Text to video
Direct the subject, action, camera, light, sound, and exclusions in one production brief.
BYTEDANCE MODEL · YILUNA WORKFLOW
Create 4–30 second clips from text, first or last frames, or mixed-media references. Choose 480p or 720p, keep generated sound on when the shot needs it, and move every render into a connected Yiluna workflow.
10 seconds · 16:9 · 720p · sound on · 480 credits
02 / THE DIRECT ANSWER
Seedance 2.5 is ByteDance's multimodal AI video model, available inside Yiluna as a complete shot workflow rather than a disconnected model demo.
It can generate 4–30 second clips from a text brief, animate a first or last frame, or use mixed image, video, and audio references. Yiluna adds the production layer around that render: editable prompts, a live credit quote, asynchronous job status, persistent history, download, and a direct path into the next iteration.
The practical difference is continuity. A promising result does not disappear into a one-off tool—you can inspect what it cost, preserve the source direction, and reuse the output in the same image-and-video workspace.
03 / INPUT CONTROL
Direct the subject, action, camera, light, sound, and exclusions in one production brief.
Anchor the opening, ending, or both with up to two frame images before describing the motion between them.
Art-direct continuity with image, video, and audio references instead of asking one prompt to carry every decision.
Multimodal ceiling: up to 30 image references, 10 video references, and 10 audio references per job. Use only what meaningfully constrains the result.
04 / PROMPT EVIDENCE
It locks the subject first, then separates camera, transformation beats, sound, and exclusions. That structure is more reusable than a pile of cinematic adjectives.
Load it in the generator ↑Photoreal live-action one-take, 16:9, 10 seconds. A maritime chart vault: teak map drawers, brass wall lamps, indigo-ink sea charts, salt-white paper edges, warm dust in the light. Real camera, real lighting, real materials. No illustration, no concept art, no template face.
Photoreal live-action one-take, 16:9, 10 seconds. A maritime chart vault: teak map drawers, brass wall lamps, indigo-ink sea charts, salt-white paper edges, warm dust in the light. Real camera, real lighting, real materials. No illustration, no concept art, no template face. Subject stays locked: an East African woman cartographer in her early thirties, a real independent actor, warm deep-brown skin, short twisted hair with one thin gold cuff on the left, a faint scar on the left brow, a fuller lower lip, one small brass hoop in the right ear, ochre linen long coat, ivory shirt, a brass compass on a cord. She walks toward camera with a calm, grounded gait. Same face, same coat, same compass for the whole shot. The camera dollies backward at her walking speed. No cuts. The corridor transforms only through chart materials: drawers slide open and charts pull themselves out; indigo ink floods the floor like a shallow tide and reflects her steps; compass roses lift into paper moths; the moths cross the lens and become salt-white terns; teak walls erode into a cliffside tide archive with stone-carved shipping lanes; the lanes bend into a canyon of sailcloth pages; the canyon narrows into a suspended bridge of stitched portolan charts; latitude lines rise into an orrery of folded paper continents. Beat sheet: establish the vault and the walk as charts tremble; ink tide and moths; terns and sailcloth canyon; she reaches toward camera, the orrery snaps shut like a brass atlas cover, and the camera pulls back to a drafting table where that atlas lies closed under a paperweight beside the same compass, ordinary harbor morning light in the window. Sound only: boot heels on teak, paper drawers, ink wash, moth wings, sea wind, terns, sailcloth, one muted brass-cover close, distant harbor rope. Low strings under, no dialogue. Avoid a walnut private library, green banker lamps, a man in a dark coat, paper birds, gulls, a manuscript canyon, an open literary book on a desk, on-screen text, subtitles, watermark, logo, UI, collage borders, face swap, costume change, plastic skin, influencer template face, painterly look, cartoon, neon city, guns, gore.
05 / FROM PROMPT TO ASSET
Use text for open exploration, frames for visual control, or multimodal references when identity, movement, and sound all matter.
Duration and resolution update the credit estimate before the job leaves the studio. There is no hidden subscription gate.
Keep the result in history, download it, revise the prompt, or carry a stronger still into the next image-to-video pass.
PAY AS YOU GO
06 / FAQ
Yes. Seedance 2.5 uses the credits in your Yiluna balance, so a recurring subscription is not required. Generation is paid by credits and the current cost is shown before submission.
No. Seedance 2.5 generation consumes credits. At the current rate, the lowest 4-second 480p preset costs 88 credits; the launcher checks the live estimate before you continue.
It supports text-to-video, first or last frame animation, and multimodal direction with image, video, or audio references. Yiluna currently accepts up to 30 images, 10 videos, and 10 audio references in multimodal mode.
Seedance 2.5 supports durations from 4 to 30 seconds in Yiluna, with 480p and 720p resolution options. It does not offer 1080p in the current integration.
Yes. Generated audio is enabled by default for Seedance 2.5 and can be turned off in the video workspace when a silent clip is more useful.
The landing page, real sample, prompt, settings, and live estimate are public. Sign-in is required only when you submit a generation job, and the prepared shot stays in the workspace while you sign in.