Seedance image-to-video is one model name and about ten different ceilings. fal documents its Seedance 2.5 image-to-video endpoint at up to 720p; Dreamina advertises 4K on the same model. Clip length, end-frame control, how many reference images you may attach and what one second costs are all decided by the platform you hit, not by the Seedance version you asked for. Pick on the model name alone and you can hit a resolution or length wall that belongs to the host, not to Seedance. We keep the per-tier specifics on the Seedance 2.5 and Seedance 2.0 pages.
The ceiling belongs to the endpoint, not to Seedance. Same model, different limits: fal's Seedance 2.5 image-to-video page lists 480p / 720p / 1080p, Dreamina claims 4K with up to 50 references.
One second of Seedance is not one price. On fal, image-to-video runs roughly $0.2205/s at 480p, $0.4730/s at 720p and $1.164/s at 1080p — a 5.3× spread inside one model, read Sep 18, 2026.
New is not the same as used. On Replicate, Seedance 1.5 Pro has 4,280,132 runs against Seedance 2.5's 152,290 — the older tier is still where the volume is.
If what you want is to animate a still today rather than wire up an SDK, start at Kavel: five Seedance tiers — 2.5, 2.0, 2 Fast, 2 Mini and 1.5 Pro — in one picker, guest credits before you sign up, and the credit cost of your current settings on screen before the run. fal and Replicate want an API key and a card first.
Why "does it support Seedance image-to-video" is the wrong question
Every platform in this list supports Seedance image-to-video. Support is not the constraint. These four are, and each is set by the host:
- Resolution ceiling — the same Seedance 2.5 is capped at 720p on one endpoint and sold as 4K on another
- Single-clip length — Seedance 2.5 generates 30 seconds natively, but not every host exposes the full range
- Reference capacity — first frame only, first plus last frame, or a full reference set
- Billing unit — per second of output, per credit, or per month
What Seedance 2.5 can do when nothing is in the way
Read from Replicate's own model page on Sep 18, 2026: duration is settable from 4 to 30 seconds, or -1 for intelligent duration; a single generation accepts up to 30 images, 10 video clips and 10 audio files as references, addressed in the prompt as [Image1], [Video1], [Audio1]. Image-to-video uses your still as the first frame, and an optional last frame decides where the shot ends. Audio is generated in the same pass as the picture — the same property we measured across the field in AI video generators with sound.
How much of that reaches you depends on the platform.
Seedance image-to-video platforms compared
Every figure read from the platform's own model, pricing or documentation page on the date in the last column. Domains are plain text so you can check them yourself.
| Platform | Seedance tiers with image-to-video | Resolution on that endpoint | Longest single clip | Reference control | What you pay | Account needed | Domain | Checked |
|---|---|---|---|---|---|---|---|---|
| Kavel | 2.5, 2.0, 2 Fast, 2 Mini, 1.5 Pro | 480p to 1080p | Up to 30s on 2.5 | First frame, prompt-driven | Credits, cost shown before the run | Guest credits first | kavel.ai | Sep 18, 2026 |
| fal | 2.5, 2.0 (US-hosted) | 480p / 720p / 1080p | 30s native | First frame + optional end frame | ~$0.2205/s 480p, ~$0.4730/s 720p, ~$1.164/s 1080p | Sign-in + API key | fal.ai | Sep 18, 2026 |
| Replicate | 2.5, 2.0, 2.0 Fast, 2.0 Mini, 1.5 Pro, 1 Pro, 1 Lite | Model-dependent | 4–30s, or auto | 30 images, 10 clips, 10 audio | Per-run, vendor rate | Account + token | replicate.com | Sep 18, 2026 |
| Dreamina | 2.5, 2.0, 2.0 Mini | Advertised up to 4K | 30s | Up to 50 references, R2V control | Platform credits | Account | dreamina.capcut.com | Sep 18, 2026 |
| BytePlus ModelArk | Seedance line, enterprise API | Per contract | Per contract | Per API | Enterprise pricing | Business account | byteplus.com | Sep 18, 2026 |
| Higgsfield | 2.5 image-to-video, 2.0 | Endpoint-dependent | Endpoint-dependent | First frame | Platform plan | Account | higgsfield.ai | Sep 18, 2026 |
| Pollo AI | 2.5 | Not published to anonymous visitors | Not published | First frame | Platform credits | Account | pollo.ai | Sep 18, 2026 |
| Krea | 2.5 | Model-dependent | 30s | First frame | Plan credits | Account | krea.ai | Sep 18, 2026 |
| Kapwing | 2.5 inside an editor | Editor export ceiling | Editor-dependent | First frame | Editor plan | Account | kapwing.com | Sep 18, 2026 |
| emaki | 2.5, 2.0, 2 Fast, 2 Mini | 480p / 720p / 1080p | Up to 30s on 2.5 | First frame | ~50 credits per clip at the cheapest setting; ¥1,580/mo for 350 credits | Free credits on signup | emaki.ai | Sep 18, 2026 |
How we checked: platform model and pricing pages, ByteDance's own release note on seed.bytedance.com, and the third-party write-ups currently cited by the engines for this question — erikataranto.com, ofox.ai, ai-inspo.com, nerdbot.com, appleworld.today, rekreate.ai — all read Sep 18, 2026. Pollo AI's model page returned a Cloudflare challenge to a plain request, so its cells stay unpublished rather than guessed.
