ByteDance video model on Lovart ME
Seedance 2.5 AI Video Generator
Build a complete sequence up to 30 seconds with precise visual, motion, and audio references.
Seedance 2.5 is designed for longer audio-video storytelling and reference-led production. Start from a written brief, anchor a first or last frame, guide camera language with a clip, or supply an audio cue. Use each reference to solve a specific production decision so the model receives clear direction instead of a pile of competing signals.
- 30s
- Longer sequence
- Generate one continuous clip up to 30 seconds.
- 720p
- Maximum output
- Choose 480p for iteration or 720p for delivery.
- 7
- Aspect ratios
- Wide, square, vertical, ultrawide, and adaptive formats.
- A/V
- Joint generation
- Create synchronized dialogue, ambience, and effects.
01 / MODEL
What Seedance 2.5 adds to a video workflow
The practical gain is more room for structured scenes and more exact control over what a reference contributes.
Thirty-second story structure
Plan a hook, development, and resolution inside one generation. Assign timing, framing, action, and transitions to each shot so the longer duration remains readable.
Intent-aware video references
Use a reference clip for camera movement, blocking, rhythm, or cinematic language. State what to adopt and what to replace instead of asking for a literal copy.
Multimodal editing direction
Combine text, frames, images, video, and audio to describe a transformation or revision. Keep identity, composition, motion, and sound instructions separate and testable.
Production-oriented control
Describe camera paths, actor marks, background treatment, or green-screen requirements explicitly. Review the output before it enters compositing or an editorial timeline.
02 / INPUTS
Assign a clear role to every input
References work best when the prompt names them and explains the decision each one controls.
First and last frames
Use the first frame to lock the opening composition and the last frame to define the destination. Describe the visual path between them, including the subject action and camera move.
Reference images
Use additional images for character identity, product shape, location, palette, wardrobe, or treatment. Refer to them as @Image 1, @Image 2, and so on.
Reference video
Guide motion, staging, shot rhythm, or camera grammar with up to three videos totaling 30 seconds. Attaching video changes both the rate and billable duration.
Reference audio
Use up to three audio files totaling 30 seconds for voice, pace, music direction, ambience, or effects. State where important cues should occur in the sequence.
03 / OUTPUT
Set the delivery constraints before prompting
Duration, resolution, and aspect ratio shape the edit. Decide them before refining language or adding references.
Duration from 5 to 30 seconds
Use a short output for motion tests. Reserve 20-30 seconds for scenes that genuinely need multiple beats, dialogue, or a complete product demonstration.
480p or 720p
Use 480p to validate blocking, timing, and reference choices at a lower cost. Switch to 720p after the structure works and inspect faces, hands, edges, logos, and small text.
Ratios for the final placement
Use 16:9 or 21:9 for wide storytelling, 1:1 for balanced feeds, 9:16 for full-screen mobile, or adaptive when the strongest input should guide framing.
04 / PROMPTS
Seedance 2.5 prompt templates for real production tasks
Replace bracketed details, keep reference labels stable between revisions, and remove any instruction that does not affect the deliverable.
Thirty-second campaign story
Plan four timed beats with a clear visual handoff between shots.
Create a 30-second [ratio] campaign film for [product]. 0-4s: visual hook, [action]. 4-12s: show [problem] in a medium tracking shot. 12-23s: product demonstration using the shape and colors from @Image 1. 23-30s: clean hero reveal with space for copy. Follow the camera rhythm of @Video 1, but use the location from @Image 2. Audio: [music direction], two timed product sounds, no generated text.Production note: Keep critical label text large and verify it frame by frame.
Performance and camera blocking
Use a motion reference to define staging without copying its subject or setting.
Create a [duration]-second scene in [location]. Preserve the lead character from @Image 1. Use @Video 1 only for the entrance timing, actor marks, and dolly path; replace its wardrobe, set, and lighting. Shot 1: [framing and action]. Shot 2: [reaction and camera move]. Maintain screen direction and eyelines. Audio: natural dialogue '[line]' with quiet room tone.Production note: Name the transferable elements and the elements that must change.
Reference-led edit
Describe a controlled visual revision with explicit preservation rules.
Edit the scene from @Video 1. Keep the original timing, camera path, and performer blocking. Replace the background with @Image 1, change the wardrobe to @Image 2, and shift lighting to [description]. Preserve facial identity and contact points. Remove [object]. Add [effect] only between [time] and [time]. Keep the original dialogue and synchronize the new ambience.Production note: Review occlusions, hand contact, reflections, and audio transitions.
Vertical tutorial sequence
Turn a process into a compact mobile-first edit.
Create a [duration]-second 9:16 tutorial showing [task]. Shot 1: final result as an immediate hook. Shots 2-4: one clear action per step, close framing on hands and tools, consistent workspace from @Image 1. Final shot: repeat the finished result with clean space in the upper third. Crisp practical sound, calm voiceover, no subtitles or generated interface text.Production note: Add exact captions later in an editor when spelling must be reliable.
05 / WORKFLOW
How to use Seedance 2.5 on Lovart ME
Treat the generator like a compact production brief and review one variable at a time.
- 01
Choose text or image mode
Start with text for a scene defined by the brief. Choose image mode when the opening frame, ending frame, character, product, or visual treatment must be anchored.
- 02
Add only decisive references
Upload the smallest set of images, video, and audio that resolves identity, composition, motion, timing, or sound. Keep each media group within its displayed limits.
- 03
Set duration, format, and audio
Choose 5-30 seconds, 480p or 720p, and the destination ratio. Leave synchronized audio enabled only when sound belongs in this generation.
- 04
Generate, inspect, and revise
Review continuity, identity, physics, lip sync, hands, text, logos, and sound timing. Change one constraint per revision so you can tell what improved the result.
06 / FAQ
Seedance 2.5 questions
These answers reflect the published Seedance 2.5 capabilities and the controls available in this integration.
What is Seedance 2.5?
Seedance 2.5 is ByteDance Seed's audio-video joint generation model for clips up to 30 seconds. Lovart ME accesses it through Kie.ai and exposes text, frame, image, video, audio, resolution, ratio, duration, and synchronized-audio controls.
Can Seedance 2.5 create a 30-second video in one request?
Yes. The current integration supports outputs from 5 to 30 seconds. Longer scenes need explicit shot timing and transitions so each beat has enough room.
Can I generate from an image or use a last frame?
Yes. In image mode, the first upload anchors the opening frame and the second can define the last frame. Additional images act as general visual references.
How many reference videos and audio files can I add?
You can add up to three reference videos totaling 30 seconds and up to three audio files separately totaling 30 seconds.
Does Seedance 2.5 generate synchronized audio?
Yes. Synchronized audio is enabled by default and can include dialogue, ambience, music direction, and effects. Turn it off when sound will be produced separately.
How does reference-video billing work?
Without video input, the selected rate is multiplied by output duration. With video input, a different rate is multiplied by reference-video duration plus output duration. The generator shows the server-aligned total before submission.
Can I use the output commercially?
Usage depends on Lovart ME terms, rights to every uploaded reference, and applicable law. Clear source material, likenesses, music, trademarks, and distribution rights before commercial use.
07 / NEXT
Compare the Seedance video workflows
Choose the model that matches the duration, reference depth, and budget of the next scene.