Luma Dream MachinePrompting basicsParameters & settings
Luma Dream Machine Guide: Ray2, Photon and Settings
Luma Dream Machine, at lumalabs.ai, pairs the Ray2 video model with the Photon image model inside a Boards-based workspace. You write a plain-language description of one shot, pick duration, aspect ratio and resolution, and generate; Ray2 Flash variants render faster for drafts while full Ray2 gives the best motion. Starting from a Photon image rather than text is the most reliable way to control the look.
Dream Machine is Luma’s creative workspace for images and video. Ray2 handles video, Photon handles images, and both live in Boards, which are canvases where generations, references and variations sit together. The experience is conversational: you describe what you want, Luma expands it, and you refine from there. This guide explains the pieces, the settings that matter, and how to prompt Ray2 so that your first session is productive. Luma ships changes often, so where a label here differs from your screen, the screen is right.
The parts of Dream Machine
- Ray2 is the video model. It produces clips of around five seconds, with longer options of nine to ten seconds, from text or from a start image. It is strong on natural, physically plausible motion and on camera moves.
- Ray2 Flash variants are faster, cheaper versions of the model. Motion quality is a little lower, which makes them ideal for testing prompts before a final render.
- Photon is the image model. Use it to make the start frames for your videos, or for images on their own. Photon is quick, handles text prompts well, and produces stills that Ray2 animates cleanly.
- Boards are the organising layer. Every generation lands on a board, and you can keep one board per project, client or idea. Boards keep the prompt and settings alongside each result, which saves a great deal of memory later.
- Editing tools around the clip: keyframes (start and end images), Loop, Extend, Reframe for changing aspect ratio, and Modify Video for restyling an existing clip. These have their own guides: keyframes, loops and extend and Modify Video.
Prompting Ray2
Ray2 takes a natural, descriptive prompt and interprets it generously. It does not need keyword lists and tends to do worse with them. A shot-description structure works well:
- Subject and action: "A red kite drifts above a windswept clifftop."
- Setting and light: "Late afternoon, long shadows, grey-green sea below."
- Camera: "The camera rises slowly, following the kite."
- Style: "Naturalistic, slightly desaturated, shot on film."
Forty to eighty words is enough. Luma’s interface can expand or enhance your prompt automatically; this helps when your prompt is thin and can over-decorate when it is already detailed, so switch it off once you are writing full descriptions and want precise control.
Ray2 understands camera vocabulary well. Push in, pull back, pan, tilt, orbit, tracking shot, crane up, handheld and static all produce recognisable results. Pick one per clip. It also responds to speed words such as slow motion and time-lapse. Avoid asking for several events, on-screen text, or named brands and people, which are declined or unreliable.
Image first, then video
The most dependable Dream Machine workflow is to make the frame with Photon and the motion with Ray2. Photon lets you iterate on composition, palette and subject quickly and cheaply, and a strong still gives Ray2 a precise starting point. When you animate a Photon image, describe only what moves: the camera and the subject’s action. Re-describing the picture is unnecessary and can pull the clip away from the still.
Keep subjects away from the frame edge if they need to move, and leave depth in the background if the camera is going to push in. A flat, edge-to-edge composition gives the model nothing to travel through.
Settings at a glance
Options vary by plan and have moved around the interface over time. These are the controls you will meet.
| Setting | What it does | Start with |
|---|---|---|
| Model | Ray2 for final quality; Ray2 Flash variants for faster, cheaper drafts | Flash to test, Ray2 for finals |
| Duration | Clip length, around 5 seconds or 9 to 10 seconds | 5 seconds until the motion is right |
| Aspect ratio | Frame shape, from widescreen to vertical | 16:9, or match your start image |
| Resolution | Output size, 720p or 1080p, with upscale options for finals | 720p drafts, 1080p or upscaled finals |
| Start image (keyframe) | Fixes the first frame of the clip | A Photon image cropped to the aspect ratio |
| Prompt enhance | Expands a short prompt automatically | On for one-line ideas, off for detailed prompts |
| Loop | Makes the clip end where it began | Off, unless the motion is continuous |
Working efficiently
Credits and time are the limits, so a cheap-to-expensive order of operations pays off.
- Make the still in Photon until the frame is right. Image generations are quick and inexpensive compared with video.
- Animate with a Flash variant at 5 seconds and 720p, with a motion-only prompt.
- Change one thing per attempt: camera, action, or light. Note what changed.
- When the draft moves correctly, rerun with full Ray2 and upscale or render at 1080p.
- Use Extend for a longer shot and Reframe to produce other aspect ratios from the same clip rather than regenerating for each format.
- Keep one board per project and name it clearly. Download the finals you approve, since the board is your working record rather than a delivery archive.
Try it
A red kite drifts above a windswept clifftop in late afternoon light, long shadows across the grass, grey-green sea far below. The camera rises slowly, following the kite as it tilts in the wind. Naturalistic, slightly desaturated, shot on film, slow and steady.
Run it as text to video on a Flash variant first. Then generate the same scene as a Photon still, animate that with only the camera and motion sentences, and compare: the image-first version usually holds its composition better, while the text-first version has more freedom in how the shot develops.