Deepfake Video That Holds Through Motion
Upload a clip and one clear face. The deepfake video comes back at full resolution with the original audio intact.
A deepfake video is the same face transfer you would run on a photograph, repeated once per frame and then held together across time. The second half is the hard part. A still only has to be convincing on its own; a deepfake video has to be convincing and then agree with the twenty-nine frames around it, because the eye forgives soft detail far more readily than a jawline that jumps a pixel between frames.
The arithmetic follows. A ten-second clip at thirty frames per second is three hundred separate face transfers, which is why a deepfake video is measured in minutes where a photo takes seconds, and why render time tracks frame count rather than file size. Sixty frames per second doubles the work for the same running time, so trimming to the shot that carries the idea is the most useful thing you can do before starting a deepfake video render.
You bring a clip in MP4 or MOV and one clear still of the face you want in it. Detection, landmark alignment, lighting match, edge blending and the re-encode happen here. The audio passes through untouched, so a deepfake video changes the face and nothing else — no voice cloning, no lip sync. Nothing is installed, nothing needs a GPU of your own, and your upload never enters a training set.
What Changes When the Input Is Footage
Every frame gets its own pass
Each frame of a deepfake video is detected, aligned and blended independently, with no track interpolated forward from the opening frame. A hard cut in the middle of a clip does not derail the sequence, and a face that leaves frame and returns is picked up cleanly.
Stability through head turns
The moment a subject turns, the model loses geometry it had a frame earlier and has to rebuild it. This is where a weak deepfake video comes apart — the jaw swims, the hairline breathes, the face pulses once per turn. Holding identity through rotation is the tell viewers register first.
The audio is left alone
A deepfake video here changes the face and nothing else. The original sound is written back unmodified, so the performance you filmed is the performance you keep. No voice cloning, no speech synthesis, no lip sync — the mouth follows the take that was recorded.
Full resolution, audio intact
The file comes back at the frame size you uploaded, with no preview-grade downgrade and the original audio untouched. The re-encode is the only quality cost, kept light enough that a deepfake video does not leave looking worse than the footage that went in.
No install, no GPU, no queue
The whole deepfake video pipeline runs on our hardware. No desktop app to configure, no model weights to download, no Python environment to break, no queue. You upload from a browser tab and download the finished clip there.
How to Make a Deepfake Video
Cut the clip down first
Choose the few seconds that carry the idea and trim to them before uploading. Every frame you keep is a frame the model processes, so a tight eight-second cut becomes a finished deepfake video far sooner than the whole scene.
Upload one clear face
The source is a still image, not a clip: one front-facing shot, both eyes visible, jawline unobstructed. That single frame sets the identity for every frame of the deepfake video that follows, so sharpness here pays off hundreds of times over.
Render, then scrub
Most clips finish in one to five minutes depending on frame count. Play the result back slowly first — a deepfake video that looks flawless paused can still stutter in motion.
Three steps, and the first matters most. If a deepfake video comes back unstable, the answer is almost always a steadier source shot rather than a second render of the same clip.
Where Footage Beats a Photograph
A deepfake video is a format, not a purpose. These are the jobs where moving footage earns the extra render time, all of them inside the consent rules below.
Dialogue and reaction shots
A face that speaks, blinks and reacts carries far more than a portrait, which is why a deepfake video lands harder than a still when it works and fails harder when it does not. Keep the take short and the camera locked off.
Casting and pitch material
Casting teams need to see an actor move before they commit. A rough deepfake video answers in five minutes what a painted concept frame only gestures at, long before anyone signs a contract.
Fan recuts and crossover edits
Recasting a scene, finishing a cosplay reel, assembling the trailer nobody funded. Trim to the shot that carries the joke — the deepfake video costs the same either way, but your afternoon does not.
Short-form vertical clips
Nine seconds of vertical video is a fraction of the frames a full scene demands, so the render is quick. Most social deepfake video work lives at this length.
Teaching people to spot the fakes
Journalists, teachers and trust-and-safety teams need to see the artefacts in motion to recognise them. Making a deepfake video and watching where it breaks is faster than any explainer, which is why detection training runs on generated material.
Working With Footage, Not Against It
Frame count is the entire cost model
The unit of work in a deepfake video is the frame — not the second, and certainly not the megabyte. A twelve-second clip at twenty-four frames per second is two hundred and eighty-eight separate face transfers; the same twelve seconds shot at sixty is seven hundred and twenty. Both files might be identical in size on disk, and the second takes two and a half times as long. When deciding what to upload, count frames rather than seconds.
That makes trimming the highest-leverage decision available. Most footage carries its idea in a few seconds surrounded by material nobody needs, and every spare frame is processing time spent for nothing. Cut to the shot, then render. As a rough guide, ten seconds of a deepfake video comes back in about a minute, a full minute of footage takes several, and anything past a couple of minutes is better split into the shots you actually intend to use.
