Skip to content
Luma logo

Luma

Visit Luma

AI video generation from text and images with cinematic, production-grade output.

ImageVideo3D

How it works / How to use

AI video generation from text and images with cinematic, production-grade output. In Dream Machine, describe scenes in natural, detailed language—think conversation, not keyword lists. Use the 6-part framework (Subject + Action + Setting + Camera + Lighting + Style), front-load the first 20–30 words, and use only one camera move per clip. Refine with Modify, @style, @character, Extend & Keyframes, and Camera Motion controls.

  1. Open Dream Machine and start from text, an image, or an existing board clip.
  2. Write a structured prompt: subject, action, environment, single camera move, lighting, and style.
  3. Generate, then refine with Modify (“make colors warmer”) or references (@style, @character) instead of rewriting from scratch.
  4. Extend, loop, or keyframe when the story needs longer motion—keep context on one board so Luma retains creative continuity.

Text-to-video with structured prompts

Use Luma’s 6-part framework: Subject + Action + Setting + Camera + Lighting + Style. Front-load critical details in the first 20–30 words.

What to provide

  • Subject with specific details (clothing, materials, scale)
  • Active verb describing motion
  • Setting, time of day, and lighting
  • One camera movement and style anchor

Details that improve the result

  • Professional prompts specify shot type, camera behavior, light source, and aesthetic—not generic categories
  • Use detailed natural-language prompts; put critical details early in the prompt
  • Never ask for readable text or logos—composite those in post

Example prompt

Slow push-in on a ceramic coffee mug on a wooden table, steam rising from hot coffee, morning sunlight streaming through window from the left, warm golden lighting, shallow depth of field, soft focus background, cozy morning atmosphere, shot on 35mm film.

If the first output is not good

  • Change only camera: swap push-in for static lock-off
  • Add one style anchor: “muted earth tones with selective color pops”
  • Regenerate 3–5 variants at lower settings before final high-quality render

Common mistakes

  • Keyword dumps without action or camera language
  • Multiple camera moves in one prompt (e.g., push-in then pan left)
  • Expecting legible on-screen text or brand logos from the model

Image-to-video from approved stills

Upload the reference image, then prompt motion and camera behavior while inheriting look from the still.

What to provide

  • Hero product or keyframe image
  • Motion description that respects the still’s lighting and composition
  • What must not change (product shape, label placement, palette)

Details that improve the result

  • Image-to-video inherits lighting, color, and composition from the reference
  • Prompt directs motion; keep background changes minimal when brand approval matters
  • Useful for turning approved photography into motion without rebuilding the scene

Example prompt

Product slowly rotates revealing all angles, soft studio lighting, minimal camera movement, premium commercial aesthetic, product details stay in focus, clean background, 5 seconds.

If the first output is not good

  • Reduce motion if the product warps—tighten to “subtle slow orbit”
  • Add “static camera” if movement introduces artifacts
  • Reframe tighter if hands or edges distort

Common mistakes

  • Asking for new objects that contradict the reference still
  • Heavy camera moves on thin product geometry
  • Skipping post for mandatory logo or typography overlays

Character reference (@character)

Upload the image, type @character followed by your prompt so Dream Machine maintains appearance across shots.

What to provide

  • Reference image of the character you want to keep consistent
  • Text description of action and scene
  • Wardrobe or props that should persist

Details that improve the result

  • Character Reference creates consistent, personalized characters across generations
  • Include character description in every prompt when perfect consistency matters
  • Generate related shots in one session when possible

Example prompt

@character A woman in a flowing red dress walks through a sunlit Victorian garden, camera tracking from the side, golden hour lighting, gentle movement.

If the first output is not good

  • Tighten framing on face/hands if anatomy drifts
  • Add “natural hand positioning, anatomically correct fingers” when hands are visible
  • Use tighter compositions when small facial variation is unacceptable

Common mistakes

  • Changing wardrobe and face descriptors simultaneously across shots
  • Wide shots requiring perfect finger detail
  • Expecting identity match without uploading reference images

Visual style reference (@style)

Upload an image, type @style, then describe the scene you want rendered with that visual DNA.

