How to Make the Ozil AI Video From One Still Photo
Ozilai TeamStep-by-step guide to making the Ozil AI video: the photo that survives image-to-video, the prompt beats, the credit cost, and where to post it.
Making the Ozil AI video means turning a single photo into a short clip of that person scoring Mesut Özil's famous 2016 goal — the chip over the goalkeeper, the two sliding defenders, the pass into an empty net. On Ozilai.video this is a photo-to-video generation: you upload the photo, describe the move in a prompt, and ByteDance's Seedance 2.0 animates it into a cinematic video. Two things decide whether the result looks real: the photo you choose and the prompt you write. One standard generation — 720p, 12 seconds — costs 10 credits, new accounts start with free credits, and you can run the whole workflow before spending anything.
Générez le vôtre ici
Choisissez une photo claire et bien éclairée de la personne à qui appartient le but.
Téléchargez une photo pour continuer.
En générant, vous confirmez que vous avez la permission d'utiliser cette photo.
Votre clip apparaît ici
Un rendu avec les paramètres par défaut coûte 30 crédits.
What the Ozil AI Video Trend Is, in One Paragraph
Mesut Özil's goal for Arsenal against Ludogorets on 2 November 2016 is five seconds of football that almost nobody needs explained. Mohamed Elneny lifts the ball over the top; Özil runs onto it, chips the onrushing goalkeeper Milan Borjan, rounds the two defenders who slide in desperately, and passes the ball into the empty net. The Ozil AI video trend takes that exact sequence — the same camera angle, the same choreography — and swaps Özil's face for someone else's, usually from one still photo. The person in the clip never played the game, and the joke is that the footage reads as a real broadcast highlight anyway.
You are not building a football scene from nothing. You are casting one face into a template the audience already recognises, and every choice below exists to protect that illusion. If you want the wider context first, the trend explainer covers where the format came from and why it spread.
Choose a Photo the Model Can Animate
Image-to-video is not a magic trick; it is a model making hundreds of small guesses about a face you gave it once. Your job is to remove guesses. A photo that answers "who is this person?" in the first glance will carry through a moving shot far more reliably than a beautiful photo taken from twenty metres away.
What to look for:
- One subject only. Group shots force the model to choose, and it often chooses badly — or blends two faces into a stranger.
- Head and shoulders, or a bust crop. The face should occupy a meaningful share of the frame, not a corner of it.
- Even, front-facing light. Soft daylight or a well-lit indoor shot. Heavy side shadow, strong backlighting, and blown-out highlights all flatten identity.
- An unobstructed face. No sunglasses, no hood pulled up, no hand or phone covering the jaw. Beards, glasses worn normally, and hair over the ears are all fine.
- Sharp focus. Motion blur and low-light noise are the two things the model will faithfully reproduce and then amplify as the clip moves.
- A plausible expression. A relaxed, neutral face or a slight smile animates into an action far better than an exaggerated grimace.
Why a head-and-shoulders shot survives the model better
The technical reason is pixel budget. During a photo-to-video generation, the face region is what carries identity from frame to frame; the body, the ball, and the pitch are largely invented. In a tight head-and-shoulders crop, the face gets most of the available detail, so the model has the strongest possible anchor to hold onto while it generates legs that sprint, a shirt that moves, and a ball that travels. In a full-body photo the face is a few dozen pixels wide, the anchor is weak, and the identity drifts by the second or third second of the clip.
There is a second reason, and it is about composition rather than pixels. A wide photo already contains its own scene, and the model has to reconcile your background with the stadium it is trying to build. A tight crop discards most of the background, so the model is free to construct the pitch and the crowd without fighting your bedroom wall for control of the frame.
