Wipe Me Down Dance AI Troubleshooting: Fix Bad Renders

When a render goes wrong, the fastest fix is usually in the input, not the model. Wipe Me Down Dance AI runs on Seedance 2.0, the same model behind every render, and it turns a photo or text description into a short MP4. When the output warps a face, changes identity, moves stiffly, or fails entirely, the issue almost always traces back to one of three things: the photo, the prompt, or the render settings. This guide sorts the common symptoms into causes and fixes so you can retry without spending extra credits.

Open the studio before you start, and keep the pricing page open if you need to check credit costs.

Symptom to cause quick reference

| Symptom | Most likely cause | First fix | | --------------------------------------- | ------------------------------------------------ | ---------------------------------------------------------- | | Warped face, melted eyes, extra fingers | Blurry photo, group shot, or subject turned away | Upload a sharp front facing photo of one subject | | Subject looks like a different person | Low detail reference or multiple people in frame | Crop to one person, improve lighting, re-upload | | Arms locked, no dance, stiff body | Prompt asks for too many movements at once | Simplify to one clear action, set length to 5 or 8 seconds | | Render fails and credits return | Blocked input or provider error | Retry once, then shorten clip and simplify scene | | Wrong frame or cut off subject | Aspect ratio does not match the platform | Choose 9:16 for TikTok, Reels, and Shorts |

This table is a shortcut. Each symptom gets a longer explanation below.

Warped face or melted features

The most common bad render is a face that warps, melts, or changes between frames. The cause is almost always the reference photo. The model needs a sharp, well lit image of one person or animal, taken from the front or a three quarter angle, with the face and body clearly visible. Blurry photos, group shots, and pictures where the subject is turned away make the limbs drift and the face change between frames.

When the model cannot hold the face, it fills the gap with guesses. Those guesses show up as warped eyes, extra fingers, or a jaw that slides around. The fix is simple: upload a better photo. JPG, PNG, and WebP files up to 10 MB are accepted. Waist up or whole body framing works best because the model needs room to move the arms and legs. If the only photo you have is blurry or includes multiple people, crop it to a single subject first.

If you have no usable photo at all, switch to text to video mode. Leave the photo field empty and describe the scene instead: who dances, where they stand, and how the camera moves. For more on getting a clean photo input, see wipe me down dance from a selfie.

Wrong identity or subject drift

A render can start with your subject and then drift into someone else. This happens when the model does not have enough consistent detail to hold onto. Group photos are the biggest trigger, because the model tries to average several faces into one dancer. The result is a subject that looks like a blend instead of the person you uploaded. Blurry reference images cause the same drift, because the model cannot lock onto a clear face.

The fix is to feed one subject per render. Use a photo where the face, hair, body shape, and clothing are visible. Those details are carried into the render, keeping the dancing subject recognisably the one you uploaded. For pets, the same rule applies: one dog or cat per photo. Two animals in one frame tend to blend into a single creature.

Do not use a group photo and hope the model picks the right person. It will not. Crop, retake, or find a single subject image. If the identity still drifts, reduce the render length to 5 seconds. Shorter clips give the model fewer frames to invent, so the face stays closer to the source.

Stiff motion or missing dance

If the subject stands still, moves only the arms, or never performs the Wipe Me Down step, the problem is usually the requested action. The default look is a full prompt you can read, copy, and edit before you spend a credit. When you add extra instructions to that prompt, you can overload the motion. Seedance 2.0 generates one continuous vertical clip from a single prompt, and it handles one clear movement per shot far better than a whole routine.

To fix stiff motion, shorten the clip and simplify the action. Pick a length of 5 or 8 seconds instead of 12. Remove any extra actions, camera moves, or background changes. One clear movement, like the bounce or the hand wipe, holds up better than three moves stacked together.