What to provide

  • Reference image capturing palette, lighting, or texture
  • Prompt describing new subject matter in that style
  • Elements that must not copy literally from the reference

Details that improve the result

  • Visual Reference extracts color palette, lighting quality, texture, and mood
  • Reuse aesthetic keywords that consistently work on your boards
  • Combine with Modify for color temperature tweaks after generation

Example prompt

@style Minimalist watercolor landscape with the same paper texture and desaturated greens. Subject: bicycle leaning on a café wall, afternoon shade.

If the first output is not good

  • Modify: “Make colors warmer and add more trees” without regenerating from scratch
  • Swap @style image if palette drift persists
  • Note winning keywords (minimalist, surreal, watercolor) for campaign consistency

Common mistakes

  • Using @style without stating the new subject
  • Conflicting style anchors (anime + IMAX grain + neon noir in one line)
  • Discarding a good style ref after one failed motion prompt

Modify and iterative refinement

Use Modify to adjust visuals by describing specific changes in natural language instead of full prompt rewrites.

What to provide

  • Existing clip that is close
  • Conversational change request (color, props, lighting)
  • What must stay the same in the frame

Details that improve the result

  • Dream Machine best practices recommend Modify for targeted adjustments
  • Board context retention helps follow-up prompts build on earlier generations
  • Iterative generations: start broad, then modify details

Example prompt

Modify: Make the colors warmer, add more trees in the background, keep the same camera path and subject pacing.

If the first output is not good

  • Chain small Modify passes rather than one massive change
  • Ask “Help me with ideas” when stuck on next story beat
  • Use Brainstorm or More Like This to explore adjacent visuals

Common mistakes

  • Using Modify to change camera move and subject and style at once
  • Starting over when Modify would fix one element
  • Ignoring board context—splitting related shots across unrelated boards

Extend, keyframes, and loops

Use Extend & Keyframes to lengthen clips with transitions toward a new visual target; add “loop” or “looping video” when seamless repetition is needed.

What to provide

  • Starting clip or image
  • End visual target for keyframes (optional)
  • Whether the clip must loop seamlessly

Details that improve the result

  • Keyframes control motion between start and end states
  • Extend smoothly lengthens videos while steering toward a new target frame
  • Loop works when periodic motion should repeat without a hard cut

Example prompt

Start: peaceful meadow at dawn. Keyframe end: starry night with fireflies over the same meadow. Transition smooth, slow camera push, looping firefly motion.

If the first output is not good

  • Extend again with narrower transition language if the middle invents objects
  • Add “loop” only after timing feels right in a non-loop preview
  • Keep related prompts on one board so context retention helps continuity

Common mistakes

  • Keyframes with incompatible lighting between start and end
  • Requesting loop before motion cadence is validated
  • Extending clips with new camera moves stacked on top of the original

Camera motion controls

Pick one camera behavior per generation—Luma’s prompt guide warns that multiple moves confuse the model. Use Camera Motion controls where available.

What to provide

  • Chosen single move: Pan, Orbit, Zoom, tracking, dolly, static, etc.
  • Speed modifier (slow, gentle, fast)
  • Subject that reads clearly during the move

Details that improve the result

  • Lead prompts with shot type and camera movement when possible
  • Use “static camera” to lock position if movement causes artifacts
  • Prefer “gentle, steady push-in over 5 seconds” over vague “push-in”

Example prompt

Medium tracking shot of a chef plating pasta in an open kitchen, camera follows from the side, slow steady movement, warm practical lights, shallow depth of field.

