Off, Bodies + Poses On, and Face On.

• Old Adhere (Max Adherence) is close to Bodies + Poses On with Structure set to 9.

• Motion is a completely new system with no equivalent in older versions.

Quick Reference Summary

• Motion (Off, 1 to 9): Use when reproducing movement paths, camera passes, or action dynamics matters more than holding exact shapes.

• Structure (Off, 1 to 9): Use when keeping real world locations, silhouettes, facial shapes, or product geometry matters.

• Bodies + Poses (On / Off): Use when a person's physical body movement is the core performance driving the scene.

• Face (On / Off): Use to transfer facial expressions onto the same actor or a brand new character.