Seedance 2.5 Upscaled

Seedance 2.5 Upscaled: Image-to-Video and Text-to-Video AI Model

What is Seedance 2.5 Upscaled?

Seedance 2.5 Upscaled is an efficient delivery path for ByteDance's Seedance 2.5 video generator. Instead of rendering at full target resolution, it generates one resolution step lower and runs an AI video upscaler to deliver the final 720p, 1080p, or 4K clip. You still get the Seedance 2.5 engine underneath: cinematic motion, multi-shot storytelling, and synchronized native audio, produced from a text prompt, a starting image, or a stack of multimodal references. It runs as a synchronous API on Segmind, so a single POST returns an MP4 or MOV with an audio track.

Key Features

• •Upscaled delivery: renders one step below the target, then AI-upscales to 720p, 1080p, or 4K for high-resolution output from a lighter render.
• •Native audio generated jointly with the visuals, so dialogue, sound effects, ambient, and music stay in sync with the motion.
• •Multi-shot storytelling in one prompt using Shot 1: and Shot 2: scripting.
• •Deep reference control: up to 30 reference images, 10 reference videos, and 10 reference audios, cited inline as @Image 1, @Video 1, @Audio 1.
• •Modes: text-to-video, image-to-video, first-to-last-frame transitions, and reference-to-video.
• •Durations from 4 to 30 seconds, seed control, bitrate mode, and codec selection.

Best Use Cases

Seedance 2.5 Upscaled fits ad spots, product demos, social shorts, music-driven clips, and pre-visualization where you want high-resolution delivery without a full-resolution render. The reference stack keeps a character, product, or color palette consistent across a shot, which matters for brand and campaign work. Longer single-pass durations suit short narrative scenes that move from setup to resolution in one take, and image-to-video mode animates a single still into a moving scene.

Prompt Tips and Output Quality

Write the shot as evolving motion: subject and action first, then scene, style, camera move, and audio. Name what each reference controls, and cite it inline. When you pass a first_frame_url, set aspect_ratio to adaptive so the output follows the source frame. Avoid real human faces in first frames, which content policy blocks; use reference images for characters instead.

How to Use Seedance 2.5 Upscaled

Seedance 2.5 Upscaled turns a prompt, a starting image, or a set of references into a cinematic clip with synchronized native audio, delivered at your chosen resolution through an AI upscaling pass. The model renders one step below the target resolution and upscales to it, so you choose the output you want and the pipeline handles the rest.

Pick your mode

• •Text-to-video: Send only a prompt. Choose a fixed aspect_ratio (16:9, 9:16, 21:9) for framing.
• •Image-to-video: Set first_frame_url to animate outward from a still. Keep aspect_ratio on adaptive so the output follows the source frame. Do not use real human faces in the first frame.
• •First-to-last-frame: Add last_frame_url with first_frame_url to guide a controlled start-to-end transition.
• •Reference-to-video: Pass up to 30 reference_images, 10 reference_videos, or 10 reference_audios for character, style, motion, or audio consistency, and cite each one inline as @Image 1, @Video 1, @Audio 1. Reference images cannot be combined with first_frame_url.

Write effective prompts

Lead with the subject and action, then describe the scene, style, camera move, and audio. For multi-shot sequences, script shots with Shot 1: and Shot 2: for continuous scene changes. Name what each reference controls so the model applies it correctly.

Recommended Settings

• •resolution: 1080p for most delivery; 4k for hero shots; 720p for quick previews.
• •duration: 5 seconds to start; raise toward 30 for multi-shot stories. Use -1 for video edits so output follows the input.
• •aspect_ratio: adaptive with first_frame_url; a fixed ratio for text-to-video.
• •generate_audio: on for speech and ambient scenes; off for silent clips.
• •bitrate_mode: high for maximum fidelity; standard for smaller files.
• •seed: fix a value to iterate the same scene; -1 for fresh variations.

Start from these defaults, review the first render, then adjust resolution and duration before scaling up.

How is Seedance 2.5 Upscaled different from Seedance 2.5?

It renders one resolution step lower and upscales to your target, delivering 720p, 1080p, or 4K from a lighter render while keeping the same engine and features.

Does it generate audio?

Yes. With generate_audio on, it co-generates synchronized dialogue, sound effects, ambient, and music in the same pass.

What is the maximum clip length?

Up to 30 seconds in a single generation; set duration to -1 for video-edit prompts so output follows the input.

Can I keep a character or product consistent?

Yes. Supply up to 30 reference images and cite them as @Image 1, @Image 2 in your prompt.

Is the output browser-friendly?

Yes. Every upscaled clip is delivered as H.264 MP4 with AAC audio, which plays in every browser and editor.