Create Slow Mo Videos Using AI Models
You can't slow existing footage down here. You can generate footage that is already slow motion, by asking for it, and models are good at that.
TL;DR
There is no frame interpolation, retiming or fps control on the platform, so existing footage cannot be slowed here. What works is generating a clip that is already slow motion by naming the technique in the prompt, choosing subjects that reward it such as splashes and fabric, keeping the camera static and the action small. For a slow controlled camera move, first-to-last frame with two close-together frames and a longer duration is the nearest thing to speed control.
You cannot slow existing footage down here. There is no frame interpolation, no retiming, no optical flow pass, and no fps control anywhere on the platform.
What you can do is generate footage that is already slow motion, by asking for it. Models render slow motion as a look, the same way they render golden hour or shallow depth of field, and they are surprisingly good at it.
Those are genuinely different jobs, and knowing which one you need saves you from hunting for a feature that is not there.
If you have footage you want slowed
This is a video editing job, not a generation one.
Any editor will retime a clip. The quality ceiling depends on what you shot: footage captured at a high frame rate slows cleanly, because the frames genuinely exist. Footage shot at 24 or 30 fps has to have frames invented, which is where ghosting and duplicated limbs come from.
That inventing step is frame interpolation, and it is what tools like Twixtor and the optical flow modes in Premiere or Resolve do. Nothing here does it. If your source is already shot, use a tool built for retiming.
The real lesson is upstream anyway: slow motion is decided at capture. Shoot at 120 or 240 fps and the slow version is free. No amount of post-processing beats having the frames.
If you are generating the clip
Then you are not slowing anything down. You are asking for a shot that looks like slow motion, and that is a prompt.
"Slow motion, water droplets falling into a still pool, high-speed camera look, shallow depth of field, camera static."
What makes this work:
Name the technique. "Slow motion" and "high-speed camera" both read as instructions. Models have seen a great deal of slow-motion footage and know what it looks like.
Pick subjects that suit it. Slow motion is compelling on things that normally happen too fast to see: splashes, fabric, hair, dust, breaking, pouring, impact. It is dull on a person walking.
Say the camera is static. A moving camera fights the effect and makes the result read as a slowed-down normal shot rather than a high-speed one.
Keep the action small. Slow motion stretches a moment. If the prompt describes a whole sequence, the model compresses it back to normal speed to fit.
Start in Text to Video, or image to video if you have a still to animate.
Why duration is not a speed control
Duration sets how long the clip runs, not how fast the action moves inside it. Asking for eight seconds does not slow a four-second action down; it gives the model eight seconds to fill, and it will invent more action to fill them.
To get slower movement, describe slower movement. "Droplet falls slowly" does more than any length setting.
What is not real
No frame interpolation. No route, no model, no capability. Claims about converting 24 fps to 240 fps are describing software that does not exist here.
No fps control. You do not set 24, 60 or 120 at generation time.
No motion strength or speed slider. Across all 37 video generation models the parameters are prompt, duration, aspect ratio and resolution, plus references, a negative prompt or first and last frames on some.
No per-model slow-motion quality ranking. Tables scoring models on slow-mo invented the column.
Getting a controlled slow move
If what you actually want is a slow, deliberate camera move rather than slow-motion action, there is a better tool for that.
First-to-last frame, on Seedance 2.0 and 2.0 Fast, takes a start image and an end image and generates the movement between. Two frames that are close together with a longer duration gives you a genuinely slow, controlled move, because you have defined both endpoints and the model only fills the middle.
That is the closest thing here to precise speed control, and it works because you constrained the distance rather than the speed.
The short version
Existing footage: retime it in an editor, and shoot high frame rate next time.
New footage: ask for slow motion in the prompt, pick a subject that rewards it, keep the camera still and the action small.
The catalog is at Models, each generation is quoted before it runs, and pack prices are on the pricing page.
Frequently Asked Questions
Can I slow down existing footage with AI here?▾
No. There is no frame interpolation, retiming or optical flow pass, and no fps control. Slowing existing footage is a video editing job, so use an editor built for retiming. Claims about converting 24 fps to 240 fps describe software that does not exist here.
How do I generate slow motion video?▾
Ask for it in the prompt. Naming the technique with phrases like "slow motion" or "high-speed camera look" works, because models have seen a great deal of slow-motion footage and know what it looks like. It is a visual style rather than a speed setting.
Which subjects work best in slow motion?▾
Things that normally happen too fast to see: splashes, water droplets, fabric, hair, dust, breaking, pouring and impacts. Slow motion is dull on ordinary actions like a person walking, since there is nothing revealed by stretching them.
Does increasing the duration slow the action down?▾
No. Duration sets how long the clip runs, not how fast the action moves inside it. Asking for eight seconds gives the model eight seconds to fill, and it will invent more action to fill them. To get slower movement, describe slower movement.
Why does my slow-motion clip look like normal speed?▾
Usually a moving camera, which fights the effect and makes the result read as a slowed-down normal shot. Specify a static camera. The other common cause is describing a whole sequence, which the model compresses back to normal speed to fit the duration.
How do I get a slow, controlled camera move?▾
Use first-to-last frame on Seedance 2.0 or 2.0 Fast with two frames that are close together and a longer duration. Because you have defined both endpoints, the model only fills the middle, which is the nearest thing to precise speed control available.
What is the best way to get good slow motion overall?▾
Capture it. Slow motion is decided at the camera: footage shot at 120 or 240 fps slows cleanly because the frames genuinely exist. Footage shot at 24 or 30 fps needs frames invented, which is where ghosting and duplicated limbs come from.
