Deepfake Photo That Survives a Zoom
One picture, one face, one pass. A deepfake photo comes back at the resolution you uploaded, because a still gets zoomed into in a way a moving frame never does.
A deepfake photo is a single frame in which one face has been rebuilt carrying somebody else's identity. One detection pass, one blend, one file out — no frame count, no tracking, nothing to average across. That makes a deepfake photo the cheapest job this engine runs and the fastest to judge: ten credits, a few seconds, a result you can open at full size. It is also the least forgiving format. Nobody pauses a video to inspect a hairline; everybody zooms into a picture.
With only one frame, every quality question collapses into photography. How many pixels does the face occupy? Has the file been through four rounds of JPEG re-saving? Is it a screenshot of a picture rather than the picture? Does the crop leave any jaw and hairline to work with? A deepfake photo is decided by those answers before a model runs, and none of them is a setting you can change afterwards.
Nothing is preloaded here — no roster of faces, no gallery, no example people. You bring a source face and a target picture, both files you are entitled to use, and the deepfake photo happens between those two uploads. The engine is FaceFusion 3.6.1 on rented GPUs, so the thirteen swap models, the pixel boost and the enhancers behind our video jobs render your deepfake photo too — a still just takes one pass instead of several hundred.
What a Deepfake Photo Gets You Here
Output at the size you uploaded
The deepfake photo keeps the target's dimensions, up to the 10 MB ceiling, and the swapped region is rebuilt pixel by pixel rather than pasted over. Image scale enlarges the file for print, image quality sets the encoder, and nothing is cropped.
Pixel boost up to 1024×1024
Most swap models train at 128 or 256 pixels, which is why cheap output goes soft the moment a head fills the frame. Pixel boost rebuilds the face region at up to 1024×1024 before compositing — on a deepfake photo, where the viewer can zoom, that is the difference between skin texture and a plastic mask.
JPEG, PNG and WebP, plus a note on HEIC
All three decode the same way, so pick by what the picture has been through: PNG once it has been edited, WebP for pixels without file size, JPEG when it came out of a camera rather than six group chats. HEIC is the exception — convert at full size first.
One face out of a group
The detector finds every face by default, which is wrong for a table of nine. Switch selector mode to reference, set the reference face position to the head you mean, and tighten reference distance if a neighbour gets caught. Everyone else leaves the deepfake photo untouched.
Private storage, nothing published
Uploads go to presigned storage scoped to your account, and results wait in your job history. No deepfake photo is shown to other users, promoted anywhere, or fed back into training. What you upload stays yours.
How to Make a Deepfake Photo in 3 Steps
Choose the source face
The source is a still: JPEG, PNG or WebP under 10 MB. Pick the shot where the face is largest, sharpest and closest to straight on. This one file sets the ceiling on how good the deepfake photo can be.
Upload the target picture
The target decides pose, lighting, framing and final dimensions. A deepfake photo inherits every photographic decision baked into that file, so choose the picture you want to end up with, not one you plan to fix later.
Render, zoom in, download
Ten credits, a few seconds. Open the deepfake photo at one hundred percent and check the hairline, the eye corners and the neck seam before anything else. Full resolution, no export tier.
Free accounts get ten welcome credits once and ten more daily, which is one deepfake photo a day at no cost. A video costs thirty, so stills stretch a free balance three times as far.
What People Make a Deepfake Photo For
Each of these starts with two pictures you are entitled to use, and each is a job a deepfake photo does better than any clip.
Print, canvas and framing
Wall work needs pixels, not motion. Push image scale up, switch frame enhancement on, and a deepfake photo holds at A3 where a screen-sized file falls apart. Work back from 300 pixels per inch at final size.
The group shot somebody missed
Someone was at the dinner and not in the picture, or blinked in the only frame that worked. Reference mode pins the deepfake photo to that one head and leaves the other eight alone.
Covers, thumbnails and posters
Creators need a face in a pose no stock library carries. Build the composition once, render the deepfake photo into it, skip the reshoot — at a third of what video costs.
Casting boards and pitch stills
A lookbook or a pitch deck needs one convincing frame, not a sequence. A deepfake photo answers the question in seconds and reads more honestly than a mood board of borrowed press shots.
Teaching people what to look for
Trust-and-safety teams, newsroom staff and students have to see the artefacts before they can spot them. Making a deepfake photo of your own face and hunting for the seam teaches more in ten minutes than a week of reading.
Getting a Deepfake Photo Right
Resolution, crop and what survives the round trip
What decides a deepfake photo is not the file's megapixel count, it is how many pixels wide the face sits inside it. A head 500 pixels across in a 2000-pixel picture gives the model far more than a head 150 pixels across in a 6000-pixel landscape. Cropping closer hands the detector a bigger face; cropping too close cuts away the jaw, ears and hairline the blend hides its edges against. Leave a head's width of margin all round.
Aspect ratio survives the round trip untouched — 3:2 comes back 3:2, a 9:16 phone picture comes back 9:16, nothing letterboxed or re-cropped. Resolution survives too, up to the 10 MB ceiling, which is where people quietly lose detail: a 24-megapixel original at maximum JPEG quality can breach 10 MB, and the reflex fix is to re-save it lower. Scale the long edge down instead, or convert to WebP. A deepfake photo from full pixels at moderate compression beats one from fewer pixels at maximum quality.