What you need versus what the model needs from you
The division of labour is simpler than it first looks. You supply identity and intent; the model supplies motion, environment, and camera.
| Input | What you need | What the model needs from you | | ------------- | ------------------------------------------------------------------------------------------------------------ | ----------------------------------------------------------------------------------------------------------------------- | | Photo quality | A photo that looks good on a phone screen and clearly reads as one person | A single, well-lit, sharp, front-facing subject with the face filling a decent share of the frame | | Prompt detail | Enough description that the clip matches the trend an audience already knows | Named action beats in order — the runner, the goalkeeper coming out, the chip, the finish — plus one camera instruction | | Aspect ratio | A destination decided up front: vertical 9:16 for short-form, or the clip's native frame for everything else | A source photo whose framing leaves room around the subject, so a vertical crop does not cut the action out of shot |
Read that table as a checklist before you press generate. In practice, most disappointing results trace back to the left-hand column being taken for granted while the prompt is rewritten for the fifth time.
Write the Prompt Around the Four Beats
The prompt is not where you get creative — it is where you get specific. The Ozil goal has four beats, and the model needs to hear all four in order. Miss one and you will get a generic striker running with the ball, which is not the trend.
- The long ball over the top. Say who plays it and that it is lofted beyond the last defender.
- The chip. Name the goalkeeper rushing out and the ball lifted over him. This beat is the trend's signature.
- Two sliding defenders. The ball is lifted or nudged past defenders who have gone to ground.
- The finish. The ball is passed into an empty net while the goalkeeper is still recovering.
Then add exactly one camera instruction. Broadcast language works best because that is what the audience's eye expects: "long lens from the halfway line", "camera tracks left to right with the run", "steady broadcast camera, wide enough to keep the goalkeeper in frame". Two camera instructions fight each other; one is enough.
A working prompt shape looks like this:
A football broadcast highlight. A lone striker runs onto a lofted
long ball played over the top of the defence, the onrushing
goalkeeper dives out to close the angle, the striker chips the ball
over him, two defenders slide in and the ball is lifted past them,
then rolled into the empty net. Long lens from the halfway line,
camera tracks left to right, floodlit night stadium.
One thing worth internalising: keep the subject count low and the sentence structure plain. Sentences that try to describe four people moving in different directions at once are how you get the limb-count failures that make AI clips obvious.
What One Generation Costs
One standard generation — 720p, 12 seconds — is 10 credits. The clip you need for the trend is shorter than that, so a single generation is your unit of work, not a minute of footage.
| Plan | Price | Credits | Standard 12s generations | | ------- | --------- | ------- | ------------------------ | | Starter | $9.90/mo | 20 | 2 | | Basic | $19.90/mo | 60 | 6 | | Pro | $39.90/mo | 200 | 20 | | Studio | $79.90/mo | 500 | 50 |
Yearly plans are cheaper per credit ($79 for 240, $159 for 720, $319 for 2,400, $639 for 6,000), and there are one-time credit packs — $9.90 for 10, $19.90 for 30, $39.90 for 100, $79.90 for 250, $149 for 600 — if you would rather not subscribe for a one-off joke. New accounts get free credits on sign-up, which is normally enough to land one or two attempts.
The practical consequence is that iterating is a budget, not a free loop. Two or three generations is a comfortable working range: one to see what the model does with your photo, one after fixing the biggest miss, one to polish. Full plan details live on the pricing page, and how credits work explains what a credit buys in more depth.
When the First Result Misses
Almost nobody gets the trend on attempt one, and that is not a sign that the approach is wrong — it is a sign that one variable is off. Diagnose before you regenerate, because a random rewrite wastes credits and teaches you nothing.
- The face drifts or melts mid-clip. Your source crop is too wide or too dim. Crop tighter onto the head and shoulders, and pick a brighter photo.
- The clip is just a person running with a ball. The chip is missing from the prompt. Name the goalkeeper, name the chip, and put that beat in its own clause.
- The body looks impossible — extra limbs, a knee bending backwards. The prompt is carrying too much simultaneous action. Cut it back to the four beats and one subject.
