Ozil Goal AI Prompt Guide: How to Prompt Seedance 2.0
Ozilai TeamWrite Ozil goal AI prompts that render cleanly on Seedance 2.0: the six prompt components, three complete prompts, and the fixes for an unnatural clip.
You do not need editing skills to put someone in Özil's boots — you need a prompt that describes motion. Text-to-video builds the whole scene from your words, while image-to-video starts from one of your photos and animates forward from it. This guide covers the six components of a working Özil-goal prompt, three complete prompts you can paste into Seedance 2.0, and what to change when the result looks wrong.
Gere o seu aqui
Escolha uma foto clara e bem iluminada da pessoa a quem o gol pertence.
Carregue uma foto para continuar.
Ao gerar, você confirma que tem permissão para usar esta foto.
Seu clipe aparece aqui
Uma renderização com as configurações padrão custa 30 créditos.
Text prompts and image prompts do not do the same job
Özilai.video runs ByteDance's Seedance 2.0 model in two modes, and the prompt has a different burden in each one.
In text-to-video mode the model casts the scene. It invents the player's face, build and kit, the pitch, the stadium and the light, because your words are the only anchor available. A text prompt therefore has to carry identity and setting as well as the action.
In image-to-video mode your still becomes the first frame. The face, kit, lighting and background already exist, so your prompt only governs what happens afterwards: the run, the camera, the speed of events, and where the shot ends.
That difference is the most common reason a prompt that works beautifully in one mode produces a mess in the other.
What the model receives in each mode
- Text-to-video: your words alone. Two runs of the same prompt can produce two different-looking players, because nothing pins the subject.
- Image-to-video: your words plus your photo, which is frame one. Identity and setting are already decided, so sentences describing them are wasted.
The practical rule: in text mode, spend words on who and where. In image mode, spend words on how it moves and where the camera goes.
The anatomy of a prompt that works
Every reliable football prompt has the same six parts. Because it is a lot to hold in your head at once, write them in the same order every time. When a clip misses, you then know which line to change instead of rewriting the whole thing.
| Prompt component | What it controls | Example wording | | ------------------- | ---------------------------------------------- | -------------------------------------------------------------------------------- | | Subject | Who is on screen and what identifies them | "the player in the photo, red-and-white kit, short dark hair" | | Action | The physical motion, written as verbs in order | "receives a long ball over the top, lifts it over the goalkeeper, keeps running" | | Setting | Surface, surroundings and time of day | "worn green pitch, packed stadium behind the goal, night match" | | Camera | Framing, height and movement | "camera tracks right at chest height, keeping him centred" | | Lighting | Mood and where the light falls | "cool floodlights from behind, soft rim light on his shoulders" | | Duration and pacing | How long the shot runs and how fast it feels | "one continuous 12-second shot, no cuts, half speed" |
Duration is the one component you also set when you generate rather than only describe in prose. A standard 720p run is 12 seconds, and that is what the rest of this guide assumes.
Describe motion, not just a scene
A scene description is a photograph written in words: "a footballer on a pitch at night". It tells the model what the frame should look like, not what should change between the first second and the twelfth. Video generation needs change, so the prompt needs verbs — receives, sprints, lifts, drops, slides, rolls, celebrates — and it needs them in the order you want them to happen.
Two habits do most of the work:
- One dominant motion per clip. When three things happen at once, the model has to resolve all of them in every frame, and limbs, balls and shirts start drifting.
- Sequence beats instead of stacking them. "Özil chips the keeper while two defenders slide in and the crowd erupts" asks for three simultaneous events. "Özil lifts the ball over the rushing goalkeeper, then the ball drops toward the empty net" asks for one thing, and then the next.
Camera language is motion too. "Camera tracks right, level with his chest" is far more useful than "cinematic angle", because it describes how the frame itself changes over the twelve seconds.
Three complete prompts you can paste
All three are written for the same famous sequence: the long ball over the top, the chip over the onrushing goalkeeper Milan Borjan, the two defenders sliding across the path, and the ball passed into an empty net. Adapt the framing rather than the structure.
1. Photo-anchored run (image-to-video)
Use this when you already have a still of your subject. The kit, face and setting come from the photo, so the prompt spends all its words on movement.
The player in the photo runs onto a long ball played over the top of the defence.
Camera tracks right at chest height, keeping him centred.
He takes one controlling touch, slows for a step, lifts the ball over the
onrushing goalkeeper, then accelerates past two defenders who slide across his path.
The ball rolls into the empty net.
Overcast late-afternoon light, cool green pitch, shallow depth of field.
Single continuous shot, no cuts, steady motion, no camera shake.
Why it works: it never re-describes the face or the kit, it gives the camera exactly one instruction, and the action is ordered as a sequence. If the run drifts or the extra limbs appear, delete the "past two defenders" clause first — it is the busiest beat in the prompt.
