Can I mix references with first and last frames?

No. Choose reference mode or frame mode. An ending frame also requires an opening frame; frame mode uses adaptive aspect ratio.

Which image formats can I upload?

The Seedance 2.5 reference uploader accepts JPEG, PNG and WebP images, up to 30 MB each. Video and audio uploads have separate limits.

Does every clip include sound?

Sound is optional. Check the sound control before submitting and describe the audio you want. Listen to the result as part of your review.

What should I check before using the video?

Check the subject, reference interpretation, framing and audio. Revise the scene or inputs when needed; a prompt or reference does not guarantee an exact result.

Can I mix references with first and last frames?

No. Choose reference mode or frame mode. An ending frame also requires an opening frame; frame mode uses adaptive aspect ratio.

Which image formats can I upload?

The Seedance 2.5 reference uploader accepts JPEG, PNG and WebP images, up to 30 MB each. Video and audio uploads have separate limits.

Does every clip include sound?

Sound is optional. Check the sound control before submitting and describe the audio you want. Listen to the result as part of your review.

What should I check before using the video?

Check the subject, reference interpretation, framing and audio. Revise the scene or inputs when needed; a prompt or reference does not guarantee an exact result.

Seedance 2.5: Plan Reference-Led Video

Seedance 2.5 for reference-led video scenes

Give a short scene a clear direction. Select Seedance 2.5 in GPT Image 2.5, describe the action, and choose text-only creation, reference materials or a first-frame setup before adjusting the output.

About

Seedance 2.5 for reference-led video scenes

Seedance 2.5 is a ByteDance video model available in the GPT Image 2.5 video generator. Start with a written scene, use reference images, video and audio to guide it, or animate a first frame with an optional ending frame. These are separate ways to prepare a request. For a product reveal, describe what moves and how the camera follows; for a character scene, explain the action and the role of each reference. Select the duration, aspect ratio, resolution and sound controls shown for your chosen mode, then review the finished clip before using it.

Ways to direct your scene

Prepare the inputs before choosing the output.

Core Features Overview

A written scene as the starting point

For text-to-video, put the main action first and add the setting, lighting and camera movement that matter. A focused scene is easier to revise than a list of unrelated events; keep dialogue or sound instructions explicit when using sound.

References with a purpose

Reference mode accepts up to 30 images, 1 video clip and 1 audio clip within the upload limits. State which material supplies appearance, movement or sound. Choose relevant inputs rather than filling every slot, and use only materials you have permission to use.

A defined opening and ending

Frame mode needs a first image; a last image can define the intended endpoint. Do not combine this mode with reference images, video or audio. Its aspect ratio is adaptive, so prepare the frames with the composition you want to carry into the clip.

Duration, format and sound

GPT Image 2.5 offers duration choices through 30 seconds and 480p, 720p or 1080p output for Seedance 2.5. Use a vertical composition for a portrait placement or a wide frame for landscape scenes where available. Check sound, framing and motion in the result before publishing.

Seedance 2.5 questions

Seedance 2.5 for reference-led video scenes FAQ

Open the GPT Image 2.5 video generator, select Seedance 2.5 and choose the input mode. Add your scene and relevant materials, then check the available settings.

Explore More AI Video Models

Veo 3.1

New

Veo 3.1 represents Google DeepMind's most advanced AI video generation technology, featuring groundbreaking native audio generation that creates synchronized sound effects, dialogue, and environmental audio alongside video content.

Try now

Wan 2.6

New

Wan 2.6 is Alibaba's video generation model delivering high-quality videos with diverse style support, smooth motion, and cinematic output from text prompts and reference images.

Try now

Sora 2

Sora 2 is OpenAI's flagship video generation model capable of producing high-quality videos from both text descriptions and image inputs. It understands complex scene compositions, character interactions, camera movements, and real-world physics to deliver cinematic results. Sora 2 represents a major leap in AI video generation with improved temporal consistency, longer duration support, and more faithful prompt interpretation.

Try now

Vidu Q4 Video Generator

New

Vidu Q4 is a video model developed by ShengShu Technology. Image-to-video turns one image into a 3–16 second clip, while reference-to-video combines 1–15 images into a 3–16 second clip. Optional audio and up to three voice references let you explore scenes in which appearance, movement and sound work together.

Try now

Kling 2.6

Kling 2.6 is Kuaishou's latest AI video generation model, recognized for its exceptional motion quality and cinematic output. Built on advanced spatiotemporal modeling, Kling 2.6 produces videos with fluid character movement, dynamic camera transitions, and rich visual detail. It supports both text-to-video and image-to-video generation, making it a versatile tool for creators seeking professional-quality AI video content.

Try now

Seedance 2.0

New

Seedance 2.0 is ByteDance's most advanced AI video generation model, unveiled in February 2026. It adopts a unified multimodal audio-video joint generation architecture supporting 4 input modalities simultaneously — text, up to 9 images, up to 3 video clips, and up to 3 audio tracks. The ground-breaking @-reference system lets you tag specific elements in your prompt and bind them to uploaded references for granular control over camera movement, character appearance, audio rhythm, and visual style. Outputs reach up to 2K resolution with native synchronized audio including multilingual lip-sync, sound effects, and background music.

Try now

Open Seedance 2.5 in the video generator

Give a short scene a clear direction. Select Seedance 2.5 in GPT Image 2.5, describe the action, and choose text-only creation, reference materials or a first-frame setup before adjusting the output.

Open Seedance 2.5 in the video generator

+1k