Seedance 2.5Text to Video APIImage to Video APIByteDanceAI Video

What Is Seedance 2.5? ByteDance’s 30-Second Video Model, #1 at Image to Video. Our First Clip: 4 Seconds at 480p for $1.01

Seedance 2.5 is ByteDance’s current video model: up to 30 seconds of video and sound in one pass, from a prompt or from a first frame. It is the #1 model on Artificial Analysis’s image-to-video arena, and it is now on MegaBrain’s media API. Here is what it does, what the API takes, what a clip costs, and the first clip we made with it.

2026-11-19·6 min read

The short version: Seedance 2.5 (ByteDance Seed, July 31, 2026) makes clips of up to 30 seconds with synchronized sound, ranks #1 at image to video (1161 Elo) and 4th at text to video (1157). On MegaBrain it runs as bytedance/seedance-2.5/text-to-video and bytedance/seedance-2.5/image-to-video: 4 to 30 seconds, 480p to 1080p, 7 aspect ratios, sound included, from $0.2536 a second. Our 4-second 480p test clip cost $1.01 and took about 3.5 minutes.

What Seedance 2.5 is

ByteDance Seed announced Seedance 2.5 on July 31, 2026 as its new-generation video creation model. It keeps Seedance 2.0’s architecture, which generates audio and video jointly rather than adding a soundtrack afterwards, and moves on three fronts:

  • Length. Up to 30 seconds in a single pass, double Seedance 2.0’s 15, and several rounds of extension for videos of a few minutes. ByteDance says shot transitions and scene changes stay more continuous.
  • References. Up to 30 images, 10 video clips and 10 audio clips in one generation, to pin down characters, motion, a clay render or a style.
  • Editing and control. Timestamp prompts that set the narrative, camera and movement for a given span of the clip; timestamp-level edits to audio and video; green-screen background replacement and camera-perspective edits.

ByteDance also says it improved textures, skin and eyes, lighting and color, keeps the look and voice of several characters consistent in one shot, and produces fewer unwanted subtitles and background music. It names its own weak spots: the physical plausibility of complex motion and the stability of scenes where several subjects interact.

How it ranks

On Artificial Analysis’s blind arenas, where voters pick the better of two clips made from the same input, Seedance 2.5 is #1 of 35 at image to video with 1161 Elo (rank range 1 to 2) and 4th at text to video with 1157. It is one of the few models near the top of both boards: Wan 3.0, #1 at text to video, is 11th at image to video. Our image-to-video arena breakdown has the full table.

What the API takes on MegaBrain

Two model ids, both on the same queue API as every other media model, both on text to video and image to video next to Kling:

FieldValuesDefault
promptWhat happens, the camera and the soundrequired
duration"4" to "30" seconds"5"
resolution480p, 720p, 1080p720p
aspect_ratioadaptive, 16:9, 9:16, 1:1, 4:3, 3:4, 21:916:9 (text), adaptive (image)
generate_audiotrue or false, same pricetrue
first_frame_urlImage to video only: https:// URL or data: URIrequired there
last_frame_urlImage to video only, optional end frame—

Not on the API yet: 4K, automatic duration, and the reference-video and editing modes. Their price depends on inputs that cannot be measured before the job runs, and we only list what we can price up front.

What a clip costs

ResolutionPer second5 s10 s30 s
480p$0.2536$1.27$2.54$7.61
720p$0.5439$2.72$5.44$16.32
1080p$1.3386$6.69$13.39$40.16

Sound costs nothing extra, so leave generate_audio on unless you add your own track. The price is charged when the job is queued and refunded in full if the generation fails. Pro and Max plan discounts apply.

Our first clip

The first Seedance 2.5 job we ran on MegaBrain after launch, from the model page’s playground: 4 seconds, 480p, 16:9, no sound. It came back in about 3.5 minutes as an 854×480, 24 fps MP4 of 1.4 MB, and the balance went down by $1.01.

Seedance 2.5, text to video, 4 s, 480p, no sound. Prompt: “A red fox trots through fresh snow in a pine forest at sunrise, slow tracking shot”

The fox keeps its shape and gait for the whole clip, the low sun lights its fur from behind, and the camera tracks it as asked. At 480p fine fur detail softens; for a hero clip, 720p costs 2.1x as much and 1080p 5.3x.