If the first output is not good

  • Switch to static lock-off if orbit distorts faces
  • Split push-in and pan into separate clips for editing together
  • Match camera vocabulary to your prompt library templates

Common mistakes

  • “Slow push-in then pan left then crane up” in one generation
  • Fast camera moves on fine facial detail
  • No shot type named before subject description

Campaign consistency and prompt libraries

Build reusable templates: [SHOT TYPE] + [SUBJECT] + [SETTING] + [BRAND CAMERA STYLE] + [BRAND LIGHTING] + [BRAND COLOR] + [FILM STOCK]. Swap variables, keep constants.

What to provide

  • Brand template with locked style constants
  • Variables per asset (subject, setting, product)
  • Platform aspect ratio (16:9, 9:16, 1:1)

Details that improve the result

  • Organize libraries by client, content type, platform, and shot category
  • Iterate on several variations before committing credits to a final render
  • Front-load brand-critical details in the first 20–30 words

Example prompt

Wide shot of [PRODUCT] on [SURFACE], handheld with gentle movement, warm natural light, muted earth tones with selective color pops, shot on 35mm film, 9:16 vertical social cut.

If the first output is not good

  • Document prompts that repeatedly succeed for the brand
  • Generate regional variants by changing background copy in post, not in prompt text
  • Use Layers-style workflows conceptually: change product, keep lighting and composition language stable

Common mistakes

  • Rewriting prompts from scratch for every cutdown
  • Inconsistent lighting language between hero and social variants
  • Skipping a prompt library when producing dozens of SKUs

How to prompt

Write Dream Machine prompts like camera notes, not tag clouds. Use natural, detailed sentences. Structure: Subject + Action + Setting + Camera (one move) + Lighting + Style. Front-load the first 20–30 words. Use @character and @style when consistency matters; use Modify for conversational tweaks. Do not ask the model to render readable text or logos.

Medium tracking shot of a woman in a flowing red dress walking through a sunlit Victorian garden, camera following from the side, 35mm lens, golden hour lighting, shallow depth of field, gentle movement, muted color grading.

Use the 6-part framework

Luma’s official prompt guide defines six elements: Subject, Action, Setting, Camera, Lighting, and Style/Aesthetic.

Example

Subject: ceramic mug. Action: steam rising. Setting: morning kitchen. Camera: slow push-in. Lighting: window light from left. Style: 35mm film, cozy atmosphere.

Front-load critical details

Put shot type, subject, and camera move early in the prompt

Example

Slow push-in on a ceramic coffee mug on a wooden table, steam rising… (lighting and style follow).

One camera move only

Luma warns that combining moves like push-in then pan confuses the model—split complex motion into separate clips.

Example

Clip A: static medium shot of subject. Clip B: slow orbit around product—cut together in post.

Anchor style to kill the generic AI look

Use film stock, color grading, and lighting recipes—e.g., Kodak Vision3, Blade Runner 2049 teal/orange, soft window light camera-left.

Example

Shot on Kodak Vision3 500T, soft window light from camera-left, slight grain, muted earth tones.

Refine with Modify and references

Dream Machine best practices: conversational Modify passes, @style/@character uploads, and board context for cohesive sequences.

Example

@style reference + prompt: “Add golden hour glow.” Modify: “Include a small cottage in the distance.”

Best output tips

Natural language over tags

Luma’s Dream Machine best practices say to treat prompts like a conversation—specific adjectives and clear descriptors beat vague mood words alone.

6-part framework

Official Luma guidance: Subject + Action + Setting + Camera + Lighting + Style. Weak prompts leave creative decisions to the model; strong prompts decide them for you.

Prompt length

Use detailed natural-language prompts and front-load critical details (shot type, subject, camera). Too little detail reduces control; very long prompts can dilute emphasis.

Single camera move rule

Use only one camera movement per clip. Combine static, push-in, tracking, or orbit—not stacked sequences in one generation.

Camera vocabulary

Models recognize shot types (ECU, CU, medium, wide) and moves (push-in, pull-back, dolly, tracking, orbit, crane). Name them explicitly.