- The action happens off-frame or too far away. Add a camera instruction and ask for a frame wide enough to keep the goalkeeper visible.
- The colours are wrong. Describe the kit colours and the light condition in plain words; do not rely on the model to infer them from the photo.
- It looks right but feels flat. Try a second generation with the same prompt but a slightly different camera line. Small, single-variable changes are how you steer.
The rule that saves the most credits: change one thing per attempt. If you rewrite the prompt, keep the photo. If you swap the photo, keep the prompt.
The Six-Step Walkthrough
- Decide where the clip is going. TikTok, Instagram Reels, and YouTube Shorts are all vertical, so if that is the destination, plan a 9:16 crop from the start and frame the photo with room around the shoulders.
- Pick and crop the photo. Head and shoulders, one subject, even light, sharp focus. Crop in the photo app on your phone before uploading — that is the single highest-leverage minute of the whole process.
- Write the prompt using the four beats, then append one broadcast camera instruction. Keep it plain and ordered.
- Generate once. A standard 720p, 12-second generation costs 10 credits. Watch the result against the four beats, not against your hopes.
- Fix the single biggest miss and regenerate. One variable at a time. Most clips are one crop or one clause away from working.
- Export, crop to vertical, and post. Name the trend in the caption so the format is legible to viewers who have not seen it yet. The TikTok posting guide covers framing and captions in more detail.
Where the Clip Goes Next
Short-form is where the format lives. TikTok, Instagram Reels, and YouTube Shorts all expect vertical 9:16, and a 12-second clip fits inside every one of them without trimming — it is comfortably shorter than the maximum length on any of those surfaces. That single clip drops into all three with the same export.
Two practical points about the vertical crop. First, keep the goalkeeper in frame: the chip only lands as a joke if viewers can see the ball going over a person, so crop vertically around the action, not around the striker's torso. Second, respect the interface. TikTok and Reels both overlay captions and buttons along the bottom and right edges, so keep the subject and the ball in the middle band of the frame. Crop too tight and the funniest part of the clip ends up underneath a caption.
For longer-form channels — a YouTube upload, a blog post, or a reel of several attempts — keep the native 16:9 output, since that is the frame the model generated and the one where the stadium reads best. Square 1:1 works on feed posts and in group chats where a vertical video gets awkwardly letterboxed.
If you want to push the same workflow beyond this one goal, turning any photo into a football highlight shows how far the photo-to-video route stretches.
FAQ
How long does it take to make the Ozil AI video?
Generation happens on Ozilai's servers — you are not editing frame by frame. The time you actually spend is on photo selection and prompt writing, which is why this guide spends most of its words on those two steps rather than on the render. Choose the photo once, write the four beats once, and the second and third attempts take a fraction of the effort of the first.
Do I need video editing software?
No. The output is a finished clip. You need a phone photo app to crop, and nothing more unless you want to add a caption or trim. Vertical cropping for short-form is a one-tap operation in the native camera roll tools on both iOS and Android.
Can I use a photo of someone else?
You control what you upload, and you are responsible for having the right to use it. The trend's humour usually comes from casting a face nobody expects in Özil's boots, which in practice means photos of friends and colleagues rather than photographs you found on the internet. Using a recognisable person's likeness without permission is a bad idea regardless of how harmless the joke seems.
How many credits does one Ozil AI video cost?
A standard 720p, 12-second generation is 10 credits, so one attempt is 10 credits and a comfortable three-attempt workflow is 30. The Starter plan includes 20 credits a month, Basic includes 60, Pro includes 200, and Studio includes 500. New accounts get free credits on sign-up, which covers your first attempts before you commit to a plan.
Why does my result look nothing like the trend?
Check the prompt for the chip. If the goalkeeper and the chip are not named explicitly, the model will happily generate a striker dribbling toward goal, which is a different video entirely. If the move is right but the person is not, the problem is the photo: crop tighter, use brighter light, and make sure exactly one face is visible in the frame.