Make one with the API

The same job from a terminal. Take your key from the API key page of your account, then submit and poll:

# 1. Submit the job (the price is charged here, refunded if it fails)
curl -s https://getmegabrain.com/api/gateway/v1/queue/bytedance/seedance-2.5/text-to-video \
  -H "Authorization: Bearer $MEGABRAIN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A red fox trots through fresh snow in a pine forest at sunrise, slow tracking shot",
    "duration": "4",
    "resolution": "480p",
    "aspect_ratio": "16:9",
    "generate_audio": false
  }'
# -> {"request_id": "18d6c836-...", "status": "IN_QUEUE", "response_url": "...", ...}

# 2. Poll until it answers 200 (202 while running, 422 if it failed)
curl -s https://getmegabrain.com/api/gateway/v1/queue/requests/$REQUEST_ID \
  -H "Authorization: Bearer $MEGABRAIN_API_KEY"
# -> {"video": {"url": "https://getmegabrain.com/api/gateway/v1/media/files/.../0",
#               "content_type": "video/mp4", ...}, "seed": 1745773368}

For image to video, send the frames under their own fields:

{
  "prompt": "The mug slowly rotates, steam rises, soft morning light",
  "first_frame_url": "https://example.com/product.jpg",
  "last_frame_url": "https://example.com/product-end.jpg",
  "duration": "6",
  "resolution": "720p",
  "aspect_ratio": "adaptive"
}

The full queue contract, statuses and refunds are in the media API docs. To wire it into an app, our product photo to video tutorial has a Python script with the same submit-and-poll loop; swap the model id for bytedance/seedance-2.5/image-to-video and the frame field for first_frame_url.

When to pick it

Pick Seedance 2.5 for clips longer than 15 seconds, for scenes where sound matters (it is generated with the picture and costs nothing extra), and for animating a first frame you already approved, where it leads the arena.

Pick Kling for short clips on a budget: Kling 3.0 Standard and Kling 2.6 Pro cost a fraction per second for 3 to 15 seconds. Draft on Kling or Seedance at 480p, and render the keeper at 720p or 1080p.

Try it

Open Seedance 2.5 text to video or Seedance 2.5 image to video, sign in, and run the playground on the page; it uses the same API and shows the cost before you press Generate.

FAQ

What is Seedance 2.5?

Seedance 2.5 is ByteDance Seed’s video generation model, announced July 31, 2026. It generates up to 30 seconds of video with synchronized sound in a single pass (Seedance 2.0 stopped at 15), from a text prompt, a first frame, or up to 30 images, 10 video clips and 10 audio clips as references, and it can extend a clip several times.

Is Seedance 2.5 good?

On Artificial Analysis’s blind image-to-video arena (AA-Video-I2V v2.1) it ranks #1 of 35 at 1161 Elo, and 4th on the text-to-video arena at 1157. ByteDance itself says complex motion and scenes where several subjects interact still have room to improve.

How much does Seedance 2.5 cost on MegaBrain?

Per second of video: $0.2536 at 480p, $0.5439 at 720p and $1.3386 at 1080p, with sound at no extra cost. A 5-second 720p clip costs $2.72, a 4-second 480p clip $1.01, and a 30-second 1080p clip $40.16. The price is charged when the job is queued and refunded if it fails.

How long does a Seedance 2.5 clip take to generate?

Our 4-second 480p test clip took about 3.5 minutes from submit to file. Longer and higher-resolution clips take longer, so call the queue API and poll rather than holding one HTTP request open.

Can Seedance 2.5 animate my own image?

Yes. bytedance/seedance-2.5/image-to-video takes the image as first_frame_url (an https:// URL or a data: URI) and an optional last frame as last_frame_url; with aspect_ratio adaptive the output follows the first frame’s shape.

Sources

Pages opened November 19, 2026. Prices are MegaBrain’s from the model pages on that date; clip costs are our arithmetic on them. The test clip is our own run.

Full disclosure: MegaBrain sells pay-as-you-go API access to image, video and text models, including Seedance 2.5 and Kling. MegaBrain is not affiliated with ByteDance, Kuaishou or Artificial Analysis.

MegaBrain for Mac · an AI coworker

Give MegaBrain the task, not the steps.

It works on your files, writes documents, spreadsheets and decks, connects to your team’s tools and asks before anything risky. Free to start.

Newsletter

Stay in the loop

Get our latest AI research and guides. No spam, unsubscribe anytime.