Use one recent photo of one adult, face front-on and clearly visible, lit by soft daylight, framed chest-up or waist-up, no sunglasses and no heavy filter, at least about 768px on the short side. That one choice decides almost everything. The lips will never match the song exactly — the model cannot hear the track — so expect a soft singing performance, not frame-accurate lip-sync.
What photo should I use for the Raindance AI trend?
The model learns your face from one still, then rebuilds it from a different angle, in different light, with a different expression. The easier you make that, the more the clip looks like you.
The ideal source photo is boring: you, facing the camera, in daylight, neutral or lightly smiling, nothing covering your face. A plain selfie by a window beats a holiday photo shot into the sun every time — so shoot one on purpose instead of digging through your camera roll. Face a window, phone at eye level, shoulders in frame, take three.
Photo checklist: do and don't
| Do | Don't |
|---|---|
| One adult, alone in the frame | Group photos or a crop of a group photo |
| Face straight on, eyes toward the lens | Profile or 45-degree angles |
| Soft, even daylight — a window or shade | Hard midday sun, harsh flash, backlight, dark rooms |
| Chest-up or waist-up framing | Extreme close-ups, or full-body where the face is tiny |
| Neutral or gentle expression | Wide grins, pulled faces, mid-laugh |
| Nothing covering the face | Sunglasses, hats pulled low, masks, hands, drinks |
| Natural colour | Heavy beauty filters, skin smoothing, strong presets |
| Sharp, in focus | Motion blur, noisy night shots, screenshots of screenshots |
| At least ~768px on the short side | Tiny saved thumbnails, compressed re-shares |
| JPG, PNG or WebP under 10 MB | Screen-recorded frames, HEIC that your browser cannot read |
| Simple background | Busy backgrounds with other faces |
A duet doubles this list: one photo per person, and the weaker photo sets the ceiling. The duet guide covers matching two photos.
The Raindance AI video doesn't look like me — what went wrong?
Almost always one of these, in order of frequency.
- Resolution. An image re-shared through a chat app a few times has lost most of its facial detail, and the model fills the gaps with a generic face.
- Angle. From a side-on photo the model guesses at half your face, and it guesses toward an average.
- Filters. A beauty filter changed your face before the model saw it, so the output reproduces the filtered version.
- Light. Deep shadow on one side, or a flash that blew out your skin, removes the structure the model relies on — cheekbones, jawline, eye sockets.
- Sunglasses or a hat. Eyes carry most of the recognition. Cover them and the face becomes a stranger's.
- A group photo. The model may blend two people or pick the wrong one.
- An old photo. It can look like you years ago and not like you now, which people read as "wrong" without knowing why.
The fix is the same every time: change the photo, not the settings. One clean front-facing picture in daylight beats any scene change or model upgrade.
Why don't the lips match the song in my Raindance video?
Because the generator never hears the song. It makes a silent clip of someone singing gently into the lens; the audio is attached later, by you, in TikTok or Reels. Nothing links the mouth to a track the model was never given.
The same is true of the other AI tools people use for this trend. Real lip-sync means feeding audio into the model, which is a different kind of tool. What the trend needs is a convincing performance — soft, unexaggerated singing that reads as "this person is singing along". Trimmed to the "I love you" bridge at eight seconds, nobody counts syllables.
Three things make the mismatch invisible:
- Keep it short. Eight seconds is enough. A longer clip gives viewers more chances to notice the mouth.
- Start the sound on the cut. Trim the audio so the vocal arrives as the clip hard-cuts to the pier; the brain forgives a lot when the entrance lands right. The sound guide has the steps.
- Pick the gentler take. Of two renders, use the one with less mouth movement. Wide-open-mouth singing invites comparison; a soft delivery does not.
How do I improve the performance itself?
The source photo shapes the performance more than people expect.
- A closed or barely-open mouth gives a controlled singing motion. A wide grin tends to produce exaggerated mouth movement.
- Eyes toward the lens produce the direct-to-camera look the trend is built on. Eyes off to the side read as distracted.
- Chest-up framing gives the model room for the slow push-in. A tight crop leaves the camera nowhere to move.
- Hair away from the face. The model animates hair as well as skin, and a face half behind hair is where artefacts appear.
- Scene choice changes mood, not accuracy. Golden-hour pier is the original look; rain on the pier is cooler and hides a little more. All four are on the generator.
When should I retry, and will it cost me again?
| What you see | Do this |
|---|---|
| Face is clearly someone else | New photo: front-facing, daylight, sharp. Do not re-roll the same photo. |
| Looks like you but a bit off | Re-roll once with the same photo; small variation between renders is normal. |
| Mouth is too exaggerated | Re-roll with a neutral-mouth photo, or keep the clip shorter. |
| Hands or background are strange | Re-roll; these vary between takes. Keeping the clip to 8 seconds helps. |
| Hair or glasses frames glitching | New photo with hair off the face and no glasses. |
| Render failed or came back empty | Nothing to do — failed renders refund the credit automatically. |
The honest economics: re-rolling costs a credit, so get the photo right first. Start with the free teaser — one 5-second 480p watermarked clip per account — to check the face reads as you before spending anything. Then unlock that teaser into the full HD clip for $3.99, or buy clips from $4.99 with cheaper bundles on pricing. Credits never expire.
One note on expectations: the clip opens on a recreation of your photo, then hard-cuts to you on the pier. Judge it on the pier section — that is the part you will post.
Does any of this change for pets or duets?
Pets. A cat or dog needs the same things — front-facing, eyes visible, good light, sharp — but mouth movement matters less, because nobody expects a dog to lip-sync. Pet mode takes one animal per photo and is the most forgiving version of the trend.
Duets. Two photos, one person each, both clearing the checklist. Mismatched light between the two is the most common duet failure — more common than either photo being bad alone.
The rules on people still apply: adults only, your own face or that of someone who agreed, no celebrities, no minors. Uploads are used only to make your video and originals are removed after 30 days. And switch on TikTok's AI-generated content label before posting.