Seedance 2.5 4K on Segmind: native 4K video with synchronized audio
Seedance 2.5 4K is ByteDance Seed's video model, rendering natively at 3840x2160. Generate 4 to 30 seconds from a text prompt, a starting image, or references — images, video or audio — with synchronized native audio, delivered as 10-bit HEVC.
Seedance 2.5 4K is ByteDance Seed's video model, rendering natively at 3840x2160. It turns a text prompt, a starting image, or references — images, video and audio — into a 4-to-30-second clip with synchronized native audio. The detail is generated at full 4K, not upscaled from 1080p, and the file ships in 10-bit colour for grading headroom. The base seedance-2.5 covers 480p to 1080p; this slug renders at native 4K only and runs as an async job.
ModelSeedance 2.5 4KNative 4K video from text or images, with audio.What it does well
Native 4K detail, generated not upscaled
Seedance 2.5 4K renders at 3840x2160 from the start, so fine texture holds at full resolution instead of softening the way an upscale does. Fast motion and small details stay legible in the same shot. This macro clip ran at native 4K, 24 fps, for 4 seconds.
A ruby-throated hummingbird hovers at an orange trumpet flower in a shaded garden, iridescent green feathers and fast-beating wings crisp in extreme close-up; it slips its slender beak into the bloom, knocks a puff of pollen into the air, then darts upward out of frame. Steady macro shot. Natural sound of humming wingbeats and rustling leaves.
Iridescent feather barbs and airborne pollen stay crisp at full 4K while the wings blur with fast motion, over ambient wingbeat audio.
Lifelike skin and texture
ByteDance Seed tuned skin, eyes and surface texture to cut the plastic look that gives AI video away. Pores, fine lines and veined hands read like real footage at this resolution. This portrait ran vertically at 9:16 for 4 seconds.
An elderly woman weaves a basket from pale rattan reeds, her weathered hands bending a strand into place; close on her lined face and attentive eyes as she looks down at the work. Steady close-up. Quiet sound of creaking reeds.
The weaver's lined face and veined hands render with natural micro-texture and light, with no plastic sheen. A tight portrait that names the material reads best.
Cinematic action from a reference video
Pass a reference video and Seedance 2.5 4K carries its camera move and pacing into a scene you describe in the prompt. Cite each clip as @Video 1; up to 10 are supported, and each is billed for its full length on top of the output. This 10-second shot followed a 5-second rooftop parkour reference while the warrior, the colossus and the crashing sound all came from the prompt.
Follow the camera movement and pacing of @Video 1. A scarred, bearded warrior in battered bronze armor, a burning chain whip wrapped around one forearm, sprints along the rooftops of a crumbling desert palace at sunset as a towering stone colossus smashes the domes behind him. He runs up a carved sandstone wall, flips over a falling beam, lashes the burning chain around the colossus's wrist and swings across a chasm, landing in a roll as debris and embers rain down. Epic, sweeping action, golden dust, deep crashing sound, roaring wind and a war cry.
The clip keeps the reference's tracking camera and pacing across all ten seconds, with golden dust, deep crashing sound and a war cry in the mix.
Examples to try
The teapot example leads with a reference image, so the model locks the product while the shot plays out.
Reference-guided product consistency
@Image 1 is the porcelain teapot. Keep its blue-and-white bird-and-branch motif, curved spout and round shape identical. A hand lifts the teapot and pours a thin stream of amber tea into a matching white cup on a pale linen cloth; steam rises from the cup. Static camera. Quiet pouring sound.
The generated teapot keeps the referenced blue-and-white motif, curved spout and round shape while a hand pours amber tea into the cup. Add up to 30 reference images and name each as @Image 1; reference images and a first frame cannot be combined.
Recommended settings
Start from the quick-preview settings, then raise duration and bitrate for a finished render.
| Setting | Value | When to use it |
|---|---|---|
| resolution | 4k | Fixed on this slug; use the base seedance-2.5 for 480p-1080p |
| duration | 4-30 | 4 for a fast test; raise it to match your beat, generated in one pass |
| aspect_ratio | 16:9 or 9:16 | Text-to-video; switch to adaptive when you pass a first frame |
| generate_audio | true | Co-generate dialogue, effects and ambience with the picture |
| output_codec | h264 | For a file that plays in every browser; leave default for 10-bit HEVC |
| bitrate_mode | high | Maximum fidelity on a finished render; standard for smaller files |
| seed | fixed value | Iterate the same scene; -1 for variety |
Good to know
- A 4K render takes several minutes and runs as an async job: submit the request, then poll the returned job URL for the finished file.
- The default file is 10-bit HEVC (H.265), which some browsers cannot play. Set output_codec to h264 for an H.264 file that plays everywhere; at 4K that re-encode adds delivery time.
- References are cited in the prompt as @Image 1, @Video 1 or @Audio 1. Each reference video is billed for its full length, counted at 4K, on top of the output.
- Content filters block recognizable real faces and copyrighted characters, so avoid a real human face in a first frame.
- This slug renders at native 4K only. For 480p to 1080p output, use the base seedance-2.5.
- No independent benchmarks cover 2.5 at launch, so the capability figures are ByteDance Seed's own.
Use it from the API
Send the prompt and settings to the async endpoint, then poll the returned job URL until the 4K clip is ready.
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"seedance-2.5-4k",
prompt="A ruby-throated hummingbird hovers at an orange trumpet flower in a shaded garden, iridescent green feathers and fast-beating wings crisp in extreme close-up; it slips its slender beak into the bloom, knocks a puff of pollen into the air, then darts upward out of frame. Steady macro shot. Natural sound of humming wingbeats and rustling leaves.",
first_frame_url="",
last_frame_url="",
reference_images=[],
reference_audios=[],
duration=4,
resolution="4k",
aspect_ratio="16:9",
generate_audio=True,
seed=42,
return_last_frame=False,
skip_moderation=False,
bitrate_mode="standard",
auto_resize=True,
output_codec="h264",
)
print(result.get("output"))# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/seedance-2.5-4k" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "A ruby-throated hummingbird hovers at an orange trumpet flower in a shaded garden, iridescent green feathers and fast-beating wings crisp in extreme close-up; it slips its slender beak into the bloom, knocks a puff of pollen into the air, then darts upward out of frame. Steady macro shot. Natural sound of humming wingbeats and rustling leaves.",
"first_frame_url": "",
"last_frame_url": "",
"reference_images": [],
"reference_audios": [],
"duration": 4,
"resolution": "4k",
"aspect_ratio": "16:9",
"generate_audio": true,
"seed": 42,
"return_last_frame": false,
"skip_moderation": false,
"bitrate_mode": "standard",
"auto_resize": true,
"output_codec": "h264"
}' | jq -r .request_id)
# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID/status" \
-H "x-api-key: YOUR_API_KEY"
# 3. Fetch the result
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID" \
-H "x-api-key: YOUR_API_KEY"Resources
- Try Seedance 2.5 4K on Segmind
- API reference
- Introducing Seedance 2.5 (ByteDance Seed) — the maker's launch post: 30-second single-pass generation, up to 50 multimodal references, native 4K, 10-bit colour, native audio and timestamp-level editing
- Seedance model family (ByteDance Seed) — the Seedance family hub: capabilities, architecture overview and sample films
- Seedance 2.5 prompt guide (ByteDance Dreamina) — the eight-part prompt formula, reference labelling (@Image/@Video/@Audio), audio and subtitle brackets, and worked prompt examples
- #text-to-video
- #bytedance-seed
- #model-launch
- #4k-video
- #native-audio
