Slop TVNewsLatest
Explainers

Seedance 2.5 guide: specs, prices and access, updated weekly

SLOP TV's running guide to Seedance 2.5: what a generation allows, what it costs on each platform, what shipped and SLOP TV's coverage of the model.

Illustration: Seedance 2.5 guide: specs, prices and access, updated weekly
SlopTV News design studio

Key takeaways

  • Seedance 2.5 generates 4 to 30 seconds of video in one pass at 480p, 720p or 1080p, and a single request takes up to 50 reference assets made of 30 images, 10 video clips and 10 audio clips.
  • ByteDance's Seed team released Seedance 2.5 on 31 July 2026 on Jimeng AI and Doubao Pro, with API access through BytePlus ModelArk under the model ID dreamina-seedance-2-5-260628.
  • fal prices Seedance 2.5 at about $0.2205 a second at 480p and $0.4730 a second at 720p with audio, while Higgsfield's console listed it from $0.144 a second on 26 September 2026 and Dreamina advertises plans from $0.04 a second.
  • In Runway's own test of 250 image-to-video prompts, always using Seedance 2.5 produced 78% usable clips at $1.80 each, against 74% at $0.61 for Runway's quality router capped at $1.00 a clip.

Updated 9 Nov 2026 · next update 16 Nov 2026

Key facts as of 26 Sep 2026

  • Spec: Seedance 2.5 generates 4 to 30 seconds of video in one pass at 480p, 720p or 1080p, with 1080p written at 10-bit color, under the model ID dreamina-seedance-2-5-260628. (source)
  • References: a single request accepts up to 50 assets, made of 30 images, 10 video clips and 10 audio clips, and the model can generate from audio references alone, up to 30 seconds in total. (source)
  • Release: ByteDance's Seed team launched Seedance 2.5 on 31 July 2026 on Jimeng AI and Doubao Pro, saying API access was coming soon through BytePlus ModelArk; the ModelArk tutorial for the model is where it now lives, under the ID dreamina-seedance-2-5-260628. (source)
  • Editing: the model takes timestamp-level instructions, supports green-screen, camera-perspective and reference-based editing, and can follow a textureless clay render for camera movement, blocking and pacing. (source)
  • API price at fal: about $0.2205 a second at 480p and $0.4730 a second at 720p with audio, billed as $0.0214 per 1,000 tokens on the formula (height x width x duration x 24) / 1024. (source)
  • Price on Higgsfield: the Seedance 2.5 playground prices text-to-video and image-to-video from $0.144 a second, and its model data lists reference-to-video, video edit and video extend across $0.1728 to $0.9553 a second for 480p to 1080p, with the discount marked as expiring on 1 October 2026. (source)
  • Consumer price: Dreamina's own plans run from $15 a month with 1,575 credits to $520 a month with 59,000 credits, and the page advertises Seedance 2.5 from $0.04 a second, with credit and plan discounts running from 23 September to 9 October 2026. (source)
  • Draft Mode: a 480p drafting model added to Higgsfield on 22 September 2026 lets you preview takes and then run the chosen prompt, references, seed, aspect ratio and duration at 1080p; Higgsfield offers it in all regions except the US, where it says the model is unavailable on the provider's side. (source)
  • Runway's comparison: in Runway's own test of 250 image-to-video prompts, always using Seedance 2.5 produced 78% usable clips at $1.80 each, against 74% at $0.61 for its quality router capped at $1.00 a clip. (source)
  • Where it runs: ByteDance's Dreamina and BytePlus ModelArk, fal and Higgsfield, and Perplexity's Computer workspace, which added it with MiniMax H3 on 21 September 2026.

Seedance 2.5 now reaches buyers through three routes at three prices - fal, BytePlus and Dreamina, the platforms bundling it into subscriptions such as Higgsfield's unlimited bonus and Cinema Studio 4.0, and Volcano Engine's own API - while ByteDance consolidates the consumer-side apps that use it. (source)

What Seedance 2.5 is

Seedance 2.5 is ByteDance's latest video generation model, released by its Seed research team on 31 July 2026 as the successor to Seedance 2.0. ByteDance's launch post frames the change as a move from generating a clip to completing a creative work: the ceiling for a single pass doubled from 15 seconds to 30, the number of reference assets a request accepts grew, and audio is generated in the same pass as the picture rather than added afterward. The ModelArk capability table puts the Seedance 2.0 series limit at 15 assets, nine images, three videos and three audios, against 50 for 2.5.

