How to put yourself into any viral video with one prompt

The same video, with you in it

The trick is to edit the video instead of generating a new one. You give the AI a real clip and tell it to replace only the people, so the moves, the timing, the camera and the room all stay as they were. That's why it looks filmed. I made mine with the video edit mode of Seedance 2.5.

When I tried to describe the movement in words instead, I looked like a creep staring into the camera and somehow had three feet. Don't describe a video the AI can already see.

How to do it from zero

1. Pick your clip. Choose a short video you have the right to use, or film your own version of a trend. Trim it to the part you want before uploading, because every extra second costs credits.
2. Choose your photos. Use a sharp photo of your face, in daylight if the clip is in daylight. Your outfit can come from the prompt. For a second person, use someone who has said yes, or invent someone with the second prompt below.
3. Open the tool. Go to Dreamina and sign in. In the left menu, click Video recreator, then stay on the Object swap tab. It says it can "replace people, objects, or backgrounds while keeping the rest unchanged".
4. Add your files. Put your clip under Add reference video. Put your photos under Add products, characters, or clothes to swap in, with your photo first so it's image 1.
5. Paste the prompt. Fill in the brackets in the prompt below and paste it into Additional instructions. If the @ tags don't link up there, write "image 1" and "image 2" instead.
6. Run one test first. Pick Dreamina Recreate at 720P and press Generate. Clips up to 15 seconds can use Dreamina Recreate Lite, which uses fewer credits. Check the result before you run anything else.
7. Finish it. Upscale the take you like, then put your own text and sound on top in CapCut. Mine came out a little too warm, so I took the warmth down slightly at the end.

The swap prompt

This is the prompt behind my video, with brackets where yours will be different. The words in the brackets are what I used, so swap them for your own. Filming with one person? Delete the second person's paragraph and change "Exactly two people" to "Exactly one person".

Edit @Video1. Keep every movement, gesture, head turn, step, position and all timing exactly as in the video, frame for frame, with the same camera and the same room. Only the people change. Keep any on-screen text exactly as it is.

The [man who sits on the table] becomes a [young woman] whose face and hair match @Image1 100%: same face, [long dark wavy hair]. She does exactly what he does, in the same places and at the same moments: [she sits on the small table with her back to the camera facing the wall, sips, then climbs down and sits on the bench]. Her body is [your real build, for example a slim, small-framed woman's: narrow sloping shoulders, small waist, slender arms and legs, small hands and feet]. Her eyes look where his eyes look, [off to the side and at the coffee], never into the camera. [Relaxed, natural, slightly absent-minded] expression. She wears [a fitted cream knit top, slim high-waisted black trousers and black ballet flats], with exactly two legs and two feet.

The [second person] becomes the person in @Image2 100%: same face, [natural curly afro], [rust wool blazer, cream wide trousers, black boots]. They do exactly what the [woman] in the video does, with [a plain black bag with no logo].

Photoreal, real skin texture, natural fabric movement, same light as the video, real-time speed. Exactly two people. No new text, no logos, no watermark.

Invent someone who doesn't look like AI

My first made-up woman looked too much like AI. It worked once I asked for an ordinary person caught on a friend's phone. Generate this in any image tool, such as ChatGPT, and use the result as image 2.

Candid phone photo of a [woman, 27], invented person, standing in [a concrete cafe], caught mid-moment, not posing for a portrait. [Slim build]. [Natural curly afro] with flyaways and an uneven shape. Real skin: visible pores, slight shine on the nose and forehead, small marks, uneven tone under the eyes, slight facial asymmetry, very little makeup.

[Relaxed rust wool blazer with real creases and cream wide-leg trousers, black ankle boots, holding a takeaway coffee]. Full-length, slightly off-centre framing, plain daylight from a window, soft phone-camera grain, slightly imperfect exposure.

Looks like an ordinary real person snapped by a friend, not a model shot. No text, no logos.

The six things that went wrong for me

Make it yours, not a copy

A swap on its own is still someone else's video. Change the outfits, the people and the on-screen text so it feels like your own version, and credit the original creator when you're using their clip.

Credits

Dreamina has free credits to start with, then runs on paid credits. Prices and deals change often, so check the amount beside Generate before each run. One 720P test before anything else is the cheapest way to find a problem.

The links

🎬 Dreamina
📖 How to use Seedance 2.5
✂️ CapCut