Comparison · Reviewed
Induce vs Luma: an agent on a canvas, or a plan for every shot?
Luma has moved on from Dream Machine to an agent workspace built around its Ray and Uni models. It can read a script and it runs several of the models Induce routes to, so the useful question is what each tool does once the story starts to change.
A Luma canvas and an Induce shot plan, step by step
How a team typically moves from idea to sequence in each tool, based on official documentation.
Induce
- Screenplay in, breakdown outScenes, cast, wardrobe, props, and shots come from the script, not from a prompt you write to an agent
- References plus story stateAlongside what must not change, each shot records what has changed since the last one
- A model per shotKling, Seedance, Veo, or Nano Banana, chosen for what the shot needs
- Checked on every takeRhapsody compares each take with the shot plan without being asked
Luma
- Brief, script, or filesUpload a script or PDF and ask the agent for a shot list
- Master Reference AssetsLabelled characters, products, and logos kept on the board
- Generate on the canvasRay3.2, Uni-1, or third-party models; restyle, extend, reframe
- Re-anchor & deliverCompare outputs with the references; stitch clips, export EXR
Luma anchors what must not change. Induce also tracks what has.
Luma’s reference guide describes Master Reference Assets as the fixed elements of a project: a character’s face and wardrobe, a logo, a product’s design. It also warns that long chains of edits drift, and tells you to re-anchor to those assets every few iterations and compare outputs with them before delivery. That keeps the baker recognisable. It doesn’t tell shot seven that his sleeves are now rolled up or that the first tray has left the oven. Induce keeps that record in the shot plan, and Rhapsody checks every take against it. A check returns a match, a mismatch, or unclear, and anything unclear goes to a person.
What a reference holds
- Tomas’s face and build
- The flour-dusted apron
- The bakery’s ovens and counter
What the shot also inherits
- His sleeves were rolled up in shot 3
- The first tray is on the rack, not in the oven
- The window behind him is no longer dark
Where Luma and Induce differ, feature by feature
| Area | Induce | Luma |
|---|---|---|
| Starting point | Screenplay or brief, broken into scenes, cast, wardrobe, props, and shots | Prompt, brief, or uploaded files. A shot list comes from asking the agent for one |
| Consistency method | References, plus the changes each shot inherits, such as rolled sleeves or a tray already out | Master Reference Assets, labelled on the board and named in each instruction |
| Checking takes | Rhapsody checks every take against the shot plan; reviewers see match, mismatch, or unclear before approving | You compare outputs with the references, and can ask the agent to evaluate alignment |
| Models | Each shot routed to Kling, Seedance, Veo, or Nano Banana, whichever suits it | Luma’s Ray3.2 and Uni-1, plus third-party models including Veo, Kling, Seedance, and Nano Banana |
| Fixing a bad take | Re-plans only what failed and keeps what worked; retake reviewed in the cut | Instruct the agent to revise, naming what to preserve, and re-anchor every few iterations |
| Editing & handoff | Review the assembled cut; confirm export and editor handoff in a pilot | Stitch clips and add voice, music, and captions in the workspace; EXR export for production pipelines |
| Team & enterprise | Team access, SSO/SAML, VPC, and API scoped with your team during a pilot | Team Workspaces and shared canvases; SSO and access controls on Enterprise; a separate API |
When Induce is the right Luma alternative
Choose Luma when
- The work is campaign assets: many variations of a product, logo, or character that must stay on brand.
- You want image, video, voice, music, and captions made in one workspace.
- You need to restyle or reframe footage you already have, or deliver EXR to a finishing pipeline.
- Your team likes to brainstorm and iterate with an agent on an open canvas.
Choose Induce when
- The project starts as a screenplay and runs to many connected shots.
- Wardrobe, props, and lighting change during a scene and have to stay changed.
- You want every take checked against the story, not only when someone asks.
- Clients sign off on continuity and want to see why a take was accepted or sent back.
The two can share a project. Teams can develop looks and reference images on a Luma canvas, then build the narrative sequence in Induce using those images as references.
Test a scene where things change in Luma and Induce
Use a six-to-eight-shot scene with one recurring character, one wardrobe detail that changes partway through, and one prop that moves.
- Upload the same script to both and compare Luma’s agent-written shot list with Induce’s breakdown before generating anything.
- Change the wardrobe detail after shot three and mark every later shot where it reverts to the reference.
- Count how often each workflow needed a person to notice a break before it was fixed.
- Log every take generated and how many your reviewers approved on first viewing.
- Compare the cost and review hours per approved minute, not per clip.