2. Text-only stadium scene (text-to-video)
Use this when you have no usable photo. Describe the player's appearance instead of typing a name, because the description is what the model actually renders.
Night match under floodlights in a packed stadium.
A midfielder in a red-and-white kit, short dark hair, medium build, receives a
long ball over the top of the defence and sprints toward goal.
The goalkeeper, in yellow, charges off his line; the ball is lifted calmly over him.
Two defenders slide in and miss. The ball rolls into an empty net.
Wide shot from the halfway line, camera pushes in slowly and tracks right.
Cool white floodlights, vivid green pitch, crowd blurred in the background.
One continuous 12-second shot, no cuts.
Why it works: subject, setting and lighting are all spelled out, because nothing else anchors them. Keep each of those lines to one detail — one kit colour pair, one hair description, one camera move — so the model is not asked to reconcile competing instructions.
3. Slow-motion celebration
Use this for the moment after the ball hits the net, which is the part most people post on its own.
Slow-motion shot of the player in the photo sliding on his knees across wet grass
toward the corner flag, arms out, head back.
Camera at knee height, pushing in slowly, 50 percent speed.
Floodlights behind him give a bright rim light on his shoulders.
Teammates run in behind him, out of focus.
Single continuous shot, no cuts, no camera shake, no zoom.
Why it works: celebrations are almost pure motion, and a low, slow camera keeps the model from inventing a second subject. If the teammates start morphing or merging into the player, delete the "Teammates run in behind him" line and the clip will usually hold.
When the output looks unnatural, change one thing at a time
Retries cost credits, so do not change five variables at once. Pick one of these, re-run, and compare.
- Shorten the duration. Twelve seconds is a long time to hold one continuous action. Generate a shorter version first, confirm the motion reads correctly, then lengthen it.
- Simplify the action to one verb. Reduce the prompt to "sprints onto a long ball over the top" or "lifts the ball over the goalkeeper". Add beats back one at a time until it breaks.
- Reduce to one subject. Remove teammates, defenders and crowd descriptions. Every extra figure is another thing the model has to keep consistent across 300-plus frames.
- Separate simultaneous events. If two things happen in the same sentence, split them into two sentences so they occur in order rather than at the same moment.
- Steady the camera. Replace "tracks right while pushing in" with a single calm move. Compound camera moves amplify every small error in the subject's motion.
If none of that helps, the problem is usually the input rather than the wording. For photo-anchored runs, read the image-to-video photo checklist before you spend more credits.
Prompt length does not change what you pay
Credits track how much video the model has to render — duration and resolution — not how many words you type. A 30-word prompt and a 300-word prompt for the same 12-second 720p clip cost exactly the same 10 credits. What you write buys quality; what you choose in the generation settings buys cost.
That means prompt writing is the one part of the workflow you can experiment with freely, as long as you are deliberate about how often you hit generate. The plan structure reflects that:
| Plan | Price | Credits per month | Standard 720p 12s runs | | ------- | --------- | ----------------- | ---------------------- | | 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 billing and one-time packs exist too, starting at $9.90 for 10 credits — exactly one standard run — and going up to $149 for 600 credits; the full breakdown lives on the pricing page. New accounts also start with free credits, which is the cheapest way to test your wording before committing. For the mechanics of how consumption works, see how AI video credits are explained.
FAQ
Do I need a photo of Özil himself to make this clip?
No. The trend is about putting whoever you like into the same sequence, and Ozilai never requires the real player's likeness. In image-to-video mode the person in your photo is the subject; in text-to-video mode you describe a player's appearance in words. Start from the Ozilai home page and use the free sign-up credits to try one run.
How long should a prompt be?
Long enough to cover the six components and no longer. Photo-anchored prompts often land in the 30-60 word range, because identity and setting are already in the frame. Text-driven prompts need more, since the model has to be told who is on screen and where the shot takes place.
Can I reuse the same prompt in both modes?
Partly. The action and camera lines travel unchanged. The subject, kit and lighting lines matter in text mode and become unnecessary noise in image mode, where the photo already shows them. Strip any sentence describing something your photo visibly contains.
Why does the end of my clip melt or warp?
Long single shots accumulate small errors, especially in hands, feet and the ball. Shorten the duration, remove secondary subjects, and keep the camera to one move. Holding a clean eight-second version beats fighting a twelve-second one. If you want a longer edit for a specific platform, see how to share an Ozil AI video on TikTok.
Does a longer prompt cost more credits?
No. Cost comes from duration and resolution. A standard 720p 12-second generation is 10 credits whether your prompt is one line or ten. The only thing a longer prompt changes is how much the model has to reconcile at once, which is why shorter prompts are often the more reliable ones.