Photo tips for a better Jawn dance video

Better input, better output — that is the whole rule. The photo you upload decides almost everything about the Jawn dance video you get back: which limbs the model has to invent, whether the face holds together, and whether the background stays stable behind the motion. The good news is that a good input photo is not hard to take, and the checklist below takes about five minutes to run through before you upload.

The five-minute checklist

Run these in order. If a photo fails two or more of them, pick a different one rather than trying to fix it in editing.

| Check | Target | Why it matters | | ------------------ | ----------------------------- | ---------------------------------------------------------- | | Face direction | Looking at the camera | A turned-away face gives the model nothing to work with | | Number of subjects | Exactly one | Groups split the tracking and faces blend | | Body visibility | Visible to the knees or lower | Hidden legs mean invented legs | | Lighting | Even, soft, front-on | Harsh shadows bake into the render | | Filters | None | Filters flatten detail the model needs | | Eyewear | None | Sunglasses hide the part of the face that sells the motion | | Pose | Facing forward, not sideways | Profile poses ruin the read of the bounce | | File | JPG, PNG or WebP, up to 10 MB | The only formats the uploader accepts |

Lighting: soft and front-on

The single most common reason a render looks flat is backlighting. If there is a bright window behind the subject, the camera exposes for the bright background and the face becomes a soft grey blob. The model then animates a soft grey blob, and no amount of retrying fixes it.

What works:

1. Stand the subject facing a window, not in front of one. Daylight on the face is the cheapest good light there is.
2. Avoid a bare bulb directly overhead — it drops a hard shadow under the brow and nose.
3. Avoid mixed light sources: warm indoor light on one side and cool daylight on the other produces a colour split the model can read as two faces.
4. Overcast outdoor light is ideal. It is soft, even and directionless.

Framing: one subject, head to knees

Frame for the model, not for the composition. The routine moves the whole body, so the body has to be in the picture.

• Put the subject in the centre of the frame with a little headroom above.
• Keep room below the knees, or the whole body if you can. Feet can be slightly cropped; legs cannot vanish entirely.
• Shoot vertical if you can. A 9:16 source fits a 9:16 render without a crop.
• Keep the background plain. A clear wall, a doorway, a plain room. Busy backgrounds are where model drift shows up first.
• Do not include a second person, even at the edge of the frame. Background people are pure risk.

The four things that make a face look wrong

These four are responsible for most of the disappointing renders people post about. Each one removes information the model needs and forces it to guess.

1. Backlight. A bright window or open sky behind the subject flattens the face into silhouette. The model has no facial detail to animate, so it produces a smooth, mask-like result. Fix: turn the subject around so the light is on their face.
2. Heavy filter. Beauty filters, smoothing filters and heavy colour grades bake into the pixels. Smoothing in particular erases the micro-detail that makes a face read as a face, and the render comes out waxy. Fix: use the original, unfiltered photo.
3. Sunglasses. Eyes carry most of the emotional read of a face, and a bouncing clip without visible eyes looks off no matter how good the motion is. Fix: glasses off, or clear lenses at most.
4. Sideways pose. A profile view gives the model almost no face to work with and hides half the body, so the animation tends to look like a shuffle from behind a wall. Fix: a front-on or very slight three-quarter angle.

If a photo has two of these, do not send it. There is almost always a better one in the camera roll, and one retry spent on a new photo is worth three retries spent on a bad one.

File formats and size limits

Short list, no negotiation:

• Accepted: JPG, PNG and WebP.
• Rejected: HEIC from an iPhone, TIFF, RAW, PDF, screenshots saved oddly, and anything over 10 MB.
• Resize if needed. A 10 MB ceiling is generous for a phone photo, but a full-resolution DSLR export can exceed it. Export a smaller copy rather than sending a huge one.

The most common upload failure is an iPhone photo in HEIC format. Change the camera setting to "Most Compatible" for JPEG, or export a JPG copy before uploading.

What to do if the first result disappoints

A failed render refunds your credits automatically, so a bad output costs you three minutes and nothing else. Work through this in order:

1. Look at the photo again. Nine times out of ten the fault is one of the four face mistakes above.
2. Replace the photo before you touch the style. Style changes polish a good input; they do not rescue a bad one.
3. Try a different frame from the same photo shoot — one step sideways, once more into the light.
4. Swap the style. The cute setting reads younger and softer, the classic setting is the cleanest read of the standard routine.
5. Re-upload at a smaller size if the first attempt stalled on the file itself.
6. Only after three genuine attempts with different photos should you conclude the subject does not suit the routine.

If the motion itself is the problem rather than the face, the framing is usually the cause — a subject photographed too close, with the lower body cropped out, forces the model to invent legs. Give it a full-body photo and try again.

Costs stay the same regardless of why you are retrying: a default 1080p · 12s · 9:16 render is 30 credits, a standard 720p · 12s video is 10 credits, and a Google sign-in grants 30 credits once — one free 1080p clip. Email accounts start at 0 and claim a check-in of 10, 20, 30, then 10 credits, once every 24 hours. The full walkthrough lives in how to make a Jawn dance video from one photo, and the same rules apply with extra weight when the subject is small — see AI baby dance videos from a single photo.

FAQ

What is the best photo for an AI dance video?

One subject, facing the camera, evenly lit, no filter, no sunglasses, and the body visible to the knees or lower. JPG, PNG or WebP, up to 10 MB. If it passes those, it is a good input.

Why did my render look waxy or smooth?

Almost always a heavy beauty filter or smoothing filter on the source photo, or a face lit from behind so it lost all its detail. Use an unfiltered photo with the light on the face.

Can I use a screenshot as my source photo?

You can if it is a clean, full-quality screenshot saved as PNG or JPG under 10 MB. But screenshots are usually compressed and often cropped too tight, so an original photo almost always renders better.

My iPhone photo will not upload. Why?

It is probably a HEIC file. The uploader accepts JPG, PNG and WebP. Set the camera to "Most Compatible" or export a JPG copy before you upload.

How many times should I retry before giving up?

Two or three times, and each retry should use a different photo, not the same one with a different style. If three genuinely different good photos all fail, the subject is probably not suited to this routine.

Does a failed render cost me credits?

No. Failed renders are refunded automatically, so you are never charged for a video you did not receive. The retries themselves are free apart from the time.