Seedance 2.0 Upscaled, Tested: Where 480p-to-4K Holds Up, and Where It Doesn't
Seedance 2.0 Upscaled from 480p to 720p, 1080p and 4K, tested: faces and fur hold up, signs and crowds don’t. Timing, prices and the API parameters.
By OceanToken Team

Seedance 2.0 Upscaled makes 1080p clips for less than native 1080p. It gets there by generating the clip at 480p and upscaling it with AI.
The question is what that does to the picture. We ran nine jobs across 720p, 1080p and 4K and looked at every frame at 100%. Faces and fur came out close to native. Shop signs, crowds and dense foliage did not.
| 9 / 9 | ~6 min | 24 fps | 720p–4K |
|---|---|---|---|
| Jobs completed, up to 5 at once | Average time to a finished clip | 97 frames for 4 seconds | Output sizes |
01 · How it works
You send a prompt as usual. Seedance 2.0 generates the clip at 480p, then an AI upscaler takes it to 720p, 1080p or 4K.
480p from Seedance 2.0 → AI upscale → 720p, 1080p or 4K
Seedance decides everything you see: composition, motion, lighting. The upscaler only enlarges the frame and fills in fine detail. It doesn't change what happens in the shot.
What we measured, on 4-second 16:9 clips:
| Output | Size | Frame rate | 4-second file | Time to finished clip |
|---|---|---|---|---|
| 720p | about 1280 × 720 (exact width a few pixels less) | 24 fps | 97 frames, 5–8 MB | 4.6–5.4 min |
| 1080p | 1920 × 1080 | 24 fps | 97 frames, about 12 MB | 5.2–6.1 min |
| 4K | 3840 × 2160 | 24 fps | 97 frames, 48–72 MB | 6.9–8.0 min |
Every clip came back at the length we asked for. Audio is optional: with generate_audio on, the file carries an AAC track.
02 · Where it holds up
Each case below shows the full frame, then a crop of the marked area at 100% of the delivered resolution. Left: the 480p clip enlarged the ordinary way. Right: the upscaled clip.
Close-up portrait, 1080p: works well
Close-up portrait of an elderly fisherman with deep wrinkles, grey stubble and weathered skin, soft window light

Wrinkles, stubble and the fine lines around the eyes all come back. This is the kind of shot the upscaler is best at: one clear subject, close to the camera, with natural texture.
Animal macro, 1080p: works well
Macro shot of a fluffy orange cat face, individual whiskers and fur strands visible, green eyes blinking

Enlarged the ordinary way, the fur is a smear. Upscaled, you can pick out single hairs and whiskers, and the highlights in the eyes are cleaner.
Lighthouse at dusk, 4K: works well
A lone lighthouse on a rocky coast at dusk, waves crashing, slow camera push in, detailed textures

Rock faces and the foam on the waves sharpen up, and the shadows keep their detail. Natural textures like stone, water and sky hold up even at 4K.
Skateboard trick, 720p: small gain
A skateboarder doing a kickflip in a sunny skate park, fast motion, camera follows

Fast motion is already blurred in the source, so there's less to recover. The board graphic and edges get crisper; the rest looks about the same.
03 · Where it doesn't
Tokyo street at night, 4K: be careful
Busy Tokyo shopping street at night, many neon shop signs with readable text, crowds walking, rain reflections

Sharper than a plain enlargement, but the shop signs don't turn into readable text, and faces in the distance become patches of colour. Up close it looks smeared. For text and crowds, don't use the upscaled 4K.
Autumn forest from the air, 4K: be careful
Aerial drone shot over a dense autumn forest with thousands of red and yellow leaves, a winding river, morning mist