The model runs on what ByteDance calls a unified multimodal audio-video joint-generation architecture, accepting text, images, video and audio in one context. Runway used Seedance 2.5 as the quality baseline in its own router test on 24 September 2026 and describes it as a state-of-the-art video model, which is a competitor's measurement of its own router rather than an independent test of Seedance. Its API home is BytePlus ModelArk, where the model cannot be activated without first buying a resource pack, and the ModelArk tutorial notes that some parameter constraints are checked only once the model has worked out the task type, so an accepted request can still fail asynchronously.

What Seedance 2.5 costs

Prices depend on who is selling it, and the spread is wide.

  • ByteDance's own app: Dreamina's price page lists Seedance 2.5 from $0.04 a second on subscription plans: Basic at $15 a month with 1,575 credits, Standard at $36 with 3,885 credits, Advanced at $159 with 17,600 credits and Ultra at $520 with 59,000 credits. The page counts those allowances as 19, 48, 220 and 737 ten-second videos a month, an estimate it bases on Seedance 2.0 Fast at 720p rather than on Seedance 2.5. First-month prices are advertised at $1.50, $22, $95 and $312, and the page dates its offer period to 9 October 2026.
  • BytePlus ModelArk: sold to developers through Seedance 2.5 resource packs; the ModelArk tutorial says the model cannot be activated until a pack has been bought.
  • fal: published rates are about $0.2205 a second at 480p and $0.4730 a second at 720p with audio, and roughly $1.164 a second at 1080p. Billing is per token at $0.0214 per 1,000 tokens for 480p and 720p, on the formula (height x width x duration x 24) / 1024, so a 30-second 720p clip works out at about $13.87 and 5 seconds at 720p at about $2.31.
  • Higgsfield: the Seedance 2.5 playground prices text-to-video and image-to-video from $0.144 a second, and its model data lists reference-to-video, video edit and video extend across $0.1728 to $0.9553 a second for 480p to 1080p, with the discount marked as expiring on 1 October 2026.

Two practical notes. The token formula is authoritative at fal, so larger frames cost more per second and smaller ones cost less. And Higgsfield's cashback offer returns API spend as credits, which changes the effective rate without changing the list price; Higgsfield's own post says 100% of API spend comes back as credits up to $100,000 per business and that unused cashback expires on 30 September, and does not say whether that date closes earning or spending. (Higgsfield)

What Seedance 2.5 does

Length and shotmaking. One pass produces 4 to 30 seconds, and the model can extend existing output repeatedly while holding characters, environments and pacing. ByteDance's example is a one-take concert sequence that runs from a dressing room to a stage reveal, and the company says the model can organize setup, development, turning point and resolution inside 30 seconds rather than extending a single moment. Longer work is built by chaining extensions, which ByteDance says is how users reach multi-minute output.

References. Up to 30 images, 10 video clips and 10 audio clips go in together, and ByteDance lists specific reference types: character, scene, style, prop, motion, audio, and a clay render that fixes camera movement, blocking and shot scale for stories that need exact staging. Lighting follows the spatial information in that clay render, per the launch post.

Editing. Timestamps control what happens when, both during generation and after it. The model supports green-screen replacement, camera-perspective changes and reference-based edits, and it keeps the subject's response to its new surroundings, including how clothing moves and how light falls on it.

Sound. Audio is generated with the video in one pass and the API accepts audio references. fal's reference-to-video page describes sound and picture as generated jointly in the same latent space rather than one after the other, and repeats ByteDance's claim of roughly 20% better prompt adherence.

What ByteDance says it cannot do yet. The launch post names the physical plausibility of complex motion and the stability of scenes with several interacting subjects as the two areas still needing work, and says plainly that "there is still room for improvement" rather than claiming the problem is solved.

Who Seedance 2.5 is for

Three different buyers, on three different routes. Creators who want a finished clip without an API buy it inside Dreamina or one of the resellers that wrap the model in an interface, where the monthly credit allowance is the real constraint. Studios and agencies buy it through ModelArk or a platform such as fal or Higgsfield when they need the model as a callable endpoint and are weighing cost per second against the quality floor. And builders who need video inside something else get it bundled: Runway's Model Router sends work to it when quality matters more than price, and Perplexity's Computer workspace added it to Pro and Max subscriber plans on 21 September 2026, Perplexity said.

