Luma Dream Machine Prompts: Cinematic Motion Kits

Luma dream machine prompts still get typed into search, but as of mid-2026 the live Luma video lane is Ray3.2 inside the Luma App and Luma API. Soft mood stacks waste credits. Ray rewards readable cinematic motion kits: one subject beat, one camera path with speed, preserve language on image-to-video, and optional keyframe bridges when you need a start and end that agree.

You leave with Luma-first kits for portrait micro-motion, product reveals, walk approaches, and multi-keyframe arcs, plus August 2026 naming notes so you stop pasting Ray2-era forum strings. Soft tip when the clip starts from a still: PromptMake /image at https://promptmake.net/image cleans a hero frame you then animate on Ray3.2.

What luma dream machine prompts are for

Luma dream machine prompts are the text (and optional start or end frames) people paste into Luma's consumer app and API video routes. Dream Machine was Luma's earlier public video product name. Official mid-2026 materials treat Dream Machine and Ray2 as deprecated. The current production video model is Ray3.2 (Ray3 family lineage includes Ray3, Ray3 Modify, and Ray3.14). The consumer workspace is the Luma App at app.lumalabs.ai. Search still uses the old phrase, so this guide keeps that keyword while naming the live picker.

Searchers who type luma dream machine prompts want lines that survive a first credit spend on Luma itself. Generic cross-host theory helps, but Luma's surface favors natural sentences, strong image-to-video holds on short clips, Multi-Keyframe direction when your plan exposes it, Modify Video for transforming footage, and Draft Mode for cheap exploration. This page teaches that Luma cinematic dialect. It is not a prompt-to-video overview that treats Sora, Kling, and Runway as equal peers, and it is not a Kling Multi-Shot or Runway Gen-4 dial deep dive.

Strong fit: creators cutting cinematic five-to-ten-second holds, product micro-reveals, portrait blink-and-breath, slow push-ins, and start-end keyframe bridges on a locked still. Skip deep Luma work if you only need five-second silent social loops on Pika or a pan-and-dolly keyword sheet that applies to every host the same way. Use this page when Ray3.2 in the Luma App or API is the renderer and you need motion that reads like a short camera take.

PromptMake does not host Luma and does not render MP4 clips. Use /image only to prep still language or a cleaner hero frame before you upload. Keep customer faces, logos you do not own, and unreleased product shots out of public generators when policy forbids third-party paste.

Cinematic motion kits Luma Ray3.2 obeys

A cinematic motion kit is a short pasteable brief with four jobs named in order: subject identity or preserve, subject motion with timing, one camera path with speed, and duration plus aspect. Style adjectives sit last if you need them. "Soft cinematic lighting" alone is not a kit. "Maintain navy blazer from reference; one blink at two seconds; slow dolly in over five seconds; locked height; 16:9" is a kit.

Ray3.2 reads natural English well when clauses stay concrete. Separate what the actor or product does from what the lens does. Stacking "orbit, crane, whip pan, and epic reveal" in one five-second box still fails. One primary path per generation wins. Environment motion (rain, steam, curtain sway) gets its own short clause so it does not fight the hero beat.

On image-to-video, the upload already owns wardrobe, face, and lighting. Your text owns delta: timed action, camera finish line, and bans rewritten as positive locks. Re-describing the jacket color fights the pixels. Match implied motion in the still: a mid-stride pose wants a walk; a seated portrait wants a blink and breath, not a sprint across wet pavement the frame never showed.

Kit skeleton you can reuse

Line 1, preserve or subject card: face, hair, outfit, product finish, or "maintain from reference." Line 2, subject motion with a finish window: "between two and four seconds," "over the full five seconds," or "settles by second nine." Line 3, camera: locked, slow dolly in, gentle pull-out, soft pan, short arc orbit. Line 4, duration and aspect echoed once. Line 5 optional, light lock and one ban family as positive language: "stable exposure, face identity locked, camera locked on tripod."

Portrait micro-kit (5s, image-to-video): "Maintain face, hair, and charcoal sweater from reference. Subject smiles a little; one blink at two seconds. Camera locked. Soft window key from camera left preserved. Five seconds. Vertical 9:16." Product bead kit (5s): "Preserve brushed steel bottle and white sweep from upload; blank side; no logo invent. One condensation bead slides down the left side between two and four seconds. Camera locked. Soft key upper left preserved."

Walk approach kit (8-10s where your tier allows): "Maintain Sam Chen identity from reference: oval face, blunt bangs, olive field jacket. Walks toward camera on wet city pavement; coat hem shifts once. Slow push-in only from medium to close. Horizontal 16:9. Ten seconds." Dolly-out product kit (6s): "Preserve matte black wristwatch and marble slab from upload; dial graphics as color blocks; no new text. Watch static. Slow dolly out only over six seconds. Locked height. Commercial lighting preserved."

