Motion blur and shutter language in prompts
Motion blur names an outcome, not an instruction. Shutter and exposure phrasing for crisp versus smeared movement, and when to add the blur in post instead.
Add "motion blur" to a video prompt and you have told the model roughly as much as adding "colour". Blur is a result, and the thing that produces it is exposure time — how long the sensor is collecting light while the subject moves across it. Prompts that name the exposure get a look. Prompts that name the result get whatever the model already wanted to do, dressed in a slightly softer grade.
That distinction decides both halves of the problem. Crisp, frozen movement is a short exposure. Smeared, trailing movement is a long one. Neither is a mood, and neither is well served by adjectives.
Blur is exposure time
The number that governs everything here is the ratio between exposure time and frame interval. Cinema's long-standing default is the 180-degree shutter: the shutter is open for half of each frame's duration. Versely's default video frame rate is 25 fps, so a frame lasts 1/25 of a second and a 180-degree shutter means roughly a 1/50s exposure. At 24 fps it works out at 1/48s, which in practice everyone shoots as 1/50 anyway.
That 1/50 is what "normal" looks like. It is the blur length your eye has been trained on by every film you have watched, which is why footage that departs from it reads as stylised in one direction or the other:
| Exposure | Look | Where it comes from |
|---|---|---|
| 1/25s and longer | Heavy smear, light trails, dreamy | Long-exposure and low-light footage |
| ~1/50s | Neutral, filmic | The 180-degree shutter default |
| 1/250s | Slightly crisp, a touch clinical | Bright daylight, sports coverage |
| 1/1000s and shorter | Frozen, edges sharp on every frame, visible judder | Sports, high-speed, harsh sun |
Two housekeeping notes before the phrasing. First, the 180-degree shutter has nothing to do with the 180-degree rule of screen direction — they share a number and nothing else. Second, blur length depends on how fast the subject actually moves, not only on exposure. A 1/1000s shutter on a slow push-in changes nothing visible, because nothing travelled far enough during the exposure to smear in the first place. Shutter language only pays off on fast motion.
Language for crisp motion
Ask for the short exposure and the consequences of it, in that order:
Shot at 1/1000s shutter in harsh midday sun. Each frame is individually
sharp: the droplets of water hang with hard edges, the runner's foot is
frozen mid-strike, no smear on the trailing arm.
The second sentence is doing as much work as the first. Naming the visible consequence — hard edges on a droplet, a foot frozen at contact — gives the model something concrete to render even where a numeric shutter value isn't a parameter it consumes. This is the same split that runs through all camera language: some tokens are backed by a real value and some are prose the model approximates, and it is model-dependent which is which. The full version of that argument is in the write-up on camera instructions models silently ignore, and it applies to exposure exactly as it applies to dollies.
Two more phrases pull the same direction: a capture context that implies a fast shutter (sports broadcast footage, bright daylight, deep depth of field), and the artefact that comes with it (slight strobing on the fast pan). The artefact is the tell — real high-shutter footage judders on lateral movement, and asking for the judder is a stronger signal than asking for sharpness, which is what the model was biased toward anyway.
Language for smeared motion
The long-exposure direction is easier, because it is further from the model's default and therefore more distinctive when it lands:
Long exposure, roughly 1/8s, handheld. The passing cars dissolve into
continuous streaks of red and white; the standing figure stays legible
while everything moving smears through the frame.
Say what stays sharp. That is the whole craft of the smeared look — a long exposure only reads as a long exposure if something in frame is stationary enough to be a reference. A prompt where everything smears just looks like a broken generation.
Directional detail helps too: horizontal streaks following the train's travel is executable; blurry is not. And be careful with the word "blurry" specifically. It sits in the same neighbourhood as out-of-focus and low-quality in most training data, so it will happily get you a soft, mushy frame rather than a directional smear.
Why "cinematic motion blur" does nothing
It names no exposure, no direction, no magnitude and no subject. What it does carry is a strong association with a grade — lower saturation, lifted blacks, more contrast — which is why prompts loaded with it come back looking moodier without moving any differently. That's the same failure pattern as "cinematic lighting", and it has the same fix: replace the mood word with a physical parameter and its visible consequence.
