Seedance output looks blurry or soft: resolution settings and upscale options that fix quality

Blurry or soft Seedance output is usually a resolution setting issue: raise the metadata.resolution value to 1080p (about $0.470 per second) or select native 4K (about $4.90 per 5 second clip) instead of 720p (about $0.190 per second). Because RelayDance is OpenAI compatible, you pass resolution in the video task metadata. Live per-model rates are published at relaydance.com/models.

Which resolution setting fixes soft video

Set a higher resolution tier in metadata.resolution to remove softness at the source rather than upscaling afterward. Seedance 2.0 720p runs at about $0.190 per second, 1080p at about $0.470 per second, and native 4K at about $4.90 per 5 second clip. Seedance Fast is priced at about $0.152 per second for lower cost drafts. Submit the change with POST /v1/video/generations, adjusting resolution alongside model, prompt, and seconds. Higher tiers render more detail per frame, which addresses perceived blur directly. According to the source documentation, 「将 base_url 改为 https://relaydance.com/v1 并保留 OpenAI SDK 即可调用」(据 relaydance.com/docs 官方文档), so no client rewrite is needed to change tiers. See relaydance.com/models.

Native 4K versus 1080p: cost and detail

Choose native 4K when 1080p still looks soft, and accept its per clip pricing model. The table below compares the documented tiers so you can match detail to budget.

Native 4K is billed per 5 second clip, not per second, so plan clip length accordingly. Clips can run up to 15 seconds per request.

Steps to re-render at higher quality

Re-render at a higher resolution by resubmitting the same prompt with an updated metadata value, following these steps:

1. Point your OpenAI SDK base_url to https://relaydance.com/v1 and set Authorization: Bearer YOUR_API_KEY.
2. Call POST /v1/video/generations with model, prompt, seconds, and metadata.resolution set to 1080p or 4K.
3. Add up to 9 reference images, 3 reference videos, and 3 audio tracks in metadata.content[], cited as @image1 to @imageN, for guidance.
4. Poll GET /v1/video/generations/{task_id} until status is succeeded or failed, then read the video url.

You can also set metadata.callback_url to receive the final state by webhook instead of polling.

Cost control when testing quality tiers

Test resolution tiers without paying for broken renders, because failed requests carry no charge. According to the source documentation, 「失败或报错的请求一律不计费」(据 relaydance.com 官方文档), so retrying a soft output at 1080p or 4K does not risk billing on errored jobs. Billing is pay-as-you-go per generated video, and payment is accepted via USDT and Stripe card. For image tasks, gpt-image-2 charges only for input (image output is free), image to image starts at about ¥0.035, and 4K output is priced the same as 1K output. Create keys at the console before testing, and confirm live rates at relaydance.com/models.

FAQ

Which resolution values remove blur? Set metadata.resolution to 1080p (about $0.470 per second) or native 4K (about $4.90 per 5 second clip) instead of 720p (about $0.190 per second).

Do failed high-resolution attempts cost money? No. Per the docs, 「失败或报错的请求一律不计费」(据 relaydance.com 官方文档).

Do I need a new SDK to change resolution? No. Keep the OpenAI SDK and point base_url to https://relaydance.com/v1, then pass resolution in metadata. See relaydance.com/docs.