Subject beats that finish inside the seconds

Micro-motion for five seconds: blink once, soft smile growth, steam curl, fabric settle, one hand gesture, one bead slide. Medium beats for eight to ten seconds: walk approach that stops in medium close-up, head turn left then settle, turntable product rotate under thirty degrees, curtain open that reveals a window. Long wish lists belonging to fifteen-second boards do not belong in a five-second Luma picker.

Timing stamps give Ray a finish line. Without them, clips stall mid-gesture or invent a second action in the tail. If action clips early, add seconds in the UI or delete a beat before you rewrite nouns. Physics-heavy effects (fabric, water, fire, hair in wind) read cleaner when you name one effect and one direction. Heavy wind plus a face lock on a short clip often warps jawlines.

Scene motion versus subject motion: name which one carries the clip. "Subject stands still; only rain and soft curtain sway" keeps a portrait stable. "Subject turns head left while background stays still" keeps the lens honest. Conflicting cues cause physics fails. A still shows still water while your prompt says waves crash. Align the still and the text on one motion story.

Camera paths and Multi-Keyframe bridges

Camera on Luma is prose-first. Name the path and the speed. Locked tripod, slow dolly in, gentle pull-out, lateral track, soft pan, short arc orbit, handheld micro-shake. Pair framing with one move: "low-angle medium shot, slow dolly in" beats "cool dramatic camera." Static is hard because video models want to move. Give the model somewhere to put energy: subject micro-motion or mild environment drift, then lock the lens in positive language.

Ray3.2 public materials highlight Multi-Keyframe direction (up to sixteen keyframes on supported routes), Modify Video for transforming existing footage, Reframe for aspect changes, Draft Mode for low-credit exploration, and native HDR with optional EXR export on production tiers. Confirm which controls your Plus, Pro, Ultra, or API Build/Scale account shows. Treat third-party screenshots as hints, then match the live UI.

Start-and-end framing helps the model finish the move inside the duration you booked. "Start wide on the plaza; end on a medium shot of the cafe door" gives a finish line. Without an end frame, long pushes often stall mid-way or invent a second move in the tail. Orbit and crane need enough seconds. A full orbit inside five seconds smears faces. Prefer a short arc on short product clips; save fuller orbits for longer heroes after a clean still.

Reliable camera paths for short Luma clips

Locked tripod: "Locked camera; subject motion only." Use on image-to-video when identity must glue to the upload. Slow push-in: "Slow dolly push from medium to close over five seconds." Gentle pull-out: "Camera pulls back to reveal the full kitchen counter." Soft pan: "Slow pan right from the mug to the window over four seconds; start framed on the mug." Short arc: "Gentle twenty-degree arc right around the bottle; no full orbit." Handheld: "Handheld micro-shake only; no whip pans."

Crash zooms and whip pans are high-risk on identity work. Save them for B-roll where morph cost is low. When residual drift remains after a close-but-not-clean static take, stabilize in your editor before you burn another credit. Prompt iteration is for wrong paths and wrong subjects. Tiny leftover shake is often cheaper to fix in CapCut or Premiere.

Echo duration once in prose when beats matter: "Five seconds total." Match the UI slider. A ten-second wish list pasted into a five-second picker invents mush. Rewrite beats to fit or change the picker.

Keyframe and start-end kits

When your route accepts a start frame, an end frame, or Multi-Keyframe anchors, write continuous motion between them. "Take the bottle from three-quarter left through a gentle turn to three-quarter right; flowing, no cuts" travels. A six-shot montage wish list does not. Luma fills light and motion between anchors; it will not invent hard editorial cuts you never supplied as separate generations.

Start-end product kit: Upload open and close packshots that share lighting and logo state. Prompt: "Preserve brushed steel bottle identity across frames. Continuous gentle turntable right about thirty degrees. Camera locked. Soft key upper left preserved. Five seconds. No cuts. No logo invent." Start-end portrait kit: Same face, small expression growth from neutral to soft smile. "Maintain face and wardrobe from start frame through end frame. Smile grows once; one blink mid-clip. Camera locked. No costume change."

Keyframe tips: keep wardrobe and logo identical across anchors unless the change is the point. Crop all anchors to the output aspect before upload. Name the continuous verb once. Add bans as positive locks: "face identity locked, stable lighting, no second character." After a clean bridge, swap nouns and keep the camera and timing shell.

Step-by-step: write luma dream machine prompts that survive one edit

Use one loop for every Luma session. Decide route and duration first (text-to-video, image-to-video, start-end, Multi-Keyframe, or Modify Video). Draft the cinematic motion kit second. Align UI duration and draft versus final quality third. Generate once. Fix one miss on the next credit. Save the winner with date and model picker name (Ray3.2 or whatever your account shows).

