How to use Seedance 2.0: create your first video
Quick answer
To use Seedance 2.0, open a provider that offers the model, choose text-to-video or image-to-video, describe one shot, set its format and duration, then generate and inspect the downloaded result. You can add image, video and audio references when the shot needs them. You do not need to start with a complicated multi-reference prompt.
For a first attempt, use a 5-second shot with one subject, one action and one camera move. The example below shows the decisions from input to final file. It is a suggested workflow, not a claim that we generated or benchmarked this example.
Create a video with Seedance 2.0
Start with a prompt or source image, choose Seedance in Magic Hour, and review the result before spending more credits.
Create with Seedance 2.01. Choose where to run Seedance 2.0
Seedance is a model made by ByteDance; the website or API hosting it controls access, payment, uploads and exports. Different hosts can expose different settings for the same model family.
Route | When it fits | What to check first |
|---|---|---|
You want a documented browser workflow | Seedance 2.0 requires Standard or higher; confirm current credits and access in your account | |
You want the official model integration and API controls | Model activation, billing requirements and the chosen variant’s supported inputs | |
You already use that platform | Exact model version, available modes, region, input limits and job price |
Runway’s official Seedance guide says to open its web app, select Video, then Seedance 2.0 from the model menu. Its modes are References, Start / End frames and Text to Video. Runway currently lists 5–15-second outputs at 480p, 720p or 1080p; 1080p requires Credits Mode. Explore availability depends on your plan.
For ModelArk, follow the official Seedance 2.0 tutorial. Standard, Fast and Mini are different variants. Their settings and rates are not interchangeable with another host’s. These details were checked September 10, 2026.
If the exact Seedance model is not a requirement and your immediate goal is to animate a still, Magic Hour image-to-video provides a browser workflow with its own available models and settings. Choose the route that supports the asset you need to make.
2. Pick the input mode for your first shot
Text-to-video: Start here when you want a new scene and do not need an existing image to be reproduced. Describe the subject, action, environment and camera.
Image-to-video: Start here when you have an approved product photo, illustration or composition. Attach it as the starting frame, then describe the movement. Prepare the image in the intended aspect ratio so important details are not cropped.
Reference-to-video: Use this when a new scene should borrow something from an asset: an object’s appearance, a camera move or an audio cue. Upload the files and state what each should contribute. Follow the labels the provider actually creates.
A reference image and a starting frame are not equivalent. The former guides the new scene; the latter defines its intended opening. The reference guide explains the actual input roles and shows examples for more complicated jobs.
3. Write one achievable scene
For this walkthrough, imagine you need a short landscape clip of a toy boat moving through a shallow pool. You can start from text or from your own approved image of that scene.
A small yellow toy sailboat moves slowly across a shallow blue pool. Gentle ripples trail behind it. The camera tracks alongside at water level in one continuous shot. Soft afternoon light and quiet water sounds. No people or writing.
The yellow toy sailboat in this image moves slowly forward, leaving gentle ripples. The camera tracks alongside at water level. Preserve the boat’s shape and the pool setting. Quiet water sounds.
The image version spends fewer words reconstructing what is already visible. Both prompts define a specific movement and camera behavior. Neither relies on an undocumented command or claims the model must obey perfectly.
Keep the first brief small enough to evaluate. “Make a cinematic advertisement” leaves too many decisions open; a boat moving through one scene gives you a clear result to accept or reject.
4. Choose settings and check the price
For the example, 5 seconds, 16:9 and 720p with audio enabled is a reasonable starting configuration where the host supports it. Choose 9:16 instead if the actual deliverable is vertical; frame the reference accordingly. Use your intended final resolution when the decision depends on fine detail.
Read the displayed job quote before generating. A subscription’s headline price is not the cost of every possible clip. Duration, variant, resolution and reference inputs can affect billing. Our Seedance pricing guide separates hosting plans from API usage.
Budget for the work, not just the winner. For three 5-second shots with two candidates each, you would generate 30 seconds to select 15 seconds of footage. This is an example budget assumption, not a measured success rate.
For a model-specific resolution table and more settings examples, use the Seedance 2.0 settings guide.
5. Generate, inspect and revise one decision
Submit once, then watch the existing job’s status. Avoid creating duplicate jobs because a result is taking longer than expected.
Review the finished clip from beginning to end. For the toy boat example, check:
- Does the boat stay the same shape and color?
- Is it moving forward, rather than the background sliding unnaturally?
- Does the camera follow the requested path?
- Are ripples and sound plausible for the movement?
- Is the framing usable in the intended placement?
If the boat changes shape, inspect the reference and simplify the motion. If it moves correctly but the camera does not, revise the camera instruction. Change one decision at a time so you can understand what the next attempt is testing.
A revised prompt is a new generation, not a guaranteed repair of the previous file. Keep the accepted result before experimenting further.
6. Download and finish the actual deliverable
Download the accepted video and save its prompt, inputs, model and settings. In ModelArk, generated video URLs expire after 24 hours, according to its video generation documentation. A saved preview link is not a durable copy of the file.
Add exact subtitles, product claims, prices and CTAs in your editor. Review the final export after cropping, upscaling or combining clips; those steps can change what the audience sees. Use source assets and commercial permissions appropriate to the project.
For a longer sequence, plan separate shots around the story. References can help carry appearance or motion between clips, but neither repeated prompting nor chaining extensions guarantees seamless continuity. Make the cut where it looks natural and keep the original approved footage.
Common starting problems
Problem | What to do next |
|---|---|
Seedance 2.0 is missing from the menu | Check the provider, account eligibility and exact model name |
Generate is disabled | Inspect the upload requirements and validation message; Runway exposes details when you hover over the disabled button |
A reference is ignored | Confirm the file is attached and the prompt uses its actual label |
A portrait is rejected | Use the provider’s permitted portrait workflow; ModelArk documents portrait-asset requirements |
A generated clip is not the requested duration | Check the selected duration and whether an automatic-duration option was enabled |
The scene is visually correct but dialogue fails | Retest the exact script with audio on; allow enough time for the line |
A completed file will not preview | Download it and check codec support before assuming generation failed |
Frequently asked questions
Not through the 2.0 workflows covered here. ModelArk documents 4–15 seconds per generation, and Runway lists 5–15 seconds. A five-minute finished piece needs multiple shots and editing. Those output limits are separate from the limits on uploaded reference clips.
Not as full-length inputs in the documented ModelArk 2.0 workflow. It allows up to three reference videos with no more than 15 seconds combined. Trim the relevant motion or camera example first. Audio references have their own limits; check the host before uploading.
The browser and API routes in this guide run generation on the provider’s infrastructure. You need a compatible device and connection for the interface, uploads and downloads; buying a faster local GPU does not accelerate a hosted generation queue.
There is no universal free allowance across hosts. Runway requires an eligible paid plan for this model, while API and other provider offers have their own billing. Verify the current offer and the price of the selected job before committing.
No. ByteDance announced Seedance 2.5 in July 2026. This article intentionally covers 2.0; do not apply its limits or instructions to another version without checking that version’s documentation.
