You have a great photo. You want the person in it to move. Or the scene to transition from day to night. Or that magic circle you drew to actually cast a spell.
Keyframe control in Luma Dream Machine is exactly what you need.
Simply put: you give the AI a “start” photo (first frame) and an “end” photo (last frame). The AI then fills in everything between them — the full transformation.
This is way more precise than text-only generation. Because the final look of your subject is locked in. The AI only has to figure out how the change happens.
Part 1: How to Set Up Start and End Frames
On Luma Dream Machine’s web interface, the workflow is straightforward:
- Enter Keyframe Mode: Click the “+” icon in the bottom-left corner, then select “Keyframe Video” mode.
- Upload Start Frame: In the Start Frame card, upload the image you want the video to begin with. This can be any existing photo or an AI-generated image.
- Upload End Frame: In the End Frame card, upload the image you want the video to end on.
- Write a Transition Prompt: To guide the transition, enter a short instruction describing the change. For example:
"face stays still""transition from sunset to starry night""flower slowly blooms"
- Generate: Click “Generate” and wait 30 seconds to a few minutes. Watch your static images flow into a smooth video.
Part 2: Pro Tips for Stable, High-Quality Results
Just uploading two random images often leads to warping or jumpy movement. Here’s how to get professional-grade stability:
1. Be Specific with Prompts (Especially for Realism)
Don’t assume the AI will “just get it.” Use prompts to guide its imagination.
- Lock the subject’s form: If you’re animating a person aging, add
"face stays still"to your prompt. Otherwise, facial features may warp. - Avoid vague verbs: Use concrete action words like
"rotate","lift","drift","unfold", or"melt". Avoid"change"or"become"— they’re too abstract.
2. Use Visually Similar Start and End Frames
For the smoothest transitions, your two images shouldn’t be dramatically different.
- Match the background/style: If you don’t have two matching images, generate them with Midjourney or similar tools. Keep the same keywords to lock in the art style and background.
- Lock characters with Character Reference: Dream Machine supports uploading a reference image of a character. In keyframe mode, upload that reference separately — the model will keep that character consistent even if your start and end frames look very different.
3. Use Loop and Extend Features
- Seamless loops: For infinite looping videos (like game effects), check the
{loop}option. If you see a bump at the seam, enable"Refine Loop"and set it between 30-50 — the AI will fix the transition. - Extend your video: The default is 5 seconds. After generation, click
"Extend"and upload another image as a new end frame. The video will continue flowing toward that new image, up to 9 seconds total.
Part 3: Advanced Use Cases
Make an Old Photo Come to Life
Start frame: A young person’s photo
End frame: An elderly version of the same person
Prompt: "face stays still, wrinkles gradually appear, hair turns gray"
The AI generates a time-lapse of aging — wrinkles forming, hair graying — all while keeping the person recognizable.
Game Effects Without Frame-by-Frame Animation
Start frame: A fireball just forming (small, concentrated)
End frame: The fireball exploding (particles spreading everywhere)
Prompt: "fireball expands outward and bursts into particles"
Drop both into Dream Machine. The AI fills in all the in-between frames — the expansion, the crack, the particle scatter. Export the 5-second video back to Photoshop and extract the exact frame sequence for your game’s sprite sheet.
Product Showcase with 3D Feel
Start frame: A black phone case on the left
End frame: A white phone case on the right
Prompt: "phone case rotates slowly in 3D space while color fades from black to white"
The AI generates a smooth product rotation with a color transition — looks like a professional 3D render but it’s just two photos.
Seasonal Magic
Start frame: Your balcony in summer (green leaves, bright sun)
End frame: The same balcony edited to look like winter (snow on the railing, bare trees)
Prompt: "leaves fall from trees, snow gradually covers the railing, sunlight fades to overcast"
The result feels like a cinematic time-lapse — autumn leaves dropping, snow accumulating, lighting shifting.
Part 4: Troubleshooting Common Issues
| Problem | Likely Cause | Solution |
|---|---|---|
| Jump cuts, warping, deformed subjects | Start and end frames have very different subject positions or sizes | Open both images in Photoshop (or any editor). Scale and reposition the subject so they roughly match in size and placement before uploading |
| Ghosting, flickering, double images | Transition strength too high, or prompt too vague | Lower the “Strength” slider (try around 0.35). Add constraints like "stable transition, smooth motion, no ghosting" to your prompt |
| Generated video doesn’t start from my first image | Frames weren’t properly aligned | After uploading both images, click "Preview Alignment". The heatmap shows red where misalignment occurs. Adjust your images and re-upload |
| The transition is too fast / too slow | Default duration isn’t right for your content | Luma currently fixed at ~5 seconds. For slower transitions, use the "Extend" feature to add more frames. For faster, generate multiple short clips |
| Background elements flicker | Background isn’t consistent between frames | If possible, keep the background identical. Use "background stays still" in your prompt |
Summary: The Philosophy of Keyframe Control
The core idea of keyframe control is “what changes vs what stays the same.”
You lock in the starting point and ending point — the certainty. Then let the AI’s creative “uncertainty” fill the space between them.
With start and end frames, you’re not just using AI. You’re directing it — telling it exactly where to begin and where to end, so it can focus all its intelligence on making the journey beautiful.
Quick Reference Card
| Step | Action |
|---|---|
| 1 | Click + → Keyframe Video |
| 2 | Upload Start Frame |
| 3 | Upload End Frame |
| 4 | Write transition prompt |
| 5 | Click Generate |
| Pro tip | Add "face stays still" for portraits |
| Pro tip | Lower Strength to ~0.35 for smoother motion |
| Pro tip | Click Preview Alignment before generating |