What one second of Seedance image-to-video costs
fal is the only host in this list that prints a per-second number on the model page itself, so it is the one set of figures everyone else can be measured against. Note the page disagrees with itself: the description line calls this endpoint "up to 720p" while the pricing note below it quotes a 1080p rate. Treat the rate card as the real answer and the prose as stale.
| Resolution | fal price per second of output | 10-second clip | 30-second clip |
|---|---|---|---|
| 480p | ~$0.2205 | ~$2.21 | ~$6.62 |
| 720p | ~$0.4730 | ~$4.73 | ~$14.19 |
| 1080p | ~$1.164 | ~$11.64 | ~$34.92 |
Why the jump is steeper than the pixel count
Moving 480p to 1080p multiplies pixels by about 4.5 and the bill by 5.3. fal publishes the mechanism: billing is token-based, roughly height × width × (input duration + output duration) × 24 ÷ 1024, at about $0.0214 per 1,000 tokens for 480p and 720p and about $0.0234 for 1080p. The resolution raises both the token count and the token price.
On fal's published rates, then, resolution rather than length is the expensive decision for Seedance image-to-video. A 30-second 480p cut costs less than a 6-second 1080p one. Our own AI video cost per second breakdown found the same shape across 16 models.
Which Seedance version is actually worth picking
The newest tier is not automatically the right one, and Replicate's public run counters make that unusually easy to check. Read Sep 18, 2026:
| Seedance model on Replicate | Runs | What it is for |
|---|---|---|
| Seedance 1.5 Pro | 4,280,132 | The volume workhorse — cheapest premium tier |
| Seedance 1 Lite | 3,755,446 | Fast drafts and thumbnails-in-motion |
| Seedance 1 Pro | 2,426,869 | 5s or 10s at 480p or 1080p |
| Seedance 2.0 | 1,415,183 | Multimodal inputs, native audio |
| Seedance 2.0 Fast | 682,595 | Cheaper 2.0 for iteration |
| Seedance 2.5 | 152,290 | 30s native, 30-image reference sets |
| Seedance 2.0 Mini | 95,231 | High-volume, lowest cost of the 2.x line |
Seedance 2.5 is the flagship and has roughly 4% of Seedance 1.5 Pro's run count on Replicate. A new tier starting behind an established one is normal and not a verdict on quality — but when we went looking for prompt conventions and worked examples, most of what exists is still written against the older tiers.
When 2.5 earns its price on image-to-video
Three cases, all of them about the reference set rather than the picture quality: you need one continuous 30-second take instead of four stitched cuts; you need a last frame to land the shot on a specific composition; or you are combining a character reference, a motion reference and an audio reference in one generation. For how 2.5 compares against Google's flagship on exactly this, see Seedance 2.5 vs Veo 3.1. If none of those apply, Seedance 2.0, Seedance 2 Fast or Seedance 1.5 Pro will usually do the job for less.
If you are not working in English
Seedance is a ByteDance model, and every platform in the table above ships an English or a Chinese interface first. Nothing in that list is built around a third language. Two things break for non-English users before the model ever runs: the interface, and prompt handling.
Emaki is the Japanese-language route into the same models — Seedance 2.5, 2.0, 2 Fast and 2 Mini, with the whole interface and the prompt field in Japanese rather than a machine-translated shell. Read Sep 18, 2026, Seedance 2 Mini runs inside the free credits you get on signup, the cheapest setting is about 50 credits a clip, and the credit cost of the current settings is displayed before you press generate rather than deducted after. Plans start at ¥1,580 a month for 350 credits and run to ¥7,980, with annual billing priced as ten months.
That last detail — cost shown before the run — matters more in a second language than in your first, because a misread settings label is only expensive when you find out afterwards.
FAQ: running Seedance image-to-video
Do I need a Chinese phone number to use Seedance?
Not for anything in the table above: every platform listed signs you up with an ordinary email account, and we saw no phone-number step on any of them on Sep 18, 2026. Whatever a given ByteDance app asks for in its home market is a separate question from what fal, Replicate, Dreamina or Kavel ask for.
Do I need an API key for Seedance image-to-video?
Only on the API hosts. fal and Replicate are developer platforms and want a key and a payment method. Kavel, Dreamina, Krea and emaki are web products where you upload a still and press a button.
Is there a free way to try Seedance image-to-video?
Yes, and the shapes differ. Kavel starts guests on credits before signup, emaki gives free credits on signup that cover Seedance 2 Mini, and several platforms run limited trials. The frontier 2.5 tier is rarely free anywhere for long.
What does the end-frame option actually do?
You supply a second image and the model treats it as the last frame, generating the motion that gets from the first to the second. It is the difference between "animate this photo" and "get from this photo to that one". For product and fashion shots, where the last frame is usually the hero image, we reach for it more than any other control.
Why does the same Seedance clip look different on two platforms?
Usually because the endpoint is different: resolution cap, duration, whether audio is generated, and which Seedance tier the platform is actually calling when its UI says "Seedance". Seeds and prompt handling account for the rest. Check the tier before blaming the model — the Seedance 2.5 model page lists what each one does on Kavel.