Because the same weights are available at very different rates, the useful question for a buyer is not which platform hosts Seedance 2.5 but which route to those weights is cheapest for the resolution, duration and volume you actually use. A 30-second 720p clip is about $13.87 at fal's token rate, while Dreamina counts a ten-second clip against a $15 monthly allowance where the per-second figure starts at $0.04, and those are not the same purchase.

The Seedance 2.5 diary

Dated entries, newest first. Each entry is added once and never rewritten; corrections are made in a new entry.

9 Nov 2026

Voyager, the desktop AI agent from Nullframe, published its own price comparison for ByteDance's Seedance 2.5 on 9 November: on its Creator plan a 720p second is listed at $0.2362 against Higgsfield's API list price of $0.4622, Runway's $0.30 and fal's $0.473, and a 480p second at $0.1064 against $0.2057 at Higgsfield. The page is a reseller's comparison of its own plan rate against rivals' list prices, so the routes are not like-for-like, and Voyager's own figure is not stable: our report of the same table on 6 November read 720p at $0.2304 a second. Source: voyager.so.

ByteDance's own cloud now sells access to Seedance 2.5 through three prepaid creation-credit packages on Volcano Engine: 200 yuan covering about 19 fifteen-second 480P clips on a 30-day life, and 500 and 1,000 yuan covering about 47 and 95 clips over 90 days, with the credits spendable against any Ark model rather than Seedance 2.5 alone, unused packages refundable within seven days and no refunds once used. The page's FAQ caps output at 720P and 30 seconds, which disagrees with ByteDance's own ModelArk documentation for the same model, where 1080p at 10-bit color is listed, so the maker's two pages do not describe the same maximum. Source: volcengine.com.

The second AI Film Festival Japan closed in Tokyo on 2 November after 730-plus submissions from more than 60 countries and regions, which the organizer's own site breaks down as 68 official selections and 21 finalists across a three-day program with CapCut, ElevenLabs and Alibaba Cloud as partners. Nippon Television's report on the festival adds the number that matters for this hub: Seedance was the tool most used across the submitted works, with Suno's music generator second. The organizer's site carries the counts and the partner list but not the tool-usage finding. Sources: aifilm.jp, news.ntv.co.jp.

2 Nov 2026

ByteDance's CapCut unit folded Xiaoyunque, its AI short-drama platform, into Jimeng while keeping Xiaoyunque's standalone entry point, with its operations moved under Jimeng's head of operations and marketing and the product team unchanged, as reported by 36Kr on 24 October and detailed by TechNode on 27 October. TechNode's reading is a concentration of resources while user growth has yet to return the investment, and it notes that revenue from Volcano Engine's Seedance model-as-a-service business is substantially higher than revenue from Jimeng and Xiaoyunque - which is why apps that compete with ByteDance's own creation tools can still be its model customers. Source: technode.com.

Higgsfield announced Cinema Studio 4.0 on 28 October as the AI filmmaking workspace that generates up to 30 seconds in a single pass from up to 50 references, up from 15 seconds and nine references in 3.5; the maker's help center says the workspace runs on ByteDance's Seedance 2.5, the model Higgsfield hosts, and its own page caps the output at 1080p. Higgsfield billed Cinema Studio generations in credits separately from the unlimited model access its plans include. Sources: higgsfield.ai, help center.

Higgsfield's 7-Day Unlimited Bonus was reworked on 30 October to drop Wan 3.0 and Wan 3.0 Prime and add Seedance 2.5 alongside Seedream 5.0 Pro, GPT Image 2.5 and Nano Banana Pro: monthly Max and Ultra plans get Seedance 2.5 at up to 20 seconds and 480p or 720p, and annual billing adds it to Pro and Plus at 720p and raises it to 720p on Max and 1080p on Ultra, per the maker's own page. The offer covers generations made on higgsfield.ai rather than through MCP, the CLI, Canvas, Supercomputer or Marketing Studio. Source: higgsfield.ai.

26 Oct 2026

On 23 October ByteDance's Volcano Engine released Seedance 2.5 4K, a version trained further on facial expression, small movements and fine detail, with 10-bit color and a Draft mode that renders a 480p storyboard first and only then runs the 4K pass; Volcano Engine gives it to a group of film and post-production companies as an enterprise trial and says wider access may follow. The 4K tier is not on ByteDance's public Seedance 2.5 product page, which still describes a 30-second, reference-driven model without naming 4K, so it is an enterprise release on top of the API model rather than a change to the standard offering. Sources: ithome.com, seed.bytedance.com, sloptvnews.com.

