Start with a clip that already has useful motion
Ray 3.2 is video-to-video. The better your source timing, camera move, and performance, the easier it is to get a controlled result.
How to Use Ray 3.2 for Controlled Video-to-Video Generation
A practical user guide for turning one source clip into a directed Ray 3.2 result: prompt the final frame, lock important beats with keyframes, tune adherence, and export the version your edit needs.
You can follow the steps with the Ray 3.2 generator open while you read.
Use a source clip up to 20 seconds
Guide the shot with keyframes
Export draft, HDR, or EXR-ready output
Before you start
Ray 3.2 works best when the source clip already has the shot you want to keep
The model can change look, lighting, character, product, or environment. It should not be asked to rescue a source take whose timing, framing, or performance is already wrong.
Decide what should change
Separate the fixed parts from the creative change: keep the motion, preserve the actor, replace the environment, relight the scene, or restyle the whole shot.
Plan the finish before the prompt
Choose whether this run is a draft, a client review, or a final delivery. That choice affects resolution, output format, and how much control you need.
Step-by-step
How to use Ray 3.2 in a real production flow
This is the cleanest path for a first successful run. Move from broad direction to exact control, then export only after the motion works.
Step 1
Upload the source video
Use a clean source clip with the framing, camera motion, and duration you want to preserve. Avoid clips with heavy compression, abrupt cuts, or overlays if you need a clean transformation.
Best for: relighting, product swap, character transformation, camera motion transfer.
Step 2
Write a target-state prompt
Describe the final shot as if it already exists. Instead of asking Ray 3.2 to remove or fix things, define the new frame, lighting, material, character, and environment.
Prompt shape: subject + environment + visual style + lighting + what must stay consistent.
Step 3
Add keyframes where timing matters
Use keyframes for important beats: opening frame, character turn, product reveal, midpoint action, and final pose. Each keyframe reduces guesswork.
Use fewer keyframes for loose exploration, more keyframes for approvals and continuity.
Step 4
Tune Motion and Structure adherence
Raise adherence when the result must stay close to the source take. Lower it when the shot can move into a more stylized direction.
Motion controls movement. Structure controls spatial layout, blocking, and composition.
Step 5
Review, iterate, and export
Run a fast draft first, check whether the shot follows the brief, then export the approved version in the resolution and format your workflow needs.
Use HDR or EXR when the clip needs to keep moving through post-production.
Detailed walkthrough
Use this workflow when the result needs to survive review
The short version is upload, prompt, keyframe, tune, export. The production version is more deliberate: read the source, protect what works, change one layer at a time, and review the full clip before calling it finished.
1. Read the source clip before writing anything
Before you touch the prompt field, watch the source clip twice. On the first pass, identify the useful production facts: camera direction, lens feeling, subject scale, action timing, product position, and the final frame. On the second pass, decide what must remain true after Ray 3.2 transforms the shot. If the camera push-in is the reason the clip works, say that it should remain. If the hand motion sells the product, protect it. If the background is the only weak part, keep the subject and motion language tight while giving the environment room to change.
2. Turn the request into a target state
Many weak generations start with command language: change this, remove that, fix the background, make it cooler. Ray 3.2 usually performs better when you describe the shot as it should look when the work is done. A target-state prompt reads like a short art direction note for a finished frame: a bright coastal product commercial, a warm interior study scene, a snow-covered mountain valley, a chrome character under blue rim light. This gives the model a stable destination instead of a list of edits that may conflict with one another.
3. Add keyframes only where control is worth the cost
Keyframes are not decoration. Use them when a beat needs to land: the opening composition, a midpoint pose, a product reveal, a facial expression, a hand placement, or the final hero frame. A simple restyle may need only a prompt and adherence settings. A client-facing product shot may need several keyframes so the bottle, label, hand, and final composition stay reliable. The rule is practical: if a reviewer would pause the clip at that moment and give feedback, it probably deserves a keyframe.
4. Choose adherence based on what the edit cannot lose
Motion adherence protects timing and movement. Structure adherence protects spatial layout, blocking, and composition. When the source clip already fits an edit, keep these controls higher so the transformed result still drops into the same timeline. When the creative direction is more exploratory, lower them slightly and let Ray 3.2 reinterpret the scene. Avoid changing everything at once. If identity, product position, and camera rhythm all matter, protect those first, then iterate on style, lighting, and environment in later runs.
5. Review the whole clip, not just the prettiest frame
A Ray 3.2 result can look strong in a still frame but fail in motion. Watch the full duration before deciding it is usable. Check whether the first frame connects to the previous shot, whether the midpoint action still reads, whether the final frame lands, and whether any transformed object drifts or changes identity. For production work, export a draft first, make notes at exact timestamps, then rerun with narrower instructions. Move to HDR or EXR only after the motion and continuity are approved.
Video lessons
Read the source clip, then write the Ray 3.2 instruction
These examples use different video problems: motion, product placement, performance, and lighting. The goal is to learn what each source clip is asking you to preserve.
