Seedance 2.5 Flash
Make 4–30 second videos with sound from a prompt, a photo, reference images, clips and audio, or edit and extend a video you already have. ByteDance's faster Seedance 2.5 model, at 480p or 720p, from 30 credits per second.
Seedance 2.5 Flash is the faster, lower-cost member of ByteDance's Seedance 2.5 video family. BytePlus, ByteDance's cloud arm, lists it as fully available in its ModelArk API under the id dreamina-seedance-2-5-flash-261115, and launch reports date its release to November 15, 2026. That is a few months after the flagship Seedance 2.5 arrived on July 31, 2026. In Dreamina, ByteDance's creative app, it appears as a separate model you can pick.
Flash keeps most of what makes Seedance 2.5 useful: text-to-video, first and last frames, mixed image, video and audio references, video editing, extension and sound generated with the picture. It gives up the higher resolutions and part of the reference capacity. BytePlus's model brief claims about twice the speed and roughly 34% lower cost than Seedance 2.5 for 720p text and image-to-video; those are vendor figures, not independent benchmarks.
On Flux1 AI, every one of those modes is in the generator above. Clips run 4–30 seconds at 480p or 720p, with sound on by default. Our test clips took about 90–110 seconds each, and all three examples on this page came back with a soundtrack. The page uses a third-party API channel for the model, so a few options from BytePlus's own API are not offered; they are listed under "Good to know" below.
Why you might pick Flash over the other video models on this site.
Set any length from 4 to 30 seconds; most video models here stop at 15. A longer single take keeps the same characters, light and sound across the shot, so you need fewer cuts and less stitching.
Flash generates ambience, effects and speech together with the picture. Write the dialogue in quotes and describe what we should hear. The night-market example below has the sizzle, the crowd and the cook's line. Switch sound off when you only need the picture; the price stays the same.
Reference mode takes up to 10 images, 5 video clips and 5 audio files at once: images for characters, products or outfits, a clip for camera movement or choreography, audio for rhythm or a voice. Call them Image 1, Video 1, Audio 1 in the prompt.
Upload a clip of 4–30 seconds and describe the change: swap the weather, restyle the scene, replace an object. Or extend a clip with 4–30 new seconds that continue its motion and sound. Both use the cheaper video-input rate.
Real outputs from this model with the exact prompts, unedited. Turn the sound on.
Text to video, 10 seconds at 720p. The camera move, the stall and the audio all come from the prompt, and the cook's line is spoken on cue. This clip cost 680 credits.
Prompt · 16:9 · 720p · 10s: A street food cook at a night market tosses noodles in a flaming wok. The camera starts on a close-up of the wok, then slowly pulls back to reveal the busy stall, steam and hanging lanterns. Sizzling oil, crowd chatter, distant music. The cook glances up and says: "Two more minutes!"
21:9 framing for a cinematic opener, rendered at 480p for 240 credits. Fog, light and the tracking move stay steady; the soundtrack is wind and hooves.
Prompt · 21:9 · 480p · 8s: Wide cinematic shot of a lone rider on horseback crossing a misty valley at dawn. Slow sideways tracking camera, soft golden light breaking through the fog. Hooves on wet grass, wind, a distant bird call.

Image to video from the same photo we use on the Grok Imagine 1.5 page, so you can compare the two models. The output keeps the photo's square shape. 6 seconds at 720p, 408 credits.
Prompt · Image to video · 720p · 6s: The corgi shakes the rain off its coat, then trots toward the camera wagging its tail. Raindrops splash on the cobblestones, the camera lowers to its eye level.
Four steps from idea to download. Credits come back automatically if a clip fails.
Text for a scene from words alone; Image to start from a first frame (and optionally end on a last frame); Reference to mix images, clips and audio; Edit or Extend to work on a video you upload.
Upload PNG, JPG or WEBP images and MP4 or MOV clips. Reference clips and audio must each be 2–30 seconds, 30 seconds in total per type. A video to edit must be 4–30 seconds long.
Describe the action, the camera and the sound. Choose 480p for tests and 720p for finals, a ratio for text and reference clips, and a length. The button shows the exact credit cost.
Most clips are ready in one to two minutes. You can leave the page; finished videos appear in My Videos as MP4 files. Rejected or failed jobs are refunded in full.
Flash follows long, specific prompts well. Describe one clear shot, and say what we hear.
All models share the same credit balance. The last column is the cost of the same 8-second 720p clip without video input.
| Model | Best for | Inputs | Length | Max resolution | Credits for 8 s |
|---|---|---|---|---|---|
| Seedance 2.5 Flash | Long takes, references, edit and extend | Text, frames, references, video | 4–30 s | 720p | 544 |
| Seedance 2.5 | The most detailed Seedance output | Text or image | 4–15 s | 720p | 896 |
| Seedance 2.0 | Multi-shot scenes and motion detail | Text or image | 4–15 s | 720p | 576 |
| Wan 3.0 | Long clips at a low price | Text, image or references | 2–30 s | 1080P | 144 |
| Gemini Omni 1.2 Flash | Cheap clips up to 4K, video edit | Text, images or video | 3–15 s | 4K | 56 |
On this site, an 8-second 720p clip costs 544 credits with Flash and 896 with Seedance 2.5. Flash also goes to 30 seconds and adds references, edit and extend, which our Seedance 2.5 page does not offer. In BytePlus's own API the flagship adds finer detail, more references, 1080p or 4K and a Draft mode. Iterate with Flash; render a hero shot with 2.5 if it needs the extra detail.