No text or logos in prompts

Luma’s FAQ-level guidance: AI cannot reliably render readable text or logos. Plan overlays in post-production.

@character consistency

Upload a reference image and prefix prompts with @character to keep personalized subjects consistent across clips.

@style visual DNA

Upload a reference and prefix with @style to transfer palette, lighting quality, texture, and mood to new subjects.

Modify workflow

Describe specific conversational changes—“make colors warmer and add more trees”—instead of regenerating entire scenes.

Extend, keyframes, loop

Extend & Keyframes lengthen clips toward a new visual target; add “loop” or “looping video” when seamless repetition is required.

Board context retention

Dream Machine retains context within a board—evolve scenes with follow-ups like “Add a golden hour glow” or ask “Help me with ideas.”

Style presets

Predefined aesthetics like Anime or Cinematic align visuals quickly; combine with your own aesthetic keyword notes for repeatable looks.

Hands and faces

When hands are non-essential, frame them out. For visible hands, prompt natural positioning; generate several variants and pick the best.

Iterate before final renders

Explore variations, refine the strongest direction, then use your best settings for the final render.

Campaign templates

Lock brand camera style, lighting, and color language in a template; swap subject and setting per asset for cohesive multi-format campaigns.

  • Write natural-language prompts with concrete subjects, actions, and lighting—not keyword soup.
  • Use one camera movement per generation; cut complex moves in post.
  • Front-load shot type, subject, and camera in the first 20–30 words.
  • Upload @character or @style references when consistency matters more than speed.
  • Use Modify for targeted fixes instead of full prompt reboots.
  • Keep related shots on one board so context retention helps continuity.
  • Never rely on AI for readable text or logos—composite in post.
  • Build prompt templates with locked brand constants and swapped variables.
  • Explore several cheap variants, refine winners, then run a final high-quality render.

Try this AI

Try Luma

Product Details

Pricing, features, limits and latest updates

Luma

AI video generation from text and images with cinematic, production-grade output.

Paid · $30 / month

Pricing Plans

Plus

$30/ Monthly

Plus

$300/ Yearly

Pro

$90/ Monthly

Pro

$900/ Yearly

Ultra

$300/ Monthly

Ultra

$3,000/ Yearly

Team

Custom

Enterprise

Custom

Key Features

Image Generation

The same credit wallet generates and edits images with Luma and partner models.

Video Generation

Dream Machine generates video with Luma and third-party models for credits.

Agents

Luma Agents plan, generate, and iterate; Pro and Ultra multiply agent usage.

Limits

  • API prompt length: 5000 characters. Maximum API prompt length (Luma API errors docs).
  • App vs API credits: Dream Machine app credits and API credits are separate pools (Luma credit system help).

Ideas / Prompt experiences

Share a prompt that worked for you. Username and email are shown with your submission. External links are not allowed.

Example prompt

Dream Machine structured video prompt

Prompt

Subject: barista pouring latte art in a small cafe. Action: slow pour, steam rising. Camera: static medium shot. Lighting: warm morning window light. Style: cinematic, shallow depth of field.

Short explanation

Luma's prompt guides recommend front-loading subject and action; one camera move per clip works best.

Example prompt

Modify with Adhere

Prompt

[upload clip] Modify with Adhere: change only the sky to overcast. Keep the subject, camera angle, and foreground unchanged.

Short explanation

Modify modes (Adhere/Flex/Reimagine) are for conversational refinement; name what must stay stable.

Share your experience

Required fields are marked. Variation, result, and explanation are optional.

Midjourney logo

Midjourney

Explore

AI image and video generation from text prompts and reference images.

ImagePaid

Runway logo

Runway

Explore

AI image and video generation and editing for motion, storytelling, and creative workflows.

ImageFree / Paid

Ideogram logo

Ideogram

Explore

Image generation with accurate in-image text, typography, and layout.

ImageFree / Paid