Skip to main content

News

5 min read

Seedance 2.0 Mini is here. Here's what it actually costs to use it for serious work.

Mini is 2x faster and about 50% cheaper than 2.0 Fast: roughly $0.07 a second at 720p, near $35 a month for 500 seconds. It caps at 720p and faces drift on close-ups, so here is when to use it, when to stay on 2.0 standard, and the seed-reuse trick that cuts iteration cost 6x.

Short version

What to remember

  1. 01

    ByteDance shipped Seedance 2.0 Mini 2026-06-15 to 2026-06-17. Seedance 2.5 is not out yet, with an early-July 2026 launch window per Caixin.

  2. 02

    Mini is about 2x faster than Seedance 2.0 Fast and about 50% cheaper per second: ~$0.07/s at 720p on Volcengine Ark. Atlas Cloud lists Seedance 2.0 standard at $0.112/s for 1080p with audio, the only primary-verified international rate.

  3. 03

    Mini caps at 720p. It is not a 4K model. For 1080p or 4K, use Seedance 2.0 standard.

  4. 04

    Face coherence on close-ups is weaker than 2.0. Don't use Mini for talking heads.

  5. 05

    The 'free draft' pitch is a third-party wrapper claim (Magnific Premium+), not a ByteDance feature. The durable cost-saver is the seed-reuse preview trick: 2-3 second preview at $0.14-$0.21, re-seed for the 10-second final at $0.70.

The cheapest verified primary-source path for 500 sec/mo of 1080p+audio on Seedance 2.0 standard is Atlas Cloud at $56/mo. Open in calculator: 500 seconds of Seedance 2.0 at 1080p via API

Prices verified 2026-06-24 against each platform's own pricing page. Where a platform does not publish a public per-second rate, the row says so instead of guessing.

What ByteDance actually shipped

The model that launched is Seedance 2.0 Mini (lowercase "mini," no "Lite" or "Pro" suffix, internal API id Doubao-Seedance-2.0-mini). It is not a separate flagship and it is not 2.5. It followed the same China-first pattern as 2.0: the Volcengine Ark experience center on June 15-16, CapCut and Dreamina on June 17, BytePlus ModelArk around June 22.

That puts the international API lag at about 6-7 days from the China experience center, or 5 days from CapCut and Dreamina. For comparison, Seedance 2.0 standard's China-to-fal.ai gap was about 8 weeks in February-April 2026. Mini shipped faster.

Deltas versus Seedance 2.0 (what changed, concretely)

Seedance 2.0 standard

Seedance 2.0 Mini

Max resolution

1080p (4K on some API tiers)

720p only

Clip length

5-12s

4-15s

Inference speed

Baseline

~2x faster than Seedance 2.0 Fast

Credit cost (~720p)

~¥1/s (~$0.14/s)

~¥0.5/s (~$0.07/s)

Face coherence on close-ups

Strong

Weaker. Measurable identity drift

Prompt adherence

Literal and reliable on complex prompts

More literal-minded and restrictive. Complex prompts may need more iterations

Multi-shot narrative

Standard capability

Same 4-15s envelope, no announced change

Audio

Native audio-video joint generation

No announced change to audio fidelity

Reference inputs

Up to 12 (6 images, 3 audio, 3 video)

Same as 2.0

marks the stronger side on that dimension.

Every dimension where Mini differs from Seedance 2.0 standard. Click it to open fullscreen.

The net trade: you give up resolution ceiling and face fidelity, you gain a real 2x speed and about 50% cost reduction. If your deliverable is 1080p or 4K, Mini is the wrong tool.

The cost math for serious work

Serious work here is 500 seconds of 720p-or-1080p video with audio per month, no annual commitment. Every per-second figure below comes from the platform's own pricing page.

What 500 seconds a month actually costs

USD per month for 500 seconds of output with audio, no annual commitment. Verified 2026-06-24.

Cheapest first. The Volcengine Ark rates are China-only. Atlas Cloud is the only primary-verified 1080p+audio rate outside China.

Not on the chart: fal.ai bills Seedance 2.0 at 4K for $1.5552/s, which is $777.60/mo for the same 500 seconds. 4K is a finishing step on a clip, not a working resolution, and it gets its own rows in the decision matrix below.

The exact per-second rates and sources
PlatformModelResolutionUSD/sec500 sec/moSource
Volcengine Ark (China)Mini720p~$0.07/s~$35/moByteDance Seed
Volcengine Ark (China)2.0 standard720p~$0.14/s~$70/moByteDance Seed
Atlas Cloud2.0 standard1080p + audio$0.112/s$56/moAtlas Cloud
WaveSpeedAI2.0 Fast1080p + audio$0.10/s$50/moWaveSpeedAI
fal.ai2.0 standard4K$1.5552/s$777.60/mofal.ai
seedance.tv Plus2.0 (resold)1080p, 10s cap~$0.0233/s$11.67/moseedance.tv

The preview trick that replaces "free drafts"

The durable trick is simpler, and it works on every platform that exposes a seed parameter:

  1. Write the prompt from the template. Use the format from the Seedance prompt template linked at the bottom of this article.
  2. Render 2 seconds at 720p as a preview. On Mini at ~$0.07/s that is $0.14 a try. On 2.0 standard at ~$0.14/s it is $0.28.
  3. Watch it, edit, render again. If the character, motion, camera, or composition is wrong, edit the prompt and re-render. Ten tries is $1.40 to $2.80.
  4. Reuse the same seed, render the 10-second final. Same seed, same prompt, longer duration. The model keeps the character, composition, opening motion, and camera move from the seed.

