
Clip quality problems are usually specification problems. A clip that looks soft, cropped, or "off" in the feed is often technically valid and specificationally wrong — right content, wrong pixel grid.
Here are the specs that matter for this format, the trade-offs behind each one, and how to verify a file before it costs you an upload.
The three numbers
Resolution. The feed wants 9:16 vertical. The target is 1080 × 1920. What the studio actually gives you depends on the mode:
| Mode | Render resolution | Credits | Upscale to 1080 × 1920? |
|---|---|---|---|
| Text to video / image to video | 720p class | 300 (5s) | Acceptable |
| Motion control, standard | 720p | 500 | Acceptable |
| Motion control, pro | 1080p | 1,000 | Not needed |
Upscaling does not add detail — it interpolates. A 720p clip enlarged to 1080p has 2.25× the pixels and the same sharpness, which is why it can look worse than the original: the interpolation softens edges that were crisp.
Aspect ratio. 9:16. If the source is landscape, crop around the subject rather than letterboxing. A letterboxed clip loses a third of the screen and puts the caption over the subject.
Duration. Five seconds is the default for a reason: it is the cheapest setting, it contains one movement, and it is short enough that face drift does not build up. Ten seconds costs 600 credits without audio and gives the model twice the frames to stay consistent across. Most creators generate at five and cut two clips together if they need ten.
Why 720p is usually fine, and when it is not
A phone screen is small. A 720p vertical clip fills it without visible softness in normal scrolling. Three situations change that:
- Text overlays. Captions rendered over a 720p source are sharp while the video behind them is soft, and the mismatch is visible.
- Full-screen playback. Tablet, TV, or a desktop browser at 1440px wide — the upscale shows.
- Paid placement. A promoted clip is watched more attentively than a feed clip, and sharpness reads as production value.
If none of those apply, 720p is the economical answer. If one applies, either pay for pro motion control or accept the softness; there is no free sharpening step that recovers detail the render never produced.
The framing specification
Resolution is not the same as framing, and framing is what most clips get wrong:
- Subject centred, in the middle third horizontally.
- Head and feet in frame for a full-body dance.
- Bottom fifth clear of essential movement — that is where captions, the sound label, and the action rail sit.
- A little headroom above the subject; a clipped crown reads as a mistake.
None of that requires a better model. It requires a prompt clause ("full body in frame, vertical 9:16, subject centred") and a crop in the editor.
Verifying a file locally
Before uploading, check four things about the actual file — not the preview, not the thumbnail:
- Pixel dimensions — should be 1080 × 1920, or 720 × 1280 for a native 720p render.
- Duration — a five-second request sometimes returns 4.8 or 5.2 seconds. That is normal.
- Codec and container — MP4 with H.264 plays everywhere. A WebM or a QuickTime file may not upload cleanly to every platform.
- File size — a few megabytes is normal for five seconds. A 40 MB file for five seconds usually means a high-bitrate export that the platform will re-compress anyway.
The browser tools on this site do that check without a paid call:
- Video converter — change container or codec, and read duration and resolution.
- Video thumbnail maker — pull the frame that will become the cover, which is also the frame that decides whether anyone watches.
- Video to GIF — useful for a preview image in an email or a listing.
- Image compressor — for the source photo, before upload, if a job was refused for file size.
All four run in the browser. None of them touches a generation endpoint, so none of them costs credits.
The cover frame is part of the spec
The cover frame — the still the platform shows before playback — is chosen from the clip or supplied by you. In a feed it does the same job as the first frame of the video, and it is worth treating as a specification rather than an afterthought:
- Pick a frame where the movement is obvious. A mid-bounce pose reads as "dancing"; a standing pose reads as "photo".
- Pick a frame where the subject is centred.
- Do not let the platform auto-pick if the middle of the clip is the least interesting frame.
The thumbnail tool exists for this: scrub, grab the frame, check it at phone size.
A specification table to keep
| Setting | Recommended | Cost | Why |
|---|---|---|---|
| Mode | Text or image to video | 300 credits / 5s | Cheapest path to a usable clip |
| Duration | 5 seconds | 300 credits | One movement, least drift |
| Audio | Off, add music in the app | 0 extra | Doubles cost and fights the platform sound |
| Resolution | Native render, upscale in the feed | 0 | A second resample adds softness |
| Framing | Vertical 9:16, full body | 0 | Decides whether the joke lands |
| Cover | Mid-movement frame | 0 | Decides whether it is watched |
Every row in that table is a decision you can make before spending anything. The expensive decisions are the ones made after the clip exists — regenerating for a crop, regenerating for a duration, regenerating because the face drifted in the extra five seconds you paid for.
What specs do not fix
A well-specified clip of a dull movement is still a dull clip. Specs set the floor, not the ceiling. If the movement is right and the framing is right, the remaining variables are the prompt and the photo — the failure guide maps defects to inputs, and the prompt mistakes guide covers the wording.
And a technically perfect clip of a child whose photo you did not have permission to use is still not publishable. That question comes before all of the above.
Frequently asked questions
What resolution should a baby dance clip be?
1080 × 1920 vertical is the target for a feed upload. Text and image modes and motion-control standard render at 720p, which upscales acceptably on a phone screen but shows on a large display. Motion-control pro renders 1080p for twice the credits.
Is 720p good enough for TikTok?
For a phone screen, usually yes — the feed plays back at a modest size and re-encodes the file anyway. It stops being good enough when text overlays are added, when the clip is played full-screen on a tablet or TV, or when it is used in a paid placement where sharpness signals quality.
How long should the clip be?
Five seconds is the right default: it is the cheapest setting, it is long enough for one repeated movement, and it is short enough that identity drift does not accumulate. Ten seconds costs twice as much and gives the model twice the frames to wobble through.
How do I check a video file without uploading it anywhere?
Use a browser tool that reads the file locally. The tools on this site — video converter, thumbnail maker, video to GIF — read duration, resolution, and size in the page and never send the file to a paid generation endpoint. Check resolution and duration there, then upload the platform copy.
Why does the file size jump after export?
Because upscaling and re-encoding both add data. A 720p clip upscaled to 1080p is roughly 2.25× the pixels and is not actually sharper. Export at the native resolution and let the platform handle playback rather than resampling twice.
![50+ Best AI Baby Dance Video Prompts That Actually Work [2026 Guide]](/imgs/blog/baby-dance-ai-prompts.webp)
![AI Baby Dance Video Generator: How to Make Your Baby Dance with AI [2026]](/imgs/blog/how-to-make-baby-dance-ai-video.webp)







