Seedance 2.5 Is Here: 30-Second Takes, Ready to Generate
We've been waiting for this one — and we kept our word: live on release day. Thirty-second single takes, prompts that read like shot lists, edits that don't mean reshoots — Seedance 2.5 moves AI video from generating to directing. It's in the generator below right now: same prompt box, same credits, text-to-video, image-to-video, and reference-to-video from 4 to 30 seconds at up to 4K.
- What it is
- ByteDance's next-generation AI video model — unveiled at the Volcano Engine FORCE conference on June 23, officially released July 31, 2026.
- Video
- 30-second single takes you can direct second by second; voice, sound effects, and music generated with the picture in one pass; native 4K. A beta long-video mode reaches 180 seconds.
- References
- Up to 50 multimodal references: images, video, audio, style — with green-screen or white-model inputs driving motion and blocking, 3D blockouts for previz, and region-level edits after generation.
- On Molyin
- Live — generate below right now: text-to-video, image-to-video, and reference-to-video, 4–30 seconds, up to 4K, on the same credits as everything else.
Facts on this page verified as of 2026-07-31.
Live on Release Day — Roll Camera
Seedance 2.5 is ready below: text-to-video, image-to-video, and reference-to-video on a 4-to-30-second slider at up to 4K, in the same prompt box on the same credits. Script your shot list right into the prompt ([0–5s], [5–10s]…) and get a 30-second take in one pass. The 2.5-exclusive input slots — expanded references, region editing — are rolling out next.
What Is Seedance 2.5?
Think of Seedance 2.5 as a handover of the director's chair. With earlier models, you wrote a prompt and the model decided how to shoot it; with this generation, the camera answers to you. Across a 30-second single take you can script the timeline second by second — what happens at [0–5s], where to cut at [5–10s] — and the model executes your shot list like a screenplay, while green-screen footage or a white-model reference dictates exactly how characters move and where the camera goes. It's ByteDance's next-generation video model, unveiled at the Volcano Engine FORCE conference on June 23 and officially released July 31 — and in our view, the upgrade most worth waiting for this year.
The rest of the numbers hold up too: up to 50 multimodal references pin down characters, products, style, and soundtrack in one shot. Edits no longer mean reshoots — the official demo is a single instruction, "delete everyone in the shot except the protagonist," with lighting, motion, and audio carrying over untouched. Voice, sound effects, and music generate with the picture in one pass, a rap verse lip-syncs accurately across eight languages, and the stray subtitles and phantom background music of earlier models are gone. If you make ads, product videos, or narrative shorts, these aren't spec-sheet lines — they're days off your schedule.
Our approach is simple: no press-release recital — live on release day. Seedance 2.5 is in our AI video generator right now, sharing one credit system with the Seedance 2.0 family, Kling, and Veo — still at some of the lowest per-clip prices anywhere. Today's shot list, today's footage.
What's New in Seedance 2.5
Six upgrades that change how you work — not a faster toy, but a crew that listens.
30 Seconds That Follow Your Cut
One prompt directs the whole take: establish at [0–5s], cut at [10–20s], resolve at [20–30s] — the model treats your shot list as a screenplay, scene changes included. A beta long-video mode stretches to 180 seconds.
50 References, Pinned in One Pass
Feed it images, video, audio tracks, and style guides all at once — with green-screen footage or a white-model reference dictating character motion and spatial blocking, so characters, products, and brand assets never drift between shots.
Edit Without Reshooting
Circle what needs fixing; everything else stays put — lighting, motion, composition, audio. The official demo is one sentence: delete everyone in the shot except the protagonist.
Sound On from the First Frame
Voice, sound effects, and music generate with the picture in a single pass, with a rap verse lip-syncing accurately across eight languages — and the old quirks, random subtitles and stray BGM, are cured too.
No Face-Melt at 30 Seconds
Characters, products, and multi-person scenes hold their identity across the full take, with prompt adherence up roughly 20% over Seedance 2.0 — long clips you can finally show a client.
Blockout First, Then Roll
Walk the scene with a 3D white-model blockout — layout, camera path, motion — and commit to the final generation only once it works. Previz habits, generative speed.
Seedance 2.5 vs Seedance 2.0
Spec by spec — and both columns are live on Molyin. Seedance 2.5 shoots 30-second takes; the 2.0 family remains the value pick.
| Capability | Seedance 2.0 | Seedance 2.5 |
|---|---|---|
| Max single-pass duration | 15 seconds | 30 seconds + second-by-second direction (180s beta long mode) |
| Reference inputs | ~12 mixed (images / video / audio) | Up to 50 multimodal, incl. green-screen / white-model control |
| Editing a generation | Regenerate the full clip | Region-level edits that keep lighting, motion, and audio |
| Audio | Native audio with lip sync | Voice + SFX + BGM in one pass; multilingual lip sync, cleaner output |
| Max resolution | Up to 4K (10-bit H.265) | Native 4K |
| Prompt adherence | Baseline | ~20% better |
Specs from official release material. Molyin currently offers text-to-video, image-to-video, and reference-to-video at 4–30 seconds; expanded references and region editing are rolling out next.
Seedance 2.5 Across the Whole Lineup
Every row in this table is live on Molyin right now — including the just-released Seedance 2.5 and the entire Seedance 2.0 family it upgrades. Pick the tier that fits; credits work everywhere.
| Model | Lineup role | Availability | On Molyin |
|---|---|---|---|
| Seedance 2.5 | Next-gen flagship, 30-second takes | Officially released July 31, 2026 | Live — generate on this page |
| Seedance 2.0 | Flagship quality, up to 4K | Generally available | Live |
| Seedance 2.0 Fast | Speed tier, 720p | Generally available | Live |
| Seedance 2.0 Mini | Lowest cost per clip | Generally available | Live |
| Kling 3.0 | Rival flagship, named elements | Generally available | Live |
| Veo 3.1 | Google flagship, native audio | Generally available | Live |
Availability verified July 31, 2026 — Seedance 2.5's official release day.
Seedance 2.5 FAQ
The questions creators actually ask, answered straight.
Start Creating with Molyin Today
Sign up free and turn your first idea into a cinematic video in minutes.