Seedance is ByteDance’s video model family, made by its Seed team. The current version is Seedance 2.5 (31 July 2026): up to 30 seconds in one generation, sound generated with the picture, up to 50 reference files, 480p to 1080p. Seedance 1.0, which this page covered in 2025 and which people still search for, made silent 5 or 10-second clips; it is still sold on ByteDance’s API but is two generations old.
You can use Seedance in CapCut and Dreamina, in Jimeng and Doubao in China, or through the BytePlus ModelArk API. Below: what each version does, how to write a prompt it follows, and what it costs.
Seedance versions
| Version | Released | Length | Resolution | Sound | Notes |
|---|---|---|---|---|---|
| 1.0 Lite | Apr 2025 | 5–10 s | up to 1080p | No | 1–4 reference images, first and last frame |
| 1.0 Pro | Jun 2025 | 5–10 s | up to 1080p, 24 fps | No | Multi-shot in one clip |
| 1.0 Pro Fast | Oct 2025 | 5–10 s | up to 1080p | No | Up to 3× faster, cheaper |
| 1.5 Pro | 18 Dec 2025 | Yes | First with audio and lip sync; retired on ModelArk 11 Nov 2026 | ||
| 2.0 (+ Fast, Mini) | 12 Feb 2026 | up to 15 s | 480p–1080p, 4K on 2.0 | Stereo | Images, video and audio as input; editing, extension |
| 2.5 | 31 Jul 2026 | up to 30 s | 480p, 720p, 1080p | Yes | 50 references, edits by timestamp, 480p draft mode |
Seedance 1.0 was a real jump in June 2025: ByteDance reported it at #1 on both Artificial Analysis video boards on 10 June 2025 and a 5-second 1080p clip in 41.4 seconds on an NVIDIA L20. The model in that headline is not the one at the top today.
What’s new in Seedance 2.5
- 30 seconds per generation, twice 2.0’s 15. Longer pieces by extending.
- Sound with the picture: speech, effects and music. Native prompts and speech in Chinese, English, Spanish, Indonesian, Malay, Thai, Arabic, Portuguese, Vietnamese, Japanese and Korean.
- 50 reference assets per request (30 images, 10 videos, 10 audio clips): characters, places, a camera move to copy, a voice.
- Edits by timestamp: “change the action from 4 to 6 seconds, leave the rest”.
- Draft mode: preview at 480p, then render the one you like at 1080p.
- Storyboards: up to 15 panels, or a sequence of keyframe images in order.
ByteDance itself lists the weak spots: physics in complex motion, and stability when several characters interact.
Where to use Seedance
- CapCut and Dreamina (dreamina.capcut.com): Seedance 2.0 arrived in CapCut in stages from March 2026 and in the US on 7 April 2026. Dreamina gives daily free credits; the amount is not published officially and differs between reports, so check in the app.
- China: Jimeng (即梦) and Doubao got 2.5 first.
- API: BytePlus ModelArk, priced by tokens. Resellers such as fal and Replicate sell it at their own prices.
Every Dreamina and CapCut video carries a visible watermark plus C2PA and invisible marks. Uploading real faces was restricted at the CapCut launch.
Seedance pricing (BytePlus ModelArk)
ByteDance’s own worked examples, 5 seconds, 16:9, no video input:
| Model | 480p | 720p | 1080p | 4K |
|---|---|---|---|---|
| Seedance 2.5 | $0.51 | $1.16 | $2.84 | n/a |
| Seedance 2.0 | $0.35 | $0.76 | $1.87 | $3.89 |
| Seedance 2.0 Fast | $0.28 | $0.60 | n/a | n/a |
| Seedance 2.0 Mini | $0.18 | $0.38 | n/a | n/a |
| Seedance 1.0 Pro | $0.12 | $0.26 | $0.61 | n/a |
| Seedance 1.0 Pro Fast | $0.05 | $0.10 | n/a | n/a |
Per second, 2.5 is about $0.23 at 720p and $0.57 at 1080p. Failed or blocked generations are not charged. Price is computed from tokens (seconds × width × height × fps ÷ 1024), so other lengths and shapes scale from these numbers.
The 2025 version of this page said 1080p cost “$0.012 per 5-second clip”. The real 1.0 Pro price was $0.61, about fifty times more.
How to prompt Seedance
From BytePlus’s official guides for 2.0 and 2.5:
- Open with one sentence: subject, place, event, style, camera.
- Then a shot list, not prose: “Shot 1… Shot 2… Shot 3…”, each with its camera move, action, expression and sound. On 2.5 you can use whole-second timestamps instead (
0-3 seconds… 3-7 seconds…, no gaps). 2.0 ignores timestamps. - Tie every character to a file: “the woman in Image 1”, “the voice from Audio 1”. Name files by upload order and say each one’s role.
- One camera move per shot. Medium shot, close-up, slow push-in, lateral tracking, fixed shot; on 2.5 also dolly zoom, FPV, bullet time, speed ramp.
- Slow, continuous actions work better than sprints and big jumps. Name the body part and the speed.
- Constraints that work: “No subtitles”, “Do not generate a logo”, “No BGM; generate only environmental sounds”. Landscape video gets fewer stray subtitles than portrait.
