The Training Season AI trend is a 15-second clip in which one selfie becomes a singer at a vintage microphone on the Royal Albert Hall stage, set to Dua Lipa's Training Season. You make it on videosi by uploading one clear photo of yourself, choosing a format and pressing Generate; there is no prompt to write and the song is already on the video.
This guide goes past the basics and answers the questions people ask once they have seen the clip: what happens in each shot, why close-ups are hard, how to pick the selfie, whether the mouth syncs, and what to do about labels and music when you post.
Key takeaways
- The clip has six shots over 15 seconds, opening and ending on a singer at the microphone, with a pull-back to the whole hall in the middle.
- A sharp, front-facing selfie with your whole face lit gives the best likeness; likeness is not guaranteed, so try a second photo if the first drifts.
- The mouth does not sync to the lyrics. It is a performance mood piece with the original sound laid over it.
- A standard clip (Standard model, 720p, 15 s) costs 13 credits, there is no free generation, and a failed render returns its credits.
- Post it with an AI label and treat the song as someone else's music.
In this guide
- What does the clip show, shot by shot?
- Why is a selfie close-up hard for AI?
- How do I choose the right selfie?
- Is the singer's mouth synced to the song?
- How is this different from a lip-sync filter?
- Can I use Dua Lipa's song when I post?
- Do I need to label it as AI?
- How does it compare with Stage Singing AI and Tiny Desk AI?
- Is the trend still worth posting?
- What does it cost, and what if it fails?
What does the clip show, shot by shot?
The structure is what makes the trend recognisable, so it helps to know it. There are six shots in 15 seconds:
- 0 to 6 s: a tight close-up of your face beside a vintage microphone, with slow head tilts and eyes closing on the long notes.
- 6 to 8 s: a red backlit silhouette of you holding the mic stand while the camera pulls away.
- 8 to 9 s: a high shot of the hall and the circular stage, with the orchestra around it and the audience in red light.
- 9 to 11 s: a wide shot of you alone on stage with your arms open.
- 11 to 13 s: a second, quieter close-up of your face.
- 13 to 15 s: a full-body shot at the mic stand, one hand on the stand as the clip ends.
Nearly half the runtime is close-up, which is why the choice of photo matters so much. If you want to see the structure before you commit, the Training Season AI page has a sample and two examples.
Why is a selfie close-up hard for AI?
A close-up fills the frame with a face, and viewers know faces better than they know almost anything else. A small drift in jaw width, eye shape or age is obvious at that size, where it would go unnoticed in a wide shot. The two close-ups here, at the start and again at 11 seconds, are the hardest shots in the clip.
There is a second problem. The motion comes from a reference clip of another singer, and the model is pulled toward that singer's face shape. Without a strong identity lock, your selfie can come out slimmer or older than you, or the reference singer's cap and jacket can leak into your result. videosi handles this by treating your photo as the only source for face, hair, build and clothes, and by telling the model never to reshape a face toward the reference singer. It reduces the problem; it does not remove it, so likeness is not guaranteed.
How do I choose the right selfie?
Pick the photo as if you were choosing a passport picture that you also like. These rules come straight from what goes wrong most often:
- Face the camera. The video opens on a tight close-up, so a straight-on selfie gives the best likeness.
- Use even light. Daylight on your whole face works best; hard shadows, strong filters and heavy retouching all drift in the close-ups.
- Leave off sunglasses and hats. The close-up needs your eyes, and a cap or glasses in the photo are copied into every shot.
- Crop to one person. A group photo can put several faces on one singer.
Your hair and clothes in the photo are the ones you sing in. If your photo shows only your face, the singer is dressed in a plain dark stage outfit, so a photo that shows your shoulders gives you more control over what you wear on stage. Only use photos of adults who agree to it, such as yourself or a friend.
Is the singer's mouth synced to the song?