When the whole frame is tiny leaves, the detail the upscaler adds looks painted, almost like watercolour. It's fine at normal viewing size. Zoom in and it stops looking like real foliage.
04 · What it costs, and how to call it
OceanToken prices per 1M video tokens, at each resolution's starting tier, as of November 6, 2026. Upscaled prices are converted by the billing formula, width × height × (24 × seconds + 1) ÷ 1024:
| Output | Official native | Native on OceanToken | Upscaled on OceanToken | 4-second upscaled clip |
|---|---|---|---|---|
| 720p | $7.00 | $6.65 | $4.90 | $0.43 |
| 1080p | $7.70 | $7.315 | $5.39 | $1.06 |
| 4K | $4.00 | $4.00 | $2.80 | $2.20 |
Prices change and differ by tier; see each model's current price on the Models page.
At every resolution here the upscaled version costs clearly less per token than native on OceanToken. At 4K it is $2.80 against $4.00 at the starting tier and $2.64 against $4.00 at the top tier.
How to call it
Seedance 2.0 Upscaled uses the same /v1/videos endpoint as every other video model on OceanToken.
bytedance/seedance-2.0-upscaledfor 720p and 1080pbytedance/seedance-2.0-upscaled-4kfor 4K- Fast and Mini have upscaled versions too:
bytedance/seedance-2.0-fast-upscaledandbytedance/seedance-2.0-mini-upscaled
bash
curl https://api.oceantoken.ai/v1/videos \
-H "Authorization: Bearer $OCEANTOKEN_API_KEY" \
-H "Idempotency-Key: $(uuidgen)" \
-H "Content-Type: application/json" \
-d '{
"model": "bytedance/seedance-2.0-upscaled",
"prompt": "Close-up portrait of an elderly fisherman, soft window light",
"duration": 4,
"resolution": "1080p",
"aspect_ratio": "16:9",
"generate_audio": true
}'
# then poll GET /v1/videos/{id} until "status" is "completed"
The parameters that matter most:
| Parameter | Values | What it does |
|---|---|---|
resolution |
720p, 1080p |
Output size. Use the -4k model ID for 4K. |
duration |
4 to 15 seconds, default 5 | Clip length |
aspect_ratio |
16:9, 4:3, 1:1, 3:4, 9:16, 21:9, adaptive |
adaptive follows the first frame |
generate_audio |
true / false |
Adds an AAC audio track |
frame_images |
up to 2 | First and last frame |
input_references |
up to 9 images, 3 videos, 3 audios | Reference media |
omni_reference_task_type |
auto, reference, edit, extend |
What to do with the reference media: build a new clip, edit the video, or continue it |
Full request and response details are in the Seedance 2.0 Upscaled docs .
05 · The takeaway
Upscaling doesn't make a better video. It makes a bigger one. Composition, motion and lighting all come from Seedance at 480p, and the upscaler only sharpens what's already there.
That's why the results split the way they did. Faces, fur, products, rock and water were drawn clearly at 480p, so they hold up at 1080p and often at 4K. Small text, faces in a crowd and thousands of leaves were never clear in the 480p frame, so no upscaler can bring them back.
For most social, ad and product videos, that makes upscaled 1080p a sensible default. Keep native for the shots where every letter or face has to be readable.
| Your shot | Use |
|---|---|
| Faces, portraits, talking heads | Upscaled 1080p |
| Pets, products, food, a single clear subject | Upscaled 1080p |
| Landscapes, sea, sky, slow camera moves | Upscaled, up to 4K |
| Fast action | Upscaled 720p is enough; the source is blurred anyway |
| Readable text, signs, packaging copy | Native |
| Crowds, wide city scenes, dense foliage at 4K | Native |
A simple workflow: draft and pick your shots on upscaled 720p, then render the final cut at upscaled 1080p. Switch to native only for the shots with text or crowds.
Seedance 2.0, its upscaled versions, and Fast and Mini all run on one OceanToken API key and one balance. Your code stays the same when you switch: only the model ID changes, so you can render most shots upscaled and the hard ones native.
Compare native and upscaled side by side on the Seedance 2.0 page →
See every video model and price →
Tests run October 29, 2026: nine jobs, six shown here, all 16:9 and 4 seconds long. Crops are at 100% of the delivered resolution. Prices are OceanToken starting prices from the live catalog as of November 5, 2026; your console shows the rate for your tier.
More from the blog
Qwen Long Context: When to Fill 984K
Qwen long context in practice: Qwen3.8 lists a 984K-token window. When filling it pays off, when to chunk or retrieve, and a planted-fact test for your data.
GPT-6 Luna Use Cases: What It Can Handle
GPT-6 Luna use cases: the scoped, high-volume jobs it fits, where to move up to Sol or Astra, its effort and tool-calling limits, and how to test the switch.
Qwen Commercial Use: Which License Applies
Qwen commercial use license, model by model: which Qwen weights are Apache 2.0, what the Qwen, research and Qwen3.8-Max licenses add, and API terms.