Why a phone screenshot makes a poor source
Screenshots are the commonest single reason a deepfake photo comes back looking plasticky. A screenshot is not a copy of a picture, it is a capture of a screen's rendering of one: resampled to fit a viewport, re-compressed by the app that displayed it, clamped to the screen's colour space, and often scaled by the device pixel ratio into something crisp on the phone and soft anywhere else. The model cannot tell resampling artefacts from skin, so it reproduces them and the enhancer sharpens them.
Chat forwards do the same damage more slowly. Every hop re-encodes the JPEG, and every re-encode discards detail exactly where a deepfake photo needs it most: eyelashes, the wet line of the lips, the fine hair at the temple. One save out of a camera is invisible. Four are blocky, and no enhancer recovers what was thrown away — it invents a replacement. Ask whoever took the picture for the original.
EXIF, colour profile and what the file says afterwards
Metadata is not decoration. The orientation flag and above all the embedded colour profile travel with your upload, and that profile explains most cases of a deepfake photo whose swapped face reads warmer or cooler than the neck beneath it: a Display P3 source composited into an sRGB target shifts. Convert both files to one profile first and that complaint disappears. Orientation matters as much: a picture a phone shows upright but stores sideways reaches the detector sideways, so bake the rotation in before uploading.
The render comes back as a clean re-encode without the original camera metadata, which cuts both ways. Good for privacy: GPS coordinates and device serials never ride along into what you publish. Bad for provenance, because a stripped deepfake photo looks, to anything reading metadata, like any other exported file. Label edited media as edited, keep the original beside the deepfake photo so you can show your working, and treat the missing EXIF as a reason to disclose rather than an excuse not to.
Six Checks Before You Spend the Credits
Almost every disappointing result arrives with the files rather than coming from the engine. The models behave identically on every job, so when one deepfake photo looks photographic and the next looks pasted on, the difference was in what you uploaded. A still also gives the model nothing to hide behind: a marginal blend in a clip is smoothed over by the next twenty frames, while a deepfake photo holds it still for anyone to study.
Read the list below before you render. It takes fifteen seconds and heads off the attempts most people burn learning these lessons in the expensive order. Change one variable per attempt — a deepfake photo costs ten credits, so guessing is cheap, but guessing at four things at once tells you nothing.
- Face is several hundred pixels wide inside the source, not several dozen
- Source is the original file, never a screenshot or a chat-app forward
- Crop leaves jaw, ears and hairline in frame with room to spare
- Both files carry the same colour profile, and any HEIC is converted first
- Target is under 10 MB as JPEG, PNG or WebP — scale down, do not re-compress
- You hold the rights to both pictures and consent for the face in the deepfake photo
What You May Not Make Here
One rule is absolute: no sexual or pornographic content, ever. It is filtered at the model level and enforced on the account, no plan or advanced setting unlocks it, and accounts that try lose access permanently. Non-consensual intimate imagery is a crime in a growing list of jurisdictions and we act on reports from the people depicted. Two more lines are just as hard: never a deepfake photo of a minor, and never one onto an identity document.
The second rule is about presentation. A deepfake photo shared as an obvious edit is parody, commentary or fan work, and all three are legitimate. The same file passed off as an authentic photograph of something that never happened is deception, and it is illegal in a growing number of countries and US states — election-integrity statutes, likeness rights and synthetic-media disclosure laws all carry real penalties. That bites hardest on a stills page, because a deepfake photo offered up as evidence travels further than any clip. Label your work, get a release before commercial use, and never publish a deepfake photo of a real person doing something they never did.
Deepfake Photo FAQ
Is making a deepfake photo free?
Every account gets ten welcome credits once and ten more every day. An image costs ten credits, so that daily grant is exactly one free deepfake photo a day, at full resolution. Free renders carry an aideepfake.io mark that a subscription removes. Paid plans buy throughput, longer video and priority, not better stills.
Which formats and file sizes work?
JPEG, PNG and WebP up to 10 MB, for both the source face and the target. HEIC from an iPhone needs converting first. A video up to 100 MB can be a target too, but that is the video route, not a deepfake photo.
Is the output good enough to print?
Within the resolution you gave it, yes. The result keeps the target's dimensions, image scale enlarges it and frame enhancement lifts detail, but nothing invents pixels the original never held. Work back from 300 pixels per inch at final size — a 900-pixel picture never becomes a poster, however good the deepfake photo looks on a phone.
How long does a deepfake photo take?
Seconds, not minutes. One still is one detection pass and one blend, with no frame count to multiply it out. Face and frame enhancement add processing time but never extra credits. Only the video route runs into minutes.
What happens in a picture with several faces?
By default every detected face is swapped. To change one person, set selector mode to reference and give the reference face position of the head you mean, then raise reference distance if a neighbour gets caught. Everyone else leaves the deepfake photo as they entered.
Can I make sexual or pornographic content?
No. Sexual deepfake material is banned outright, filtered at the model level, and grounds for permanent loss of access. No plan or setting changes that, reports are acted on, and the ban covers a deepfake photo of anybody at all — including one of yourself.
Is making a deepfake photo legal?
It depends on what you do with it. Parody, commentary, research and clearly labelled fan work are broadly protected. Passing a deepfake photo off as a genuine photograph, using a real likeness commercially without a release, or producing sexual material is not — a growing number of countries and US states prosecute exactly that.
Related Tools
The same engine behind this deepfake photo, pointed at a different starting point.
Make a Deepfake Photo
Upload a face and a picture. Today's free credits cover one full-resolution deepfake photo.
Start rendering