No. The singer's mouth does not sync to the lyrics. The trend is a performance mood piece: head tilts, held breaths, eyes closing on the long notes, and then the wide shots. The original sound is laid over the finished video, lined up with the shots rather than with individual words.
That matters for how you post it. Do not caption it as if you are singing the lyrics word for word, and do not expect a viewer to read lips. Describe it as what it is, an AI performance clip.
How is this different from a lip-sync filter?
A lip-sync filter is built around the mouth: it takes a face and an audio track and moves the lips to match the words. The Training Season clip is built around the whole scene instead. It gives you a hall, an orchestra, six camera shots and a pull-back from close-up to wide, and your face is placed into that scene.
So the two answer different wishes. If you want a face to speak or mouth a specific line, a lip-sync tool is the right kind of tool. If you want to look like the star of a concert film, with the camera work and the red light, this is closer. The singing here is performed, not spoken, and it is not meant to be read word by word.
Can I use Dua Lipa's song when I post?
The sound used in the trend is Dua Lipa's Training Season, the live orchestral version recorded at the Royal Albert Hall. The videosi video comes with that original sound already on it. The song belongs to its owners; videosi does not claim it, and the AI part of the clip is the picture only.
The practical advice is to use the music the way your platform expects. If you post on TikTok or CapCut-linked apps, you can swap the sound inside the app and choose the track from its own library, which is the cleaner route if you are worried about rights. A copy of the song you downloaded and uploaded yourself carries the same copyright risk as any other upload. Crediting the artist in a caption is polite but is not a licence, and videosi is independent and not affiliated with TikTok, CapCut or Dua Lipa. If you plan to use the clip commercially, get proper permission for the music rather than relying on a caption.
Do I need to label it as AI?
You should. The clip looks like real concert footage, and a real hall with a real orchestra is exactly the kind of scene people assume is genuine. A short note in the caption, plus the platform's own AI label where it offers one, removes any doubt and protects you if the clip travels beyond your followers.
Keep the caption plain, for example "me singing at the Royal Albert Hall (AI)". It tells viewers the joke before the first beat. Also keep to the rule about people: use your own photo or one from an adult who agrees. videosi turns one adult's photo into the singer; it does not put two real people together or make anyone say words they never said.
How does it compare with Stage Singing AI and Tiny Desk AI?
All three turn a photo into a short performance clip, and they differ in the room. The Training Season clip is a grand, red-lit concert hall with an orchestra, a vintage microphone and a camera that pulls back to show the whole room. It is the most cinematic of the three and the most dependent on the face close-up.
The Training Season page is the one to pick for the Albert Hall look. The Stage Singing AI effect is the alternative if you prefer a concert stage with LED screens and a short live singing clip. The Tiny Desk AI page goes the other way: a cosy office where you and a band perform at a desk. Choose by mood: epic hall, bright stage show or small room.
Is the trend still worth posting?
It is early. Our trend radar marked it as spreading on 2026-11-06. The earliest AI version we found on TikTok is dated 2026-10-09, and the biggest clips we found are dated 2026-11-02 and 2026-11-03: a template clip with about 700,000 views and one with about 250,000 that shows an actress singing the song. No single original creator is confirmed, so we do not credit anyone for it.
It spread through CapCut templates and in-app AI filters, and templates change or vanish often, so check inside the app for what is available now. Because the biggest clips are only days old, a clip posted soon has a better chance than one posted in a month, and it may fade within weeks.
What does it cost, and what if it fails?
It is not free: there is no free generation. A standard 15-second video (Standard model, 720p) is 13 credits, and better models and higher resolutions cost more. Credit packs start at $6.90 for 10 credits and never expire. The MP4 has no watermark, and a failed, timed-out or blocked render returns its credits.
Rendering takes several minutes. If you close the page, the clip is saved in My creations, so you can come back for it. To start, open the Training Season AI page, upload one selfie and choose 9:16 for TikTok, Reels and Shorts, or 1:1 or 16:9 for other places.
