Text to Video
Seedance 2.5 Text to Video
Turn a script, concept, or shot brief into a video workflow with one consistent SandBase model route.
Open workflowBYTEDANCE VIDEO MODEL FAMILY
Seedance 2.5,
from brief to video.
ByteDance’s new-generation video model for one-take storytelling, flexible multimodal references, and precise editing. See the official examples, then run the same workflows on SandBase.
SINGLE PASS30s
Audio-video clips up to 30 seconds, with multi-round extension.
REFERENCES50
Up to 30 images, 10 video clips, and 10 audio clips per request.
EDITINGTimestamp
Target a specific segment for character, action, or camera changes.
CONTROLClay render
Block out space, poses, and camera paths with textureless 3D.
AVAILABLE WORKFLOWS
Start from the input you have.
Each route keeps its own request schema and model details. Open a workflow to review supported parameters and run it in SandBase.
Image to Video
Seedance 2.5 Image to Video
Begin with a keyframe when composition, product placement, or character styling needs a clear visual starting point.
Open workflowReference to Video
Seedance 2.5 Reference to Video
Assign separate roles to character, product, environment, motion, and audio assets for more controllable multi-reference work.
Open workflowOFFICIAL EXAMPLES · BYTEDANCE SEED
What Seedance 2.5 can do.
Nine examples from ByteDance Seed’s Seedance 2.5 launch. Filter by capability, play the official result, and open the matching SandBase workflow.
ByteDance Seed
Long-form · Text to Video
Backstage to stage in one take
A single handheld tracking shot follows a singer from the dressing room, past the dancers, and onto a live arena stage.
Official prompt“One-take handheld gimbal tracking shot. The camera slowly pushes in through a gap in a heavy red curtain…”
Try Text to VideoByteDance Seed
Long-form · Reference to Video
Extend a clip, keep the world
Continues an existing clip with a new 30-second chase across a platform and street, keeping characters, scene, style, and sound.
Official prompt“Extend the video. Continue from the visuals and subjects in @Video 1 and generate another 30-second clip…”
Try Reference to VideoByteDance Seed
Long-form · Reference to Video
Continuous camera across three performers
The camera circles between three referenced opera performers without a cut, while costumes and sleeves move with believable physics.
Official prompt“16:9 widescreen, cinematic texture, single continuous take, smooth camera movement, no cuts.”
Try Reference to VideoByteDance Seed
References · Reference to Video
An 18-reference concert hall
Venue, soloist, instruments, orchestra, choir, and audience each come from separate references and resolve into one coherent performance.
Official prompt“Use @Image 1 for the venue. Reference @Image 2 for the pianist. Reference @Image 3 for the cello…”
Try Reference to VideoByteDance Seed
References · Reference to Video
From clay render to animated short
A textureless 3D blockout fixes camera path, pacing, and blocking; a style image supplies characters, materials, and light.
Official prompt“Refer to @Clay Render 1 for camera movement, pacing, shot-size transitions, subject trajectory, and blocking.”
Try Reference to VideoByteDance Seed
Editing · Reference to Video
Green screen to a finished story
A green-screen performance moves through three timed scenes, from training ground to match day, while the athlete and motion stay intact.
Official prompt“Using @Video 1, render the green-screen background, obstacles, wardrobe, and supporting characters.”
Try Reference to VideoByteDance Seed
Editing · Reference to Video
Re-shoot the camera, not the scene
Characters, action, and style stay the same; only the camera plan is rewritten in four timed segments, from FPV skim to top-down sweep.
Official prompt“Edit @Video 1. Keep the characters, actions, and visual style unchanged. Adjust only the camera movement.”
Try Reference to VideoByteDance Seed
Industry · Reference to Video
A history lesson brought to life
A classroom scene in an Eastern painterly style: a Southern Song street built around a referenced historical figure.
Official prompt“Expressive Eastern painterly style. A street scene in Lin’an during the Southern Song dynasty.”