Why a cheap deepfake video flickers
Flicker is not a blending problem, which is what makes it so persistent in tools that treat it as one. It comes from detection. On a handful of frames the face is blurred, turned too far, or half in shadow, the detector does not register it confidently, those frames pass through unchanged, and the deepfake video strobes. One unswapped frame in thirty is invisible as a still and glaring in motion, because human vision is built to notice change over time far more keenly than detail inside a single image.
The three reliable causes are motion blur from a moving camera or a fast turn, hard profile angles where one eye disappears behind the nose, and shadow crossing the face as the subject moves through a light. All three are properties of the footage rather than of the model, which is why a steady, evenly lit shot produces a stable deepfake video on the first attempt and a handheld take in mixed light rarely does, however many times you re-run it.
Review a deepfake video the way an editor would
Watch the deepfake video back at quarter speed before you judge it, and check in a fixed order rather than staring at the whole frame. Hairline first, during a turn — that is where the blend mask ends and where a halo or a hard edge announces itself. Then the eyes across a blink, then the teeth during speech, then the neck seam while the head moves, then the ears, which models routinely soften into shapes that do not survive a profile.
The last consideration is delivery. Whatever platform you upload to re-encodes the file again, and compression is hardest on the high-frequency detail around a blended edge. A deepfake video that holds up in your own player can develop visible blocking around the jaw once a social platform has finished with it, so export at the highest quality allowed and let the platform do the reducing.
Picking Footage a Deepfake Video Can Use
The source still deserves more attention than its file size suggests. One sharp, front-facing frame — both eyes visible, no sunglasses, no microphone across the chin, lit from the front rather than hard from one side — sets the identity every frame of the deepfake video inherits. A soft or angled reference does not degrade the output slightly; it degrades every frame equally.
The target clip is judged on how the face behaves over time. Watch for the face shrinking as the subject walks away, for a hand or a cup passing in front of it, and for a second person the detector could latch onto. A shot where the face stays the same size, stays unobstructed and belongs to one person yields a usable deepfake video far more reliably than a prettier shot that breaks any of the three.
- Trim to the one shot that carries the idea before you render
- Prefer a locked-off camera — handheld motion blur is the main cause of flicker
- Keep the face a usable size for the whole clip, not just the opening frame
- Use one sharp, front-facing still as the source face for the deepfake video
- Avoid takes where a hand, microphone or hair crosses the face
- Scrub the finished deepfake video at quarter speed before you publish it
Rules That Apply to Every Clip
Sexual and pornographic deepfake video content is banned here. Not discouraged, not gated behind a plan — banned, filtered at the model level, and grounds for losing access to the account that attempted it. No setting, tier or workaround changes this. The ban applies whether or not the person depicted is famous, because non-consensual intimate imagery is the harm this technology does the most damage with.
The second rule is presentation. A deepfake video published as an obvious edit is parody, commentary or fan work, and all three are legitimate. The same file published as authentic footage is deception, and it is already illegal in a growing number of jurisdictions — election-integrity statutes, non-consensual imagery laws and likeness rights all reach it, and the list of countries prosecuting it grows every year. Label edited footage as edited, clear likeness rights before anything commercial, and expect us to honour takedown requests from the people depicted.
Deepfake Video FAQ
How long does a deepfake video take to render?
One to five minutes for most clips. Frame count is what matters, not file size: a ten-second deepfake video at thirty frames per second is three hundred passes and usually lands in about a minute. Sixty frames per second doubles that.
What can I upload, and how long can the clip be?
MP4 or MOV up to 100MB for the footage, plus a still image up to 10MB for the source face. Running time is limited only by that budget, but longer clips mean more frames — trim a deepfake video to the shot you need.
Does the voice change?
No. The audio track is written back untouched and at the original level. A deepfake video from this tool is a face transfer only: no voice cloning, no speech synthesis, no lip sync, so the mouth follows the take that was filmed.
Why does my deepfake video flicker?
The detector drops the face on a handful of frames. Motion blur, a hard profile turn and shadow crossing the face are the usual culprits, and all three come from the footage rather than the model. A steadier, evenly lit shot fixes a flickering deepfake video; re-rendering does not.
Is there a watermark, and what happens to my files?
Free renders carry a small aideepfake.io mark and an active subscription removes it — the switch is in the tool, labelled before you upload. The clip itself comes back at the resolution you uploaded on every tier. Your face reference and footage produce that one render and nothing else — not training data, not resold, not published. A deepfake video you make here stays yours.
Can I make explicit or pornographic content?
No. Sexual deepfake video content is banned outright, filtered at the model level, and grounds for immediate loss of access. No plan, setting or support request changes that, and it applies to every face, famous or not.
Is making a deepfake video legal?
It depends entirely on what you do with it. Parody, commentary, research and clearly labelled fan work are broadly protected. Passing a deepfake video off as genuine footage is deception and is now illegal in a growing number of jurisdictions, as is sexual material and uncleared commercial use of a likeness.
Related Tools
The same engine behind this deepfake video tool, pointed at different source material.
Make Your Deepfake Video
Trim your clip, upload one clear face, and have the deepfake video back before the caption is written.
Start rendering