How to Use Seedance 2: Seedance 2.0 vs 2.5, Settings, Prompts and Prices

Short answer: "Seedance 2" is two models. Seedance 2.0 (February 2026) makes clips up to 15 seconds; Seedance 2.5 (July 2026) makes up to 30 seconds in one run. Cicadas runs both: open the Seedance 2.5 page (or the Seedance 2.0 page), choose text or one image, write one paragraph, pick 4 to 30 seconds and 480p or 720p, check the credit quote, generate, and download the MP4 from My videos. A 5-second 480p clip on Seedance 2.5 is 276 credits, about $2.73 on the $9.90 first pack.

We make Cicadas. This guide names what each version is from ByteDance's and fal.ai's own pages, says which endpoint each of our pages calls, and shows the settings as they are in our generator on October 27, 2026. There are no benchmark numbers here, because we have not measured any.

1. What "Seedance 2" means today

Seedance 2.0. ByteDance's video model announced on February 12, 2026, available in Dreamina, Doubao and through Volcano Engine's Model Ark (secondary coverage: NYU Shanghai RITS, KrASIA, read 2026-10-27). fal.ai's Seedance 2.0 text-to-video page (read 2026-10-27) lists durations from 4 to 15 seconds or "auto", 480p or 720p in the input schema, seven aspect ratios including "auto", synchronized audio on by default, and dialogue in double quotes for lip-synced speech. The same page also mentions 1080p pricing, which its schema does not offer; we offer 480p and 720p only.

Seedance 2.5. Announced at Volcano Engine's FORCE conference in June 2026 (GIGAZINE, 2026-06-24) and rolled out to Jimeng AI and Doubao in late July (TechNode, 2026-07-31). The headline change is length: "a 30-second high-quality video clip in a single run", roughly double the 2.0 ceiling, plus up to 50 reference inputs in ByteDance's own apps. fal.ai's Seedance 2.5 text-to-video page (read 2026-10-27) lists durations from 4 to 30 seconds or "auto", 480p or 720p, the same seven aspect ratios, synchronized audio that "doesn't change the token count", and a list price of about $0.473 per second at 720p and $0.2205 at 480p.

So when a search says "seedance 2", the honest reading is: 2.0 if the page is from February to June 2026, and usually 2.5 after that. If you want the longest single take, 2.5 is the one.

2. Which version Cicadas runs

Both, each on its own page, each calling the exact fal.ai endpoint of that version and never swapping one for the other:

| Page | Endpoint (fal.ai) | Lengths | Resolution | Sound | |---|---|---|---|---| | Seedance 2.5 | bytedance/seedance-2.5 text-to-video and image-to-video | 4, 5, 10, 15 or 30 s | 480p or 720p | optional, no surcharge | | Seedance 2.0 | bytedance/seedance-2.0 text-to-video and image-to-video | 4, 5, 10, 15 s | 480p or 720p | optional, no surcharge |

Both pages have been through a live generation on this site; the recorded clips are listed on each page and in the test log. Neither page offers "auto" duration, reference videos or audio, seeds or 1080p: you choose an exact length so the price is known before you submit.

3. Step by step on the Seedance 2.5 page

1. Open the Seedance 2.5 page and sign in.
2. Choose Text or Image. In image mode, upload one JPG, PNG or WebP you are allowed to use; it becomes the first frame and the clip keeps its shape (the aspect-ratio control is for text mode).
3. Write the prompt as one paragraph (see the next section). The field takes up to 2,000 characters.
4. Pick the length (4, 5, 10, 15 or 30 seconds), the resolution (480p or 720p) and, in text mode, the aspect ratio (21:9, 16:9, 4:3, 1:1, 3:4 or 9:16).
5. Leave sound on if you want synchronized audio; it costs the same either way.
6. Read the credit quote under the button, then generate. Keep the page open or come back later: the MP4 appears under the button and in My videos.
7. If the generation or the file check fails, the credits return automatically. A delivered clip is charged even if you dislike it, so start short.

4. The prompt shape that works on both versions

Five things, in one paragraph, in English: subject, action, camera, look, sound. Say what must stay stable and what must change. Ask for one continuous shot unless you want cuts; Seedance 2.0 and 2.5 both take multi-shot prompts, but a single shot is easier to judge. Put any spoken line in double quotes, as fal's model page describes for lip-synced speech, and add "no text, no logos" at the end so the model does not invent lettering.

A recorded example from our Seedance 2.5 page, image-to-video, 4 seconds, 720p, silent, generated on September 23, 2026 from the site's own illustration of a graphite coupe on a coastal road:

Its prompt asked for "one continuous four-second cinematic shot", a car that "drives steadily away from the high camera along its existing lane", a "restrained descending push forward" and "no cuts, no new vehicles near the hero car, no deformation, no text, no logos". The result keeps the car's silhouette and the road; the near foliage moves faster than the distant city, as the prompt asked.

And a recorded Seedance 2.0 text-to-video example, 4 seconds, 720p, with sound, generated on September 27, 2026:

Prompt: "At a night market, a street-food cook flips sizzling dumplings in a black wok. Flames leap and steam rises into warm lantern light. Handheld close-up, crackling oil and busy market ambience. One continuous shot, no text or logos." The flames and steam read clearly; the recorded audio is a low roar rather than distinct market chatter.

5. Prompt examples to adapt

These three are from our Seedance prompt builder page, which writes the labelled shape above for free in your browser. They are untested on this site: use them as starting points, not as promised results.

• Cinematic: "A lone astronaut walking across red dunes at sunset, wide anamorphic shot, dust lifting with each step, wind and a distant hum, slow lateral dolly. 10s, 16:9."
• Product: "A handmade ceramic mug on a café table, gentle steam rising, slow camera push, warm morning light and quiet room ambience. 5s, 9:16."
• Dialogue: "Two friends on a rooftop at night, city lights behind; one says "we should have done this years ago", the other laughs. 10s, 16:9."

For 30 seconds on Seedance 2.5, describe the whole arc in order (start, middle, end) and keep one subject; a long clip with a vague prompt tends to wander.

6. What it costs

Prices on Cicadas are set per second from fal.ai's list rates and quoted before you generate. The first order is a 1,000-credit pack for $9.90, so one credit is about one cent.

| Version | 480p | 720p | |---|---|---| | Seedance 2.5, 5 s | 276 credits (about $2.73) | 592 credits (about $5.86) | | Seedance 2.5, 10 s | 552 credits (about $5.46) | 1,183 credits (about $11.71) | | Seedance 2.5, 30 s | 1,654 credits (about $16.37) | 3,548 credits (about $35.13) | | Seedance 2.0, 5 s | 170 credits (about $1.68) | 378 credits (about $3.74) | | Seedance 2.0, 15 s | 510 credits (about $5.05) | 1,134 credits (about $11.23) |

Sound on or off costs the same. Test at 4 or 5 seconds and 480p before you buy a 30-second take. The AI video cost index compares these with the other models on the site.

7. Limits to know before you spend

• No reference inputs here. ByteDance's own apps take reference images, videos and audio; our pages take a prompt or one starting image.
• No "auto" length. You pick the seconds, which is what makes the quote exact.
• 480p or 720p. 1080p is not offered, whichever version.
• Text in the picture is unreliable. Ask for none and add titles in your editor.
• Faces and rights. Use only images you own or may use; the site refuses prompts that name real people or ask for face replacement.

Related

• Seedance 2.5 vs Wan 3.0 and MiniMax H3 vs Seedance 2.5 for side-by-side recorded clips.
• Seedance prompts to build the prompt first, then paste it into the generator.