The related trap is trying to remove blur with a negative. Writing no motion blur is weak for two reasons. On flow-matching models running at low guidance there is often no negative channel to push against at all, so the phrase is close to a no-op — the background on that is in the negative prompt glossary entry and the longer treatment of what replaced negative prompts. And instructive "no X" phrasing tends to leak X back into the frame regardless, because the token you are trying to suppress is still the most salient thing in the sentence. The general form of the fix holds everywhere: describe the desired state rather than issuing a prohibition. every frame individually sharp beats no motion blur every time.
When to add the blur in post instead
There is a hard asymmetry here that should drive the default: you can add blur afterwards, you cannot take it out. Blur is baked into the pixels of a generation. Sharpening a smeared frame recovers nothing, and upscalers asked to work on heavily smeared footage tend to invent detail that was never there.
That gives you a clear rule. When you are unsure, prompt toward the crisper end and add blur downstream. Four situations where post is straightforwardly the better answer:
- Matching across models. A six-shot sequence assembled from three different models will arrive with three different native blur characteristics. Normalising in the edit is far cheaper than rerolling until they agree.
- Blur on one element only. A directional smear on the passing car but not the standing figure is a masked, layered operation. No prompt gives you that separation.
- Frame-rate conform work. If clips are being retimed or interpolated to sit on a 25 fps timeline, blur added before the conform gets stretched along with everything else. Do it after. The arithmetic behind that conform is in 24fps clips on a 25fps timeline.
- Anything heading for an upscale. Order the finishing pass as clean, then 1080p, then 4K. Upscaling a blurred source is the worst-value pass in the pipeline, because there is no high-frequency detail left for it to reconstruct.
The one case where prompting is clearly right is when the blur is intrinsic to the shot — a whip pan, a crash zoom, anything where the camera itself is the fast-moving thing. There the smear is the move, and faking it later never quite matches. Those moves have their own failure modes, covered in the piece on prompting high-velocity camera moves.
A ten-minute test
Whether a numeric shutter value does anything on a given model is an empirical question, and it is quick to answer. Take one fast-motion prompt and generate it three ways, holding everything else identical:
- No exposure language at all.
shot at 1/1000s shutter, each frame individually sharp.long exposure, roughly 1/8s, moving elements smear into streaks.
Step through the results frame by frame rather than watching them play. If versions 2 and 3 differ from each other in blur length — not just in grade — the model responds to exposure language and it belongs in your template for that model. If all three look the same, exposure is prose there, and the honest answer is to generate crisp and handle blur in the edit. Run it once per model you rely on and write the answer down.
FAQ
What shutter speed should I ask for by default?
Roughly 1/50s, which is the 180-degree shutter at 25 fps and close enough at 24. That is the neutral, filmic amount of blur, and unless you want the shot to read as stylised, it is what you want. Deviate deliberately: shorter for a frozen, sports-like or aggressive feel, longer for dreamy or trailing motion.
Does writing a specific shutter number actually work?
It depends on the model, and it is worth testing rather than assuming. Numeric exposure values behave like the rest of camera language — sometimes a real parameter, more often prose the model approximates. What raises the hit rate everywhere is pairing the number with its visible consequence: name what the short exposure does to a droplet or what the long exposure does to passing headlights, and you have given the model something to render even when the number itself is ignored.
Should I add motion blur in the prompt or in post?
Default to post, because blur is additive and irreversible. Generate toward the crisper end, then apply blur in the edit where you can control amount, direction and which elements it touches. The exception is blur that comes from the camera itself — whip pans, crash zooms — where the smear and the move are the same thing and a post-applied blur never quite convinces.
Why does my crisp, high-shutter footage look juddery?
Because that is what real high-shutter footage does. With little blur to bridge the gap between frames, fast lateral movement steps rather than flows, and at 25 fps the steps are visible. It is a property of the look, not a defect. If you want the crispness without the stutter, slow the movement in the shot rather than raising the frame rate — generating above 30 fps tends to cost temporal stability rather than buy it.