Motion-heavy source clip
This fast amusement-ride shot is useful for learning Motion adherence. The camera and track movement are the point of the clip, so a Ray 3.2 prompt should protect the ride path, speed, and perspective before asking for a new visual world.
Product handling and placement
The beverage pour shows why Structure adherence matters. The bottle, glass, hand, ice, and horizon all create a composition that must stay readable. Use this kind of clip to practice product swaps, label changes, and lighting variations without losing the original staging.
Performance and character continuity
This split comparison shows a performance being carried into a more stylized character result. It is a good reminder to describe both the new character surface and the original performance details that should survive: head turn, hand gesture, expression timing, and framing.
Lighting and mood transformation
The warm interior writing scene is a clean relighting example. It has a clear subject, desk, window, and gesture. For Ray 3.2, this is the kind of source where you can change time of day or color temperature while preserving the calm action and composition.
Prompting
Write the prompt as a finished shot, not a repair instruction
The fastest way to learn how to use Ray 3.2 is to describe the final state with concrete visual details. Name the subject, the environment, the lighting, the motion that must stay, and the exact thing that should change.
Too vague
Make this look better.
Better
A night market product shot with cool neon reflections, realistic wet pavement, the same camera push-in, and the same product position.
Too vague
Change the actor into a robot.
Better
The same performance and body motion, transformed into a polished chrome android with expressive eyes, soft studio rim light, and preserved facial timing.
Too vague
Put it in a different place.
Better
A desert sunrise environment with long warm shadows, the same subject scale, the same camera orbit, and natural atmospheric dust.
Controls
Use adherence and locks to decide how far Ray 3.2 can move from the source
The same source clip can produce a subtle relight or a full stylized transformation. The difference is how tightly you hold motion, structure, and identity.
A simple control rule
Motion adherencetiming
Structure adherencelayout
Character locksidentity
Use this checklist
Use high Motion adherence when timing, gestures, or camera rhythm must match the source.
Use high Structure adherence when composition, blocking, or product placement must stay locked.
Use character locks when identity, pose, body proportion, or facial continuity matters.
Use Reframe after the creative direction is approved, not before you know the shot works.
Avoid these
The common Ray 3.2 mistakes are usually workflow mistakes
If a result feels random, the issue is often the source clip, prompt shape, or control settings. Fix the workflow before blaming the model.
01
Starting from a source clip with the wrong timing.
02
Writing a prompt that describes the editing command instead of the final image.
03
Using too many changes in one run before checking the first direction.
04
Lowering adherence when the shot still needs continuity.
05
Exporting a final before reviewing the motion at full duration.
Final review
Do not export the final until the clip works in motion
A good Ray 3.2 result should not only look impressive as a still. It should behave like a usable shot across the full duration.
The source duration still matches the edit you planned to use.
The first, middle, and final frames all support the same creative direction.
Faces, hands, products, and readable labels do not drift across the shot.
The camera motion feels intentional rather than newly invented by accident.
The output format matches the next step: draft for review, HDR for high dynamic range delivery, EXR for finishing and grading.
FAQ
How to use Ray 3.2 FAQ
Short answers for the questions people usually have before their first Ray 3.2 run.
How do I use Ray 3.2 for the first time?+
Start with a source video that already has useful timing or motion. Then write a target-state prompt, add keyframes for important moments, tune Motion and Structure adherence, and review a draft before exporting anything final.
Is Ray 3.2 text-to-video or video-to-video?+
Ray 3.2 is best understood as a controlled video-to-video workflow. You give it a source clip, then direct how that clip should transform instead of asking the model to invent the whole shot from nothing.
When should I use keyframes?+
Use keyframes when a shot has specific beats that must land correctly, such as a product reveal, character turn, camera move, expression change, or final pose.
What is a good Ray 3.2 prompt?+
A good prompt describes the final state clearly: the subject, environment, style, lighting, and what should remain consistent from the source clip. It reads more like art direction than a list of fixes.
Should I change everything in one generation?+
Usually no. Start with the most important transformation first, such as relighting or environment change, then iterate. Trying to change product, character, camera, lighting, and style all at once makes the result harder to judge.
How do Motion and Structure adherence differ?+
Motion adherence protects timing, gestures, and camera rhythm. Structure adherence protects layout, blocking, subject position, and composition. Raise the one that protects the thing your edit cannot afford to lose.
What should I check before exporting?+
Watch the full clip, not just the best frame. Check the beginning, midpoint, and final frame; look for object drift, identity changes, broken hands, or motion that no longer matches the source take.
When is a draft good enough?+
A draft is good enough when you are still choosing direction. Move to higher-quality output only after the motion, composition, and creative intent are approved. That saves credits and keeps review focused.
Turn the guide into a usable first run
Independent SaaS workflow built on Luma Ray 3.2. Not affiliated with Luma Labs.