Try Reference to VideoByteDance Seed
Industry · Reference to Video
Clay render to photoreal car assembly
A blockout fixes part positions, assembly order, and motion paths; a material reference turns it into a photoreal production sequence.
Official prompt“…turn the clay render into a high-end, photorealistic car assembly sequence.”
Try Reference to VideoVideos and prompt excerpts are from ByteDance Seed’s Seedance 2.5 announcement (31 July 2026) and are embedded from ByteDance’s servers; all rights remain with ByteDance. Titles and summaries are written by SandBase. Read the original for full prompts and the end-to-end short film.
BUILD A REFERENCE PACK
Give every asset one job before you generate.
Reference-to-video works best when the brief separates identity, world, movement, and sound. This prevents one image from being asked to control every part of the shot.
Character or product
Provide the approved face, wardrobe, product silhouette, material, and logo treatment. State what must stay unchanged.
Prompt role: “Use Image 1 only for the product shape and finish.”Location and light
Use a separate environment image for composition, time of day, palette, and the quality of practical light.
Prompt role: “Use Image 2 only for the location and lighting palette.”Blocking and camera
Describe subject trajectory, shot size, camera path, and the order of actions in short timed beats.
Prompt role: “0–4s entrance; 4–8s orbit; 8–12s hero frame.”Dialogue and ambience
Define the desired voice, sound effects, or ambience separately from visual direction when the route supports audio.
Prompt role: “Keep ambience quiet; use a soft room tone under the action.”PROMPT OPTIMIZATION GUIDE
Write for direction, not just description.
Strong video prompts make the subject, sequence, camera, sound, and continuity constraints explicit. This structure is designed for iteration: change one control at a time, evaluate the resulting shot, then keep only the details that improve the brief.
- 01Outcome and format
State the channel, aspect ratio, duration, and intended audience before describing the scene.
- 02Subject and setting
Name the subject, its distinguishing details, the location, time of day, and visual treatment.
- 03Reference roles
When using source files, assign each one a job: character identity, product appearance, location, motion, or sound.
- 04Timed story beats
Break a longer video into concise time ranges so actions and scene transitions have a clear order.
- 05Camera and audio
Specify shot size, movement, pace, lighting, and the desired dialogue, ambience, or music direction.
- 06Non-negotiables
Close with the details that must remain stable, such as product geometry, character identity, logo legibility, or continuity.
PROMPT STARTERS
Four production-ready ways to begin.
These are original SandBase starter prompts, written for adaptation rather than as performance guarantees. Replace bracketed values with your own assets, product rules, and approved claims.
TEXT TO VIDEO · PRODUCT AD
Single-product reveal
Use when the product can be described precisely and the goal is a concise paid-social opening.
9:16 vertical product film, 8 seconds. A [PRODUCT] rests on a matte [COLOR] pedestal in a bright studio. 0–2s: macro close-up of its signature detail. 2–5s: smooth orbit as soft daylight reveals the material. 5–8s: pull back to a clean hero frame with space for a headline. Premium, tactile, realistic lighting. Keep the product proportions, finish, and label placement consistent. No readable text generated in frame.
Improve it: add one measurable product detail, then test alternate camera paths before adding more adjectives.
REFERENCE TO VIDEO · BRAND STORY
Asset roles, not a pile of files
Use when a product, talent, setting, and motion reference must stay distinct.
16:9 cinematic brand moment, 12 seconds. Use @Image 1 only for the [PRODUCT] shape and materials; use @Image 2 only for the lead talent's wardrobe and appearance; use @Image 3 only for the location palette. 0–4s: the lead enters from frame left, holding the product. 4–9s: medium tracking shot as they use it naturally. 9–12s: wide sunset hero composition. Warm practical lighting, gentle handheld movement, natural room ambience. Preserve the product design and lead identity across every shot.
Improve it: give every reference a singular purpose; conflicting instructions create less controllable shots.
VIDEO EDIT · TIMELINE CONTROL
Change one layer at a time
Use when the source performance should stay intact but the shot needs a new setting or camera treatment.
