Seedance 2.5 is ByteDance's newest video model, and it is running here now. Give it a prompt or a single photo and it returns one continuous take of up to 30 seconds — six times the length most video models hand back, generated in a single pass with no seams to hide.
Draft at 480p, then re-run the take you like at 720p. Sound can be generated with the clip. Everything runs in the browser and you pay per second in credits.
First Frame
Click to upload
18 credits per run at these settings

Customer Value
A basketball player driving to the hoop in a packed arena, low tracking shot, stadium lights.
No timeline, no keyframes, no render queue. Type or upload, set the length, and a finished MP4 comes back.
Name the subject, the camera move and the light. Seedance 2.5 rewards direction — 'slow push through a neon alley in the rain' lands better than a noun.
Drop in a first frame and the model animates the image you already approved, keeping its subject, colours and framing intact.
Anywhere from 4 to 30 seconds on a slider, in 9:16 for TikTok and Reels, 16:9 for YouTube, or 1:1 for the feed.
Pick 480p to find the shot and 720p for the one you are keeping. The clip arrives as an MP4, ready to post.
Seedance 2.5 is a text-to-video and image-to-video model built around length. Where the other models on Zorq AI end a take at 8 or 15 seconds, this one keeps going to 30 — long enough to hold a whole scene.
The headline capability, and it is the one that changes what you can make. Stitching two clips together is where identity slips — the face shifts, the jacket changes shade, the light jumps at the join. One pass has no join.
Turn audio on and the track is produced alongside the video rather than dropped underneath it afterwards, so the sound belongs to the scene it came from.
Upload a first frame — a product shot, a character, a landscape — and Seedance 2.5 adds the motion while the frame you designed stays the frame you get.
480p at 19 credits per second for finding the shot, 720p at 41 for the finished take. Switching tier does not mean re-typing anything.
9:16, 1:1 and 16:9 all render natively, so a vertical clip is composed vertically from the first frame instead of being cropped out of a wide one.
Long takes punish weak motion modelling — cloth, hair and momentum have thirty seconds to fall apart. Holding them together across a full scene is exactly what this generation was trained to do.
What the Seedance 2.5 generator accepts and what a run costs, so you can price a project before you start it.
4 to 30 seconds in one continuous take, set on a slider. Cost scales per second, so a 30-second piece costs six times a 5-second one at the same tier.
480p on fast, 720p on pro. Both run from the same box, and the usual workflow is to lock the motion cheaply and re-run the winner at 720p.
16:9, 9:16 and 1:1. Portrait is the default because most clips made here go straight to a vertical feed.
95 credits for a 5-second 480p take, 205 for the same clip at 720p, 570 for a full 30 seconds at 480p. Credit packs start at $6.9, and the $35.9 monthly plan carries 1,620 credits.
30 sec
longest single take on the site
19/sec
credits at the 480p tier
2 tiers
480p draft, 720p final
MP4
with optional generated sound
The previous generation, with a 480p mini tier and 1080p pro output.
The default video model here, strong on short cinematic beats.
Another text and image to video option, up to 15 seconds.
The workflow this model plugs into when you start from a photo.
Seven replacements compared, this model among them.
Push a finished clip to a higher resolution afterwards.
Length, cost, tiers and output format — the questions people ask before their first Seedance 2.5 render.
Seedance 2.5 is ByteDance's newest Seedance video model. It generates a short film-like clip from a written prompt or a single image, and its defining capability is length: one continuous take of up to 30 seconds instead of the few seconds most video models return.
Yes. It is wired into the generator on this page and into the main video tool, and you can pick it from the model list like any other. It runs online in the browser with nothing to install.
You pay per second of output. The 480p tier is 19 credits per second, so a 5-second take is 95 credits and a full 30-second take is 570. The 720p tier is 41 credits per second. Credit packs start at $6.9 for 170 credits, and the $35.9 monthly plan carries 1,620.
Between 4 and 30 seconds, in a single generation. That is the longest take available anywhere on Zorq AI — Veo 3.1 stops at 8 seconds and Kling O3 at 15 — which makes Seedance 2.5 the one to pick when a scene needs room to breathe.
The model is served through the same provider API this site runs on, which is how the generator above works. If you are looking to call it yourself rather than through a tool, that is the layer to look at; if you just want output, the generator here is the shorter path.
Pick Seedance 2.5 when length matters: a 30-second take, a full scene, anything where cutting between two generations would show. Pick Seedance 2 when you want a 1080p master of a short beat, since its pro tier renders at 1080p. They share the same prompt box, so trying both costs you nothing but credits.
Both. Start from a written prompt for a scene that does not exist yet, or upload a first frame to animate something you already have. Image to video holds your subject steady; text to video gives you more range.
It can. Switch the audio option on and a track is generated with the clip instead of being added in an editor afterwards. It costs extra credits, so most people leave it off while drafting and turn it on for the take they are keeping.
480p on the fast tier and 720p on pro. Draft at 480p until the motion is right, then spend the pro credits once on the version you are actually going to post.
Longer than a short clip from a smaller model, because you are asking for far more footage — a 30-second take is six times the frames of a 5-second one. Set the length you want, leave the tab open, and the finished MP4 lands in your library.
Direct it like a shot, not a search query. Name one subject, one clear action, the camera move and the light: 'a lone skater crossing an empty plaza at dusk, low tracking shot, long shadows'. Over-loaded prompts with five competing ideas are the main cause of muddy output.
Write the beat, not just the frame. Thirty seconds is long enough for something to change, so say what changes — 'she walks to the window, then the light shifts to evening'. A static description will give you a static half-minute.
Yes. Clips generated on a paid balance are intended for creator content, ads and client work, subject to the model's content policy. If a clip features a real person or a client's product, make sure you have permission to publish it.
Thirty seconds is the ceiling for one Seedance 2.5 generation, and it is the ceiling worth working to — a full minute is built by cutting two takes together in an editor, which is exactly the seam this model exists to avoid. Plan the piece as scenes of up to half a minute and cut on a deliberate change of shot rather than on a technical limit.
A standard MP4 that plays everywhere and uploads straight to social platforms with no conversion. The output keeps the aspect ratio you chose, so a vertical prompt stays vertical on export.
Usually the input. A blurry source frame or a prompt asking for too much at once leaves the model little to hold onto over a long take. Start from a sharp, well-lit image and one clear motion instruction, then add complexity once the base take looks clean.
Yes — a frame from any image generator works as a starting image. Build the character or scene as a still first, then animate it, and the clip keeps the colours, composition and style you designed in the source frame.
On paid Zorq AI balances, generated clips are yours to use for creator content, ads and client work, subject to the content policy.
Type a prompt or upload a photo, set the slider anywhere up to 30 seconds, and get a single continuous take back as an MP4. Draft at 480p while you are still finding the shot, then re-run the winner at 720p with sound. Credits work across every model on the site, so nothing is locked to a plan.