Two selfies, a photo of your street and your meme images: a 30-second vlog in one single take where memes sip coffee at the terrace and cross at the light behind you. The full workflow is visible here without an account.
Real workflow, real results. Pan and zoom freely.
Use this workflowThe workflow is right here: the character sheet, the street photo, six meme cards to replace with your images, the two instructions and the Wan 3.0 video with live street sound. Pan, zoom, open the nodes.
At a glance
The format is everywhere on TikTok and Reels: a vlogger walks down his street filming himself on the front camera, while memes live their lives in the scenery behind him, sitting at the cafe terrace, buying a book, waiting at the pedestrian crossing. Nobody reacts, and that deadpan is what makes the video. Thirty seconds, one single take, no shooting, no editing, no extras.
What holds the format together is not the memes, it is the discipline of the shot: a multi-angle character sheet that locks your face and outfit from the first frame to the last (the consistent character guide explains why), a front camera that never flips, one meme per scene and a spoken line declared word for word. The template ships all these rules in its instructions: you only replace the images.
On the model side, Wan 3.0 is the only one in the catalog that holds thirty seconds in a single take with ten reference images, audio generated with the video and, above all, a real face accepted as a reference. Seedream 5 Pro builds the character sheet from two or three selfies, and the public calculator gives the exact cost before any signup.
Generate your character sheet
Two or three selfies into an Image node with Seedream 5 Pro: full body front, three-quarter, profile and back, plus face close-ups, neutral background, same outfit everywhere. That sheet is what holds your face across the thirty seconds.
Photograph your street
One single photo, with depth: wide sidewalk, shops on both sides, ideally a terrace or a shopfront. Two photos of the same spot would make the model invent a street that does not exist.
Import YOUR memes
One image per card, character large in the frame, no caption over it. Drawings and animals work great; photos of real people are often refused by providers, and never photos of children.
Adjust the opening line
The vlogger speaks only once, at the start, and the line is declared word for word in the instructions: that is what keeps the model from inventing dialogue. Stick to common words, rare words come out mispronounced.
Generate in 9:16 with sound
Thirty seconds, vertical, audio on: the street ambience, the footsteps and the spoken line are generated with the picture. The format posts as is, no editing.
The same canvas, tuned for different needs: each variation is two or three lines of prompt away.
The feed-native format: 9:16 vertical, thirty seconds in one single take, street sound generated with the picture. No cut to hide and no editing to do, the video posts as is, and the long take holds attention better than a string of shots: you wait for the next meme like a running gag.
The character sheet is generated from two or three selfies: your face and outfit stay identical across the whole take, with no shooting. Same principle as the put yourself on screen workflow, pushed to a full single take: it is you walking down the street, not a rough lookalike.
The animation-meets-live-action blend of hybrid films: the drawing keeps its flat colors and ink outlines, yet casts a real shadow on the pavement and passes real pedestrians. The template's instructions lock that contract: a drawn character stays drawn from the first frame to the last, never replaced by an actor.
The whole set is generated: the street comes from one reference photo, the golden-hour light, the crowd and the street sound are made with the picture. The render keeps the flaws that read as real, micro-shake, hunting autofocus, white balance catching up, the amateur-vlog grammar covered in writing a prompt that works.
Swap the memes for your mascot: it sips a coffee at the terrace, waits for the bus, steps out of your shop with a bag. Same mechanics, professional use: the mascot lives in your customers' real world, and the reference card keeps it identical from one video to the next.
Never ask the vlogger to flip the camera: the model cannot represent the gesture and resolves it by filming him from behind, which is impossible with the phone in his own hand. The body pivots to show the scene over the shoulder, the camera stays front-facing from start to finish.
Two characters sharing one five-second block swap attributes or melt into one. One meme per scene is the most important rule of the grid: the only group moment is the finale, where characters that already appeared fall in line.
Any name or text in the instructions can come out written in the picture, as a comic bubble or a shop sign. Describe the meme's pose and attitude, never its name, and keep the template's text rule: no invented writing, realistic and unreadable signage.
If the sheet goes in as a first frame instead of a reference, the clip opens on the multi-angle sheet itself. Every image in this template is an identity reference: they guide the render without ever serving as the opening frame.
The cards accept any image you import: the template ships invented examples, and the images you use are your call. Keep in mind that famous memes are protected works and many show real people, which providers often refuse as references: drawings and animals work best.
Some video models refuse any real face as a reference image, as a provider policy. Wan 3.0 accepts them: that is one of the reasons the template runs on it, along with its thirty seconds in a single take.
No, the line is performed by the model with a generated voice that fits the character. To hear your real voice, wire an audio sample as a voice reference on the Video node: Wan 3.0 accepts up to five.
The price depends on the model and resolution you pick: the public calculator shows the exact cost of the workflow before any signup, and the quote updates in the Video node with every setting. Wan 3.0's thirty seconds in 720p remain among the most affordable long takes in the catalog.
No: the character sheet generated from your selfies is enough, the vlogger and his walk are created by the model. If you want your real gait and your real street, Wan 3.0 also accepts a video as a reference: film yourself in selfie mode and the model uses it as a motion guide.
The template's grid holds six five-second scenes, one meme per scene. To show more, generate a second clip with the same character sheet and the same street photo, then join the two in editing: the shared references keep the face and scenery identical.
Whichever language you write the line in: it is performed word for word, and the ambient chatter follows the requested language. Keep common words and a short sentence, that is what comes out most naturally.
AI meme video: memes come alive in your street
Use this workflowFree account, no card required. The workflow opens pre-filled in your canvas.
Wan 3.0
Seedream 5 Pro
MiniMax H3