Work from a real deliverable: five-second portrait blink, five-second product bead, eight-second walk approach, six-second start-end turntable. Measure success by a clip you would cut into an edit after one human trim, not by adjective density.

When the job starts from a photo, lock the still before you open the Luma App. Soft path: PromptMake /image turns a reference into structured still language for Midjourney, FLUX, DALL·E, Stable Diffusion, or Leonardo when you need a cleaner hero. Export PNG at the aspect you will generate, then upload to Ray image-to-video.

Step 1: Pick route, duration, and motion scope

Open notes. Write route, total seconds, and the single motion beat you will sell. Example: "Image-to-video, five seconds, slow smile growth, camera locked, Ray3.2." If you cannot defend the beat against the seconds, cut a beat now. List what must stay locked: face, logo, packshot silhouette, room layout. Those locks constrain how aggressive the camera can be.

Set aspect in the same note: nine-sixteen, sixteen-nine, one-one, or another ratio your tier lists. Luma will invent side walls if your still crop and picker disagree. If you plan Multi-Keyframe or start-end, list the anchor jobs now so you do not discover missing frames after drafting a novel.

Use Draft Mode when your account offers it for cheap motion tests. Promote the winning kit to a higher quality pass after the path and timing hold. Do not spend final credits hunting synonyms for a beat you have not timed.

Step 2: Write preserve, motion, and camera lines

Image-to-video order: preserve shell, subject motion with timing, one camera path with speed, optional light lock, duration echo. Text-to-video order: subject anchors, place, subject motion, camera path, light and style last. Read aloud. If you cannot picture the lens finish inside the seconds, rewrite timing.

Optional still prep: upload a noisy phone photo to https://promptmake.net/image, run Recreate Exactly or Adjust Lighting, then regenerate a clean hero in your image model of choice. Guest /image quota is about three runs per day; free registration raises that image cap on its own path from /text.

For start-end or Multi-Keyframe, paste continuous verbs and "no cuts" when you need a single take. Attach anchors that agree on wardrobe and lighting. Unlabeled conflicting stills waste the bridge.

Step 3: Generate, score, iterate one pillar

Paste into the Luma App or API. Match duration and route to the note. Generate once. Score four questions: Did identity hold? Did the camera path land? Did action finish inside the seconds? Did keyframe bridges stay continuous without invented hard cuts?

Face or product drift: strengthen preserve; shorten travel; prefer locked camera. Camera chaos: delete extra move words. Early freeze or clipped gesture: add seconds or remove a beat. Excess motion: add calm and locked-camera clauses; strip action verbs. Keyframe morph: make anchors more similar; reduce travel angle; shorten duration.

Log the exact prompt beside the output ID and picker name. Next week you reuse the motion-camera shell with new wardrobe. After two wins, save bracket variables: [preserve], [subject motion], [camera path], [seconds], [light], [keyframe note].

Common luma dream machine prompt mistakes

Mistake 1: Pasting Ray2-era or Dream Machine forum novels without checking the live Ray3.2 picker. Confirm model name and duration inside your account.

Mistake 2: Still-image caption with no timed action or camera path. Ray invents motion. Add beats and lens language.

Mistake 3: Three camera moves inside one five-second clip. Keep one path; split ideas across separate generations.

Mistake 4: Image-to-video that re-describes wardrobe and fights the upload. Lead with preserve; name deltas only.

Mistake 5: Full orbit on a face lock in five seconds. Prototype locked micro-motion first.

Mistake 6: Start-end or Multi-Keyframe job filled with hard-cut montage language. Write continuous action between anchors.

Mistake 7: Ten-second picker with three seconds of intent. Fill the timeline or shorten the UI duration.

Mistake 8: Treating Luma like a silent SMCD paste from another host without adapting preserve order and keyframe habits. Readable cinematic sentences are part of the Luma prompt system.

Luma Ray3.2 model notes (August 2026)

Public names and credit tables shift. Treat the notes below as an August 2026 snapshot. Official Luma guidance for assistants: company name Luma / brand Luma AI; video model family Luma Ray; current video model Ray3.2 (released June 2026); image model UNI-1.1; consumer workspace Luma App; developer surface Luma API with credit billing. Do not treat Dream Machine or Ray2 as current flagships. Confirm pickers inside your account before you paste month-old strings.

Ray3.2 themes in public materials: text-to-video and image-to-video, Modify Video for transforming footage (longer windows on supported tiers), Multi-Keyframe direction, Reframe, Draft Mode, 1080p output with native HDR and optional EXR on production paths. Individual plans often labeled Plus, Pro, Ultra; Enterprise for business. API tiers often labeled Build and Scale. Pricing and exact duration caps move; hedge numbers and verify live.