Also check your aspect ratio. If the output is 16:9 or 1:1 and you want a vertical clip, the dance may look cropped or stiff because the model is framing for a different shape. Set 9:16 for TikTok, Reels, and Shorts. The studio recalculates the credit badge before you press Generate, so you can see the cost change when you adjust resolution, length, or format.

For a longer explanation of how the Seedance 2.0 model handles dance motion, read Seedance 2 dance motion explained.

Failed render and credit refunds

A failed render is not a charge. The credits return automatically to the balance they came from, in full. A render that was never produced is not charged, and retrying does not double charge. If you see a failed message in the studio, you do not need to contact support to get the credits back. They are already back.

If renders keep failing, the fastest fix is to change the input. Shorten the clip or simplify the action. One clear movement per shot holds up far better than a whole routine. A 1080p render can also take longer than 720p, and heavier renders may be more likely to time out. Drop to 720p or 480p and retry.

Some failed renders are caused by a blocked input. Before every render the prompt and the photo are screened by a moderation provider, and a blocked input is refused without spending credits. If your photo or prompt is refused, do not repeat the same input. Change the wording or use a different photo.

You can also check your credit balance in the studio. The badge next to Generate reads the exact cost for the current settings. The default spec, Standard · 720p · 12s, is 26 credits. 480p is 12 credits and 1080p is 64. Longer clips scale linearly from the twelve second table and are rounded up to a whole credit. The figure updates the moment you change a setting.

Checklist before you retry

Run through this list before spending another credit.

1. Photo quality: sharp, well lit, one person or animal, front or three quarter angle, face and body visible.
2. File format: JPG, PNG, or WebP up to 10 MB.
3. No group shots: one subject only. Two pets in one frame will blend.
4. If no clear photo: switch to text to video and describe the scene.
5. Length: set 5 or 8 seconds for a cleaner result.
6. Action: keep one clear movement in the prompt.
7. Aspect ratio: choose 9:16 for vertical platforms, 16:9 for landscape, 1:1 for square.
8. Credits: confirm the render cost on the Generate button before you start.
9. After a failed render: retry once, then change the input if it fails again.

These steps cover most bad outputs without needing to change the model or your plan.

What not to do

Do not upload the same blurry photo and expect a different result. Do not feed a group shot and hope the model picks one face. Do not stack four actions into one prompt and wonder why the motion is stiff. Do not buy more credits to fix a bad render before changing the input. The problem is almost never the credit balance.

Also, do not present an AI render as a real recording of a real person or as the original trend video. The model generates a new performance from your inputs. It does not copy anybody's recorded footage or choreography.

FAQ

Why does my face look warped in the dance video?

A warped face comes from a weak reference photo. Use a sharp, well lit image of one subject, taken from the front or a three quarter angle, with the face and body clearly visible. Blurry photos, group shots, and pictures where the subject is turned away cause the face to change between frames. Re-upload with a clear single subject photo, or switch to text to video if you have no usable photo.

Why did the render fail and take my credits?

A failed render does not take your credits. The credits return automatically to the balance they came from, in full. A render that was never produced is not charged. If you see a failed message, the credits are already back. If renders keep failing, shorten the clip or simplify the action, and retry.

Can I get a refund for a bad render?

A bad render that was produced is not automatically refunded, because credits are spent when the render is generated. A failed render that never produced a video returns its credits automatically. For payment refunds on unused credit purchases, check the refund policy linked from the pricing page. The studio does not charge for a render that was never produced.

How do I stop the model from changing my identity?

Use a single subject photo with sharp detail. Group photos and blurry references cause the model to drift into an average face. Make sure the face, hair, body shape, and clothing are visible. If the drift continues, reduce the clip to 5 seconds and avoid adding extra prompt instructions that fight the identity.

Does a shorter clip always render better?

Not always, but shorter clips reduce the number of frames where the model can invent errors. If you see warping or stiff motion, drop from 12 seconds to 8 or 5 seconds and keep one clear movement. The cost scales down with length, so you also spend fewer credits on the retry.