On 26 October the University of Hong Kong Business School's AI Evaluation Lab published a study of more than ten video models in which Seedance 2.0 Pro ranked first overall and topped all six generation dimensions - visual continuity, physical plausibility, aesthetic quality, instruction alignment, audio quality and audio-visual matching - but only sixth on safety and responsibility, while Hailuo-2.3 took first place on safety. The lab said the gap shows the industry prioritising quality over safeguards, and it also found that picture quality fell for most models once reference images were added to a prompt. Sources: lead.hku.hk, sloptvnews.com.

19 Oct 2026

On 13 October Cascade, the AI-native production platform owned by Mollyroe plc, launched the Greenlight Film Competition with BytePlus and Seedance under the theme "Great stories deserve a shot". Filmmakers pitch a teaser for a story made on Cascade's platform; four winners share a US$100,000 cash pool funded jointly by Cascade, BytePlus and Seedance, and take a production greenlight and a package of platform credits, with BytePlus providing credit support. Entries close on 4 December 2026 and the winners are expected in the new year. Source: investegate.co.uk.

12 Oct 2026

BytePlus ModelArk's pricing page, last updated 8 October 2026, now lists Dreamina Seedance 2.5 at $0.103 a second at 480p, $0.231 at 720p and $0.569 at 1080p, the official list price for the route through ByteDance's own API. Source: docs.byteplus.com.

Dreamina's campaign running 23 September to 9 October 2026 gave 90% off the first month of a Basic membership, which it priced at an effective $0.035 a second for Seedance 2.5 at 720p; the offer window has now closed, and its standing annual first-year rate for the model is $0.047 a second. Source: dreamina.capcut.com.

26 Sep 2026

The hub opens. This first entry records the news the page starts from.

On 26 September the Higgsfield Seedance 2.5 playground priced text-to-video and image-to-video from $0.144 a second, and its model data listed reference-to-video, video edit and video extend across $0.1728 to $0.9553 a second for 480p to 1080p, with the discount marked as expiring on 1 October 2026.

On 24 September Runway published a benchmark of its Model Router in which Seedance 2.5 was the quality baseline: 78% of clips usable at $1.80 each, against 74% at $0.61 for a quality router capped at $1.00 a clip, in Runway's own test of 250 image-to-video prompts. Source: Runway.

On 22 September Higgsfield added Seedance 2.5 Draft Mode, which generates 480p previews so a take can be chosen before the 1080p final, and said the model is unavailable in the United States on the provider's side. Source: Higgsfield changelog.

On 21 September Perplexity said its Computer workspace could create video with MiniMax H3 and ByteDance's Seedance 2.5, available to Pro and Max subscribers. Source: Perplexity.

On 31 July ByteDance's Seed team launched Seedance 2.5 with 30-second single-pass generation, 50 reference assets and timestamp-level editing, rolling out on Jimeng AI and Doubao Pro and saying API access was coming soon through BytePlus ModelArk, where it now runs under the ID dreamina-seedance-2-5-260628. Sources: Seed launch post, ModelArk tutorial.

What changed this week

Against the 2 Nov 2026 edition: Voyager put its own Seedance 2.5 price comparison up against Higgsfield, Runway and fal, and ByteDance's Volcano Engine now sells the model through three prepaid creation-credit packages with 30- and 90-day lives, whose FAQ's 720P ceiling contradicts ByteDance's own ModelArk documentation. The ModelArk tutorial for Seedance 2.5, re-opened on 9 November, still gives 4 to 30 seconds, 480p, 720p and 1080p at 10-bit color and up to 50 reference assets per request, and still requires a purchased resource pack before the model can be activated. The fal, Dreamina and Higgsfield rates in the key-facts block were not re-quoted this week.

SLOP TV's Seedance 2.5 coverage

Sources

  1. docs.byteplus.com
  2. seed.bytedance.com
  3. fal.ai
  4. console.higgsfield.ai
  5. dreamina.capcut.com
  6. higgsfield.ai
  7. runway.com
  8. x.com
  9. x.com
  10. fal.ai
  11. docs.byteplus.com
  12. dreamina.capcut.com
  13. investegate.co.uk
  14. ithome.com
  15. seed.bytedance.com
  16. lead.hku.hk
  17. technode.com
  18. higgsfield.ai
  19. higgsfield.ai
  20. voyager.so
  21. volcengine.com
  22. aifilm.jp
  23. news.ntv.co.jp
  24. higgsfield.ai