- Use 4–5 references, not 50. One or two character images, one place, one camera-move video, one audio. Filling every slot makes styles fight. Use a clean headshot for faces, not a multi-view character sheet, which produces duplicates.
- Don’t mix languages inside one line of dialogue.
On 1.0, the API also read flags at the end of the prompt: --resolution 1080p --duration 5 --camerafixed false.
Seedance prompt examples
The first five are from ByteDance’s documentation; the last three we wrote with the same rules.
Seedance 2.0 shot list:
Shot 1: Side shot of a street alley; the man slowly starts running, with a sense of rapid breathing. Shot 2: The man knocks over a fruit stand; the camera shakes quickly and gives a close-up of the man's frightened face. Shot 3: The man climbs over a low wall and disappears; the camera slowly pulls back and freezes on the empty street.
Similar recipe in the library →
Seedance 2.5 edit of an uploaded video:
Change the man's action from drinking coffee to mopping the floor from 4-6 seconds in Video 1, and leave the rest of the content unchanged.
Seedance 1.0 Pro, style:
A 1980s TV drama with an old-fashioned and cheap makeup and costume style: a man is writing under a table lamp.
Seedance 1.0 Pro, camera move inside the action:
On the leaves of the garden, a group of pixies reside. As the camera pans right, the protagonist steps out from home, clad in a petal cloak and holding a grass-leaf wand. At the tip of the wand is an embedded glowing yellow gem, styled in a microcosmic world.
Similar recipe in the library →
Seedance 1.0 Pro with API flags:
At breakneck speed, drones thread through intricate obstacles or stunning natural wonders, delivering an immersive, heart-pounding flying experience. --resolution 1080p --duration 5 --camerafixed false
Seedance 2.5, 15 seconds with timestamps and sound:
A street-food vendor in Bangkok at night flips noodles in a wok. 0-4 seconds: wide shot, slow push-in through steam and neon signs. 4-9 seconds: close-up of the wok, flames flare, sizzling and clanging. 9-15 seconds: medium shot, she hands the plate to a customer and says in Thai "Enjoy!", crowd chatter. Realistic, warm tungsten light. No subtitles. No BGM.
Similar recipe in the library →
Seedance 2.0 with references:
The woman in Image 1 walks through the market in Image 2. Shot 1: medium tracking shot, she slowly picks up an orange and smells it. Shot 2: close-up, she smiles slightly and looks toward the camera. Soft morning light, natural colors, cinematic texture. Keep it subtitle-free. Do not generate a logo.
Similar recipe in the library →
Seedance 2.0 Mini, a cheap draft at 720p:
Product shot: a white ceramic mug on a wooden table, coffee being poured in, steam rising. Fixed camera, slow motion, soft window light, rich details.
Similar recipe in the library →
Seedance vs other models
On Artificial Analysis (with audio, read 2 November 2026), Seedance 2.5 is #1 for image-to-video and #4 for text-to-video; Seedance 2.0 is #3 and #9. MiniMax H3 is #3 text-to-video, Kling 3.0 Pro #17–19, Google Veo 3.1 #21–22.
On price, MiniMax H3 at 768p costs $0.08 per second against about $0.23 for Seedance 2.5 at 720p; Seedance 2.0 Mini matches H3 at $0.08. More in our guides to Hailuo and H3, Veo 3, Kling and Midjourney video.
FAQ
What is the latest version of Seedance?
Seedance 2.5, released on 31 July 2026. It makes up to 30 seconds per generation with sound, takes up to 50 reference files and outputs 480p, 720p or 1080p.
What was Seedance 1.0?
ByteDance’s June 2025 video model: text or image to silent video, 5 or 10 seconds, up to 1080p at 24 fps, with several shots inside one clip. Seedance 1.0 Pro and Pro Fast are still on the BytePlus API.
Is Seedance free?
Dreamina and CapCut give free daily credits that can be spent on Seedance; the amount is not published officially. The API is paid, from about $0.05 for a 5-second 480p clip on 1.0 Pro Fast.
Does Seedance generate audio?
Since version 1.5 Pro (December 2025), yes: speech with lip sync, effects and music. Seedance 1.0 was silent.
How much does Seedance cost per video?
On BytePlus, a 5-second 16:9 clip costs $1.16 at 720p and $2.84 at 1080p on Seedance 2.5, $0.76 and $1.87 on 2.0, $0.26 and $0.61 on 1.0 Pro.
Can I use Seedance in CapCut?
Yes. Dreamina Seedance 2.0 rolled out in CapCut from March 2026 and reached the US on 7 April 2026.
Sources
Checked on 2 November 2026:
- ByteDance Seed: Seedance 1.0 tech report (arXiv 2506.09113)
- BytePlus: Seedance 1.5 Pro
- ByteDance Seed: Seedance 2.0 launch
- ByteDance Seed: Introducing Seedance 2.5
- CapCut newsroom: Dreamina Seedance 2.0
- BytePlus ModelArk: pricing
- BytePlus prompt guides: Seedance 1.0 Pro, 2.0, 2.5
- Artificial Analysis: text-to-video and image-to-video