Wan 3.0 also does text, frames and references up to 30 seconds, and it costs much less: 144 credits for 8 seconds at 720p against 544. Wan also renders 1080P. Flash adds video editing and extension, and is the closer match to Seedance's look and motion. If budget is the main concern, start with Wan; if you need to edit or continue footage, use Flash.
You pay per second of video. Sound costs nothing extra. When a clip goes in (reference clips, edit, extend), a lower rate applies to the output plus the input seconds, which is how the model provider bills it.
| Clip | Settings | Credits |
|---|---|---|
| Shortest | 4 s at 480p | 120 |
| Default | 5 s at 480p | 150 |
| Default, sharper | 5 s at 720p | 340 |
| Longest | 30 s at 720p | 2040 |
| With a reference clip | 5 s at 720p + a 5 s clip, 41/s | 410 |
| Extend | 5 new seconds on a 10 s clip at 720p | 615 |
| Edit | A 10 s clip at 720p | 820 |
Basic · 300 credits/mo
≈ 2 clips of 4 s at 480p, or 1 at 720p
Pro · 800 credits/mo
≈ 6 clips of 4 s at 480p, or 2 at 720p
Max · 2000 credits/mo
≈ 16 clips of 4 s at 480p, or 7 at 720p
Worked example: a 10-second 720p clip without video input costs 68 × 10 = 680 credits. With input video, each second of output and input costs 18 credits at 480p or 41 at 720p. An edit returns a clip as long as the original, so it is charged for both. Credits are taken when you press Generate and refunded automatically if the job fails. New accounts get 10 free credits, not enough for a clip (120 credits minimum).
Where Flash and this page stop, compared with Seedance 2.5 and BytePlus's own API.
Flash renders 480p or 720p only, with no native 1080p or 4K. That is a limit of the model, not of this site. For a higher-resolution master, use Wan 3.0 at 1080P or Gemini Omni 1.2 Flash up to 4K, or upscale the result with our video upscaler.
Flash has no Draft mode; that cheap preview step belongs to Seedance 2.5 in BytePlus's API. The API can also return the last frame and use web search; this page does not expose either. To chain shots, use Extend instead.
An extended clip contains only the 4–30 seconds that were added, not your original joined to them. Put the two files together in any editor. Edit, extend and frame-based clips keep the shape of your input, so the ratio picker only appears for text and reference clips.
Prompts and images are screened before generation and outputs are moderated by the model provider. Sexual content, real people without consent and copyrighted characters are refused, and BytePlus restricts realistic human faces as references. Refused jobs are refunded.
Not on Flux1 AI. A clip starts at 120 credits (4 seconds at 480p), and new accounts get 10 free credits, enough to try image models but not a Flash clip. Dreamina advertises its own free entry offer; check your account there for the current terms.
BytePlus lists Seedance 2.5 Flash as fully available since mid-November 2026, and launch reports date it to November 15, 2026. You can use it in BytePlus's ModelArk API, in ByteDance's Dreamina app and on third-party sites such as this one. The flagship Seedance 2.5 came out on July 31, 2026.
Flash is the faster, cheaper variant: BytePlus claims about twice the speed and roughly 34% lower cost for 720p text and image-to-video. In exchange it stops at 720p, takes fewer references (10 images, 5 clips and 5 audio files against 30, 10 and 10) and has no Draft mode. Both make 4–30 second clips with sound.
Flash belongs to the newer 2.5 generation and does more: up to 30 seconds instead of 15, reference clips and audio, edit and extend. On this site it also costs slightly less for the same clip: 544 credits for 8 seconds at 720p against 576 for Seedance 2.0.
Yes. BytePlus offers it through ModelArk as dreamina-seedance-2-5-flash-261115, billed at $7 per million tokens, or $4.30 when the input includes video. BytePlus estimates $0.76 for a 5-second 720p clip. This page is a no-code way to use it with the same credits as the rest of Flux1 AI.
No. Flash renders at 480p or 720p. If you need a sharper file, upscale the finished clip or pick a model with higher native resolution, such as Wan 3.0 (1080P) or Gemini Omni 1.2 Flash (up to 4K).
Yes. In Edit mode, upload a 4–30 second clip and describe the change; the result is as long as your clip. In Extend mode, the model adds 4–30 seconds that continue your clip. Both are billed at the video-input rate on the input plus the output seconds.
Sexual content, violence against real people, copyrighted characters and realistic faces of real people used without consent. Requests are screened before generation and outputs are moderated; anything refused is refunded automatically.
Write a prompt, add a photo or a clip, and get a video with sound in about two minutes.