TryOn
One physical event per take
Sora resolves a single event properly. A walk, a turn and a door opening is three renders, and asking for all three in one prompt means the last one happens off screen.
Sora 2 video generator
Sora 2 is the OpenAI model people ask for by name, and it earns it on physical realism: weight, gravity, hair and cloth behave the way they do in a real room. Four, eight or twelve second takes at 720p, from a written prompt or from one of your photos.
Model
Image
0 / 2400
Aspect ratio
Duration
Resolution
Number of Results
Details21 credits
Log in to create and keep your videos.
Recent videos
The same model runs in both video studios: start from a photo here, or switch to the text studio and write the shot instead. Text to video
How to make a Sora 2 video in 3 steps.
01
A frame with weight in it
Sora 2 is judged on how things fall, pour and settle, so the first frame should contain something with mass: a body, cloth, liquid. A flat graphic gives it nothing to be physically accurate about.
02
Describe what resists the movement
Say what moves and what fights it — the weight of the coat, the drag of the water. Sora answers a sentence about resistance better than a list of adjectives, and twelve seconds is long enough to show the difference.
03
Check the idea at four seconds
Renders are slower and pricier per second than the budget models. Queue the four second version to test the action, then the twelve second one for the shot that has to pass as footage.
How to write a Sora 2 prompt.
OpenAI’s model is the one to reach for when a clip has to look photographed: weight, gravity, cloth and skin behave the way they do in a room.
Weight, cloth, skin.Something with mass in the frame
A person standing, a garment hanging, water in a glass: Sora 2 needs a subject whose weight it can model. A backlit silhouette or a flat lay gives it nothing to simulate, and the render comes back looking synthetic.
- ↗Start from a real photo, not a render — it inherits artefacts
- ↗Visible ground tells the model where gravity points
- ↗Skip mirror selfies; reflections take a whole take to resolve
Action plus resistance.Name the resistance, not the beauty
Describe what the action pushes against: the stiffness of the hair, the weight of the fabric, the drag of the water. Sora 2 turns that into believable motion, and it is the one thing cheaper models consistently miss.
What Sora 2 is used for
The expensive route earns its price on shots that have to look photographed: fashion, food, water, and anything a client will zoom into.
Fashion and fabric
Cloth is where cheap models show their seams. Sora 2 keeps a hem falling and a sleeve bending through a camera move, which is what makes a garment clip usable in a lookbook rather than a demo.
What people say about Sora 2
Notes from sellers, editors and designers who render stills instead of booking another shoot.
“I run ads from photos I already own. One product shot now gives me three versions to test, and a failed render comes back as credits instead of a lost afternoon.”
“The prompt is the whole job. Once I wrote "slow push in, soft window light" instead of "make it cinematic", the takes stopped needing five attempts.”
“The first frame decides everything. A clean frame with room around the product renders in one pass; a tight crop always comes back jittery.”
“I animate illustrations for a newsletter. Line work and palette stay put, which is more than most generators I tried managed.”
“The free daily render is what got me in. I tried a shot without paying, liked it, and bought credits the same week.”
“A 9:16 render at 1080p drops straight into the timeline. I use it for B-roll between the shots we actually filmed.”
“I run ads from photos I already own. One product shot now gives me three versions to test, and a failed render comes back as credits instead of a lost afternoon.”
“The prompt is the whole job. Once I wrote "slow push in, soft window light" instead of "make it cinematic", the takes stopped needing five attempts.”
“The first frame decides everything. A clean frame with room around the product renders in one pass; a tight crop always comes back jittery.”
“I animate illustrations for a newsletter. Line work and palette stay put, which is more than most generators I tried managed.”
“The free daily render is what got me in. I tried a shot without paying, liked it, and bought credits the same week.”
“A 9:16 render at 1080p drops straight into the timeline. I use it for B-roll between the shots we actually filmed.”
Sora 2 tips worth knowing
Physical accuracy is expensive per second: four ways to spend it on the right frame.
720p, three lengths
Takes come in 4, 8 or 12 seconds at 720p. Choose the length before the prompt: a four second prompt with three beats will lose two of them, and the refund does not come.
Silent by default
This route returns picture only. Plan sound in the edit, and keep spoken lines out of the prompt — the model will try to lip-sync them and drift off the words.
Draft elsewhere, finish here
Iterate the motion on Grok or Seedance and move the winning sentence to Sora 2. Paying twelve second prices for a test render is the fastest way to burn a credit balance.
Sora 2 questions, answered.
What the OpenAI model is better at, what it costs per second, and where to draft instead.
What is Sora 2?
Sora 2 is OpenAI’s video generation model, available here at 720p with 4, 8 or 12 second takes. It works in both studios: describe the shot in words, or upload a photo and let it animate from that frame. Renders take longer than the budget models, so queue a few and come back.
Can I use Sora 2 for free?
The daily free render is reserved for the cheapest model in the catalogue, so Sora is paid from credits — the price of the exact duration is shown before you submit, and a failed job is refunded. A free account is enough to top up.
What is Sora 2 actually better at?
Physics. Liquid, smoke, falling fabric and the way a body carries weight are where cheaper models show seams, and it is also unusually good at keeping a face consistent through a camera move. Keep prompts concrete: one subject, one action, one camera move per take.
Compare with another model
All AI modelsOne free try-on, one free video and one free image every day — no sign-up needed.