Prompting principles that hold on Ray: write natural sentences with concrete motion; separate subject, camera, and scene motion; lead image-to-video with preserve; pair movement terms with speed; lock the camera with positive language when you want stillness; use Draft Mode to test kits cheaply when available.

How Luma differs from Kling, Runway, and Hailuo

Kling VIDEO 3.0 leans into flexible duration and labeled multi-shot narrative inside one generation. Runway Gen-4 targets cinematic lanes with rich camera vocabulary and UI motion dials beside the text. MiniMax Hailuo leans into timed audiovisual briefs and native stereo audio in one pass. Luma Ray3.2 leans into readable cinematic sentences, strong short-hold image-to-video, and keyframe or modify routes when your plan exposes them.

Porting prompts across hosts needs rewrite. Strip Shot labels when moving a Kling board into Luma unless you split into separate generations. Strip score and soundscape blocks when moving a Hailuo cue sheet into a silent Luma test. Expand preserve shells when moving a Pika three-liner into a longer Luma walk. Keep subject and motion clarity universal; change duration stamps, keyframe habits, and audio expectations per host.

For cross-host subject-motion-camera-duration theory, use the prompt-to-video AI guide or ai video prompt structure articles on this blog. For Kling camera-plus-duration dialect, use the Kling AI prompts guide. For Runway Gen-4 dials, use the Runway Gen-4 prompts guide. Stay here when you need Luma cinematic motion kits on Ray3.2.

Soft /image prep before Luma image-to-video

Luma image-to-video inherits pixels from your upload. A soft, noisy, or wrong-aspect still forces the model to invent edges while it also tries to obey your camera path. Clean the hero first.

PromptMake /image helps when you need structured still language from a reference photo, a lighting fix before motion tests, or a variation set that shares one Luma motion shell. Recreate Exactly holds composition. Adjust Lighting fixes exposure. Create Variation builds alternate heroes for A/B camera tests.

Workflow: finalize still at target aspect, upload to Ray image-to-video (and optional end frame or keyframes), paste preserve plus timed motion plus one camera path, generate. PromptMake never replaces Luma credits; it shortens the still step.

FAQ

What are luma dream machine prompts?

Luma dream machine prompts are the instructions people paste into Luma video generation. As of August 2026 the live model is Ray3.2 in the Luma App or API; Dream Machine is the legacy search name. Strong luma dream machine prompts name a preserve or subject card, one timed subject beat, one camera path with speed, and duration plus aspect. Soft mood words alone do not steer Ray3.2.

Is Dream Machine still the product name in 2026?

Official Luma materials treat Dream Machine and Ray2 as deprecated. The current video model is Ray3.2. The consumer workspace is the Luma App. Searchers still type luma dream machine prompts, so keep that phrase in your notes while selecting Ray3.2 (or the live Ray picker) inside the product. Confirm the label in your account; public names shift through mid-2026.

How should I structure a Ray3.2 cinematic motion kit?

Lead with preserve or subject anchors, then subject motion with a finish window, then one camera path with speed, then duration and aspect. Add light lock and short positive stability language last. Keep one motion story per generation. That kit pattern travels across text-to-video, image-to-video, and start-end bridges on Luma.

How do image-to-video prompts work on Luma?

Upload a still you trust. Lead with preserve language so face, wardrobe, or product finish stay locked. Add one timed motion beat and one camera path. Crop to output aspect before upload. Soft prep for a cleaner hero still: PromptMake /image at https://promptmake.net/image. Subtle motion on uploads lands cleaner than action verbs the still never showed.

What is Multi-Keyframe on Luma Ray3.2?

Multi-Keyframe lets you supply multiple anchor frames so Ray bridges continuous motion between them on supported routes (public materials cite up to sixteen keyframes). Write continuous verbs and say no cuts when you need a single take. Keep wardrobe and lighting consistent across anchors. Confirm the control exists on your plan before you design a sixteen-frame board.

Do Luma prompts need negative keywords?

Short positive locks help more than long NO walls: face identity locked, camera locked on tripod, stable lighting, no costume change when you need that freeze. Put bans after the spine in the main prompt when a separate negative field is missing. For a deeper artifact ban library across video hosts, see the video prompt negative keywords article on this blog.

How do I start for free before spending Luma credits?

Draft route, duration, motion kit, and camera path offline first. If you need a hero still, use PromptMake /image with the free guest or registered quota (about three /image runs per day as a guest; registration raises the image cap separately from /text). Use Draft Mode inside Luma when available for cheap smoke tests, then spend higher-quality credits after the kit holds.