GuidesSeedance Generation Failed? Causes and Fixes
Makernice Team ·
Prices as of November 10, 2026 (price list v12).
A Seedance video usually fails for one of four reasons: the safety checks of ByteDance's Seedance models rejected a real human face, protected characters or sensitive content; an input image was outside 300–6000 px per side or not a still JPEG, PNG or WebP; the run passed its time limit; or the provider was unavailable. This applies to Seedance 2.5, Seedance 2.0 and Seedance 2.0 Fast.
Why does Seedance say generation failed?
Most Seedance failures are content checks, not bugs. BytePlus, ByteDance's cloud platform, says requests can be blocked by two layers: an input content filter and the model's own output moderation. Some platform-level safety policies can't be turned off (BytePlus help). Apps such as CapCut, Dreamina and Higgsfield show these failures in their own words; this guide covers the model side and what you see when you call Seedance on Makernice through the API or the web tool. The checks travel with the model, so the same prompt or image can be rejected wherever you use it. On Makernice, every failure has an error.code, and content rejections add an error.reason:
| What you see | Cause | What to change |
|---|---|---|
| 422 content_rejected, reason prompt | The prompt failed moderation at create | Rewrite the flagged wording |
| content_rejected, reason input_image | A reference image failed moderation | Use a different image |
| content_rejected, reason model | Seedance's safety policy refused it | Remove real faces, brands and sensitive wording |
| invalid_request, image side under 300 or over 6000 px | Seedance rejects images outside 300–6000 px per side | Crop or resize to 300–6000 px per side, get a new quote and create a new generation |
| invalid_request with error.input | Image unreachable, redirected, animated or over 10 MB | Use a direct https still JPEG, PNG or WebP |
| timeout | The run passed 60 minutes | Get a new quote and create a new generation |
| upstream_unavailable or generation_failed | The provider was unavailable or the run failed | Later, get a new quote and create a new generation |
Why does Seedance reject a real person or say face detected?
Seedance refuses input images that may show a real person. ByteDance's Volcano Engine Ark documents a 400 error whose message says the input image may contain a real person and asks you to replace it (Ark error codes). ByteDance's CapCut said in April 2026 that during the initial rollout, Dreamina Seedance 2.0 restricts making videos from images or videos that contain real faces (CapCut newsroom). Messages such as "face detected in your media" are the app's wording; they usually mean this input check.
The same check applies on Makernice; calling Seedance through the API doesn't turn it off. A rejected image returns content_rejected with error.reason input_image (our moderation) or model (the provider's safety policy). Our acceptable use policy also limits uploads: only images you own or may use, showing you or an adult who has clearly agreed, never a minor or a public figure. Don't try to get a real face past the check; use a designed or AI-generated character instead.
What does protected content is not allowed mean in Seedance?
"Protected content is not allowed" is also app-side wording, and it usually points to the copyright check. ByteDance's Ark error codes include rejections where the input text, image, video or audio, or the generated video, may be related to copyright restrictions (Ark error codes), and CapCut says it blocks unauthorized generation of intellectual property. Remove named characters, franchises, logos and brands from the prompt and the reference images, and describe an original character. "Sensitive content" messages match the broader sensitive-content checks on input and output in the same list: rewrite violent, sexual or other policy-violating wording.

Why was my Seedance input image rejected?
In image modes (image-to-video, first and last frame, reference-to-video), Seedance 2.5, Seedance 2.0 and Seedance 2.0 Fast need each image to be 300 to 6000 px per side. An image outside that range is rejected: the generation fails with error.code invalid_request and the hold is released. Crop or resize it to fit, then get a new quote and create a new generation. Each image must also be a public https URL that serves a still JPEG, PNG or WebP of at most 10 MB directly; redirects aren't followed, and animated WebP and GIF aren't supported. If a file can't be used, error.input names the image: image_url, last_image_url or reference_images[0] (zero-based). For how first frames and reference images are used, see image-to-video and first and last frame.
Why does my Seedance video have no sound?
On Makernice, Seedance 2.5, Seedance 2.0 and Seedance 2.0 Fast always generate video with audio, and there is no switch to turn it on or off: Seedance doesn't take the generate_audio option. A silent clip isn't a Seedance setting you can change here. If you need a silent video, remove the audio track after you download it. See the audio row on the Seedance 2.5 and Seedance 2.0 Fast model pages.
Why is Seedance stuck at 99% or still generating?
Makernice reports a status, not a percentage. A video moves from queued to generating, then to moderating (generation is done and the amount is charged; the outputs are in a safety check, so keep polling) and succeeded. A video can take a few minutes, longer for long or high-resolution clips. If it runs past 60 minutes, it ends as failed with error.code timeout. If your own client times out while waiting, don't create a second generation: poll GET /v1/generations/{id}, or repeat the create request with the same Idempotency-Key, quote and input to get the original generation back. The statuses are listed under Check the status.
Are failed Seedance generations charged on Makernice?
No. Makernice doesn't charge for a failed Seedance generation: the hold is released in full. Note: if a finished output is blocked by output moderation for a policy violation, the run is still charged. You can contact support for a manual review. A prompt blocked when you create the generation (422 content_rejected, reason prompt) creates no generation and holds nothing. In the Makernice web tool, a failed generation shows the cause in plain words: the prompt, a reference image, the model provider's safety policy or a timeout. Every case is in Are failed generations charged?, with billing examples.
Where do these Seedance facts come from?
| Fact | Value | Source |
|---|---|---|
| Input image size (Seedance 2.5, 2.0, 2.0 Fast) | 300–6000 px per side | Makernice model specs |
| Input image file | Still JPEG, PNG or WebP, up to 10 MB, no redirects | Docs: Quote |
| content_rejected reasons | prompt, input_image, model | Docs: Status |
| Image that can't be used | Fails with invalid_request, hold released | Docs: Quote |
| Maximum video runtime | 60 minutes, then error.code timeout | Makernice service setting |
| Audio on Seedance models | Always on, no switch | Models |
| Real person in an input image | Rejected: input image may contain a real person | Ark error codes |
| Copyright | Input or output may involve copyright restrictions | Ark error codes |
| Two safety layers | Input filter plus output moderation | BytePlus help |
| Failed generation billing | Released; output blocked by moderation is charged | Docs: failed generations |
Checked on 2026-11-09. Next, check the limits on the Seedance 2.5 and Seedance 2.0 model pages, read Are failed generations charged? in the docs, and if you also use Kling, see why a Kling video fails.
Frequently asked questions
Will Seedance accept my video if I use it on Makernice instead of an app?
What is the difference between reason input_image and reason model in a Seedance content_rejected error?
How do I retry a failed Seedance generation through the Makernice API?
Why did my Seedance prompt fail before the video started?
Models in this post
VideoSeedance 2.5
ByteDance's flagship for long takes: up to 30 seconds, 480p to 4K, and up to 16 reference images to keep every shot on-model. Audio is always on (no toggle).
VideoSeedance 2.0
The all-rounder for ads and social: 4–15 s clips at 720p or 1080p from text, an image, first and last frames, or 9 references. Audio is always on (no toggle).
VideoSeedance 2.0 Fast
-15%Seedance 2.0 at one 720p tier and a lower per-second price: same modes, same 4–15 seconds. Audio is always on (no toggle). Draft here, finish on 2.0 or 2.5.