Edit @Video 1. Keep the performer, action timing, and overall visual identity unchanged. Change only the environment and camera treatment. 0–3s: close follow shot through a modern gallery. 3–7s: side tracking shot as projected light moves over the walls. 7–10s: slow crane out to reveal the gallery audience. Photorealistic lighting, stable subject proportions, continuous motion, no new text or logos.
Improve it: explicitly say what must not change before describing the edit; keep each iteration limited to one primary adjustment.
VIDEO EXTEND · CONTINUITY
Continue with a clear next beat
Use after a successful first shot when the story needs another action without resetting the world.
Extend @Video 1 by [DURATION]. Continue from the final camera position and preserve the same lead character, wardrobe, location, color grade, time of day, and ambient sound. The [SUBJECT] notices [NEW EVENT], reacts with a brief pause, then walks toward it as the camera follows from behind. End on a stable medium-wide frame that can support a further extension. Keep motion physically plausible and transitions smooth.
Improve it: describe the next event and the required ending frame so the extension has an intentional handoff.
BEFORE YOU SHIP
Evaluate output against the brief.
IdentityDoes the person, product, and any approved brand asset remain recognisable?
StoryDoes every timed beat happen in the requested order and end on a usable frame?
CraftAre framing, motion, lighting, and audio appropriate for the intended placement?
ComplianceAre claims, logos, likeness rights, and supplied assets approved for the campaign?
Capability boundaries and reference-workflow guidance are based on ByteDance’s public Seedance 2.5 announcement and the BytePlus prompt guide.
INDEPENDENT BENCHMARK SNAPSHOT
Ranked for text-to-video quality.
In the October 2026 search snapshot of Artificial Analysis’ AA-Video-T2V v2.0 leaderboard, Dreamina Seedance 2.5 was listed #3 with an Elo of 1142. The leaderboard uses blind preference votes and changes as new models and votes are added.
AA-VIDEO-T2V V2.0#3Elo 1142 · snapshotOpen leaderboard ↗
Benchmark placement is independent of SandBase and should be treated as one decision signal alongside your own brief, asset set, controllability needs, latency, and live price.
COMMON QUESTIONS
Plan your first Seedance 2.5 workflow.
Use these answers to turn an idea into a smaller, testable first run before committing to a larger campaign or integration.
What is Seedance 2.5 best for?
Start with Seedance 2.5 when the brief needs a story-led video, a product or character that stays recognisable, or references that guide composition, motion, and sound. Compare the result against your own assets and delivery requirements before making it a production default.
Which Seedance 2.5 workflow should I choose?
Use text-to-video for a prompt-led concept, image-to-video when a keyframe should guide composition, and a reference-led workflow when character, product, location, movement, or audio assets each need a defined role. Open a workflow card above to see its live schema.
How should I organise reference assets?
Give each asset one purpose. For example, reserve one image for the product design, one for the performer, one for the setting, and a video only for motion or camera language. Avoid asking multiple references to control the same detail.
Why does my prompt lose character or product consistency?
Reduce competing instructions. Start with one subject, one setting, and one camera plan; explicitly name the details that must not change. Add references only after the baseline composition is working, then change one control at a time.
How do I extend or edit a successful video?
Describe the next narrative beat and the ending frame you need, while explicitly preserving the previous character, wardrobe, setting, color treatment, and ambience. For edits, state what must stay unchanged before naming the single element to replace.
How should I evaluate cost and output quality?
Run a short representative test before scaling. Track usable-shot rate, reruns, manual cleanup, and final delivery fit alongside price. Use the model detail page for current parameters and pricing rather than relying on a fixed price in this guide.
Were the example videos generated on SandBase?
No. The official cases are ByteDance Seed’s published Seedance 2.5 examples, embedded from ByteDance’s servers with attribution. They show what the model can do; your own results depend on your prompt, references, and settings.
CONTINUE ON SANDBASE
Keep exploring the right route.
Move from this family guide into the video catalog, a model-level workflow, pricing, or API documentation—without restarting your research.