Ten previews plus one seeded 10-second final on Mini costs $2.10. Ten full 10-second attempts on 2.0 standard, at $1.40 each, costs $14. Same result, 6-7x cheaper.

Why it works: Seedance seeds encode the latent motion and composition, not just the frame. Same seed, same prompt, longer duration gives you the same shot extended. That is documented model behavior, not a hack.

Decision matrix

WorkloadCheapest verified pathCost / monthNotes
100 sec/mo, 720p + audioVolcengine Ark Seedance 2.0 Mini~$7China-only path; international Mini is not yet on a verifiable per-second rate
500 sec/mo, 720p + audioVolcengine Ark Seedance 2.0 Mini~$35Cheapest verified, China-only
500 sec/mo, 1080p + audioAtlas Cloud Seedance 2.0 standard$56Only primary-verified per-second rate outside China as of 2026-06-24
2000 sec/mo, 1080p + audioAtlas Cloud Seedance 2.0 standard$224Linear; no volume discount retrieved
50 sec/mo, 4K (finishing only)fal.ai Seedance 2.0~$78Token-billed. 4K is a finishing step, not a workflow
100 sec/mo, talking head, 1080pDo not use Mini. Seedance 2.0 standard on Atlas Cloud~$11Mini's face coherence is too weak for close-ups
Any volume, 4K, 30+ daysDo not start here. 4K is finishing, not iteration.n/aRender at 1080p, upscale to 4K via Replicate's ByteDance Video Upscaler as a separate paid step

The "Any volume, 4K" row is the one most teams get wrong. The right way to ship 4K at scale is two passes at different resolutions, not one pass at 4K. That is the difference between $78 and $777 for the same 50 seconds of finished output.

When NOT to switch from Seedance 2.0 standard to Mini

  1. Your deliverable is 1080p or 4K. Mini caps at 720p, and there is no upscale that turns 720p into commercial 1080p without visible artifacts.
  2. You shoot talking heads or close-up character work. Identity drift on Mini is measurable. A face that looks right in a 2-second preview can look like a different person by frame 180.
  3. You need a complex multi-element prompt (more than three named objects or two characters in motion). Mini is more literal-minded and restrictive. Expect more iterations fighting the prompt.
  4. You are rendering for a platform that requires native 4K (broadcast, for example). Mini is not a 4K model, and ByteDance has not announced 4K for Mini.

Get the Seedance prompt template

The full Seedance prompt template is one of the free brains in Scopeful. The same template drives Flora, Magnific, BytePlus ModelArk, Volcengine Ark, Dreamina, and fal.ai's Seedance endpoints without changes, and the seed-reuse preview trick is built into it, with an explicit SEED: field and a hard-constraint entry that calls out the iteration pattern.

Create a free Scopeful account and the template is in your library. Free plan, no card. The in-app Copilot fills it in for the model and platform you pick.

Questions

Questions & Answers

Is Seedance 2.5 out yet?
No. It is in late-stage beta per Caixin, with a public launch window in early July 2026. ByteDance has confirmed 4K is a native 2.5 feature but has not published a rate.
Can I keep using Seedance 2.0 standard?
Yes. ByteDance has not deprecated 2.0 standard or 2.0 Fast. All three coexist on Volcengine Ark, and 2.0 standard is the only one with a verified per-second international rate as of 2026-06-24.
Will my existing generations break?
No. Generations are not deprecated when a new model ships.
Is there a free trial of Mini?
No documented free trial for international users. The seed-reuse preview trick in the article is the practical free-trial equivalent for any user.
Does the preview trick work on Kling 3.0 and Veo 3?
Different seed semantics. Kling's image-to-video reference is the more reliable preview path. Veo 3 has no documented public seed-reuse preview path.
How does Mini compare to Veo 3, Sora 2, and Hailuo?
The closest verified comparison is Veo 3 on the seedance.tv Plus plan at ~$0.0187/s (third-party storefront). Sora 2 at $0.70/sec on OpenAI's API is the most expensive verified 1080p line item. For a like-for-like 1080p+audio comparison on a primary source, the choice is between Volcengine Ark Mini (720p only, ~$0.07/s) and Atlas Cloud 2.0 standard (1080p with audio, $0.112/s).
What if I need 4K, 30s clips, or longer than 10s?
Mini tops out at 720p and 15s. 2.0 standard tops out at 1080p and about 12s on the consumer tier. For 4K or longer clips, use 2.0 standard on BytePlus ModelArk (no public per-second rate for that SKU) or run a separate 4K pipeline via Replicate's ByteDance Video Upscaler on a separately-rendered 1080p clip.