The crying AI trend is a 10-second AI video that looks like the saddest scene of a romantic drama: a person smiles through welling tears, lets one tear fall, and walks away into a night street. You make it from one photo of yourself on videosi (videosi.io), pick city lights or rain, press Generate, and download the clip with no prompt to write.
This guide goes past the basics and answers the questions people ask once they have seen the first clip: why AI tears so often look wrong, why the background is always blurry lights, how this differs from an old live filter, which photo to use, and how to post it honestly.
Key takeaways
- The trend works because the crying is small: wet eyes, a trembling smile and one tear. Big sobbing is where AI faces fall apart.
- One sharp, front-facing photo with hair and shoulders visible does more for the result than any setting.
- It is a generated video, not a live camera filter, and it renders in a few minutes.
- A standard 10-second video (Standard model, 720p) costs 9 credits, and the first clips appeared on 24 and 25 September 2026, so the trend is past its peak.
- Make it of yourself, your pet, or an adult friend who agreed, and label it as AI-generated when you post.
In this guide
- Why do AI crying faces look over the top?
- Why is it always a night street full of blurry lights?
- How is this different from the old Snapchat crying filter?
- Which photo should I choose?
- Rainy night or city lights: which version should I make?
- What does the video model actually do with my photo?
- What sound should I add when I post it?
- How do I label it as AI on TikTok?
- Is it still worth making now that the trend is fading?
- What does one attempt cost?
- How do I make it without hurting anyone?
Why do AI crying faces look over the top?
Ask a video model to "make this person cry" and it tends to reach for the loudest version of crying it has seen: an open mouth, a scrunched face, streaks running down both cheeks. A face in motion is hard to keep stable, and exaggerated expressions are where the likeness slips first. The result can look distorted or like a different person entirely.
The crying AI trend avoids this by asking for very little. The script on the page calls for wet eyes, a trembling smile, a single tear and a turn away. There is no sobbing, no running make-up and no exaggerated distortion. Small emotional beats are easier for a model to hold steady across ten seconds, and they read as more believable to the viewer. If you have tried a one-line "crying" prompt elsewhere and hated the result, this is usually the reason.
Why is it always a night street full of blurry lights?
It is part style and part practicality. A shallow depth of field and a night setting give the clip the look of a drama film, with warm lamps and tail-lights melting into large soft circles behind you. That look signals "movie goodbye" at a glance, which is the whole joke of the trend.
It also hides things. Soft, out-of-focus backgrounds are forgiving: the model does not have to render sharp signs, crowds or fine street detail that could wobble between frames. Dark scenes with warm highlights also flatter faces and make wet eyes catch the light. The foreground figure you see over the shoulder is kept as a blurred shape for a similar reason, and its face is never shown. That keeps the focus on you and avoids a stranger's face appearing by accident.
How is this different from the old Snapchat crying filter?
Older crying effects were real-time lenses. They tracked your face through the camera and laid animated tears over it while you moved, so you saw the result live and filmed yourself. Nothing new was created: your own video simply had a layer added on top.
The crying AI trend is a different thing. You upload one still photo, and a video model generates a new ten-second scene with its own camera moves, lighting, a changing expression and a walk into the distance. You never film anything. That is why it can look like cinema, and also why it takes a few minutes to render instead of working live.
Which photo should I choose?
The close-ups sit tight on your eyes, so the photo matters more than any other choice. A few rules make the biggest difference.
- Use a sharp, front-facing photo with even light and a neutral or slight smile.
- Keep your face uncovered. Sunglasses, a hand over the mouth or hair across the eyes make the likeness drift, because eyes and mouth move the most.
- Show head and shoulders, with some hair visible. The last shot is from behind, and your hairstyle and top carry the likeness there. A photo cropped tight to the face leaves the model to invent them.
- Prefer evening or indoor photos. Both settings are dark, and a bright beach selfie looks out of place in a night street.
If your first result does not look like you, change the photo before changing anything else. Likeness is never guaranteed, but a better source image is the cheapest fix.
Rainy night or city lights: which version should I make?
The two settings on the crying AI trend page tell slightly different stories with the same five beats.
City lights at night is the classic. Dry pavement, cool blue air and warm highlights on the face, with lamps, shops and car lights turned into big bokeh circles. It suits a soft, bittersweet mood, and it blends in with how most people posted the trend.
Rainy night adds wet hair and shoulders, falling rain made visible by a cold white backlight, and shiny reflections on the ground. The tears are harder to separate from the rain, so it feels heavier and more cinematic. It is a good pick if your photo is already moody or outdoors. The trade-off is that rain adds more for the model to animate, so a clean, simple photo helps.
What does the video model actually do with my photo?
The clip is made with ByteDance's Seedance video model. According to the Seedance 2.0 page from ByteDance Seed, the model supports text, image, audio and video inputs, and it can take images, audio and video as references for control over performance, lighting, shadow and camera movement. That is the capability this trend leans on.
On videosi, your photo is the source for your face, hair, build and outfit, and a short motion reference supplies the camera: the over-the-shoulder start, the slow push-in, the cut to profile and the walk away. A five-part prompt ties the scene together, covering identity, motion, setting, a timeline of beats and a list of rules. The full prompt is not published, but the structure is the lesson. Put identity first, borrow camera work from a reference, and keep the emotion restrained.
What sound should I add when I post it?
The downloaded MP4 is silent and has no watermark, so the sound is your choice. No single creator or song is confirmed for this trend, and posts use many different sad or dramatic sounds. That is good news: you do not need to hunt for one official track.
How do I label it as AI on TikTok?
You should. TikTok announced in a September 2023 newsroom post a tool that lets creators label content that has been completely generated or significantly edited by AI, and said its synthetic media policy requires people to label AI-generated content that contains realistic images, audio or video. A generated video of a realistic person crying fits that description.
In practice, use the AI-generated option the app offers when you post, and say it in the caption too, for example "made with AI from one photo". Do the same on Reels and Shorts using whatever disclosure option each app has. Check the current wording inside the app before you post, because policies and menus change. Labelling costs you nothing, keeps the joke honest, and means nobody mistakes a drama scene for a real moment from your life.
Is it still worth making now that the trend is fading?
Here is the honest timeline from the trend radar behind videosi. The earliest clips we found are dated 24 and 25 September 2026, when it appeared as a CapCut AI template. It spread in early October through tutorials in several languages, and from 17 October it was already declining.
That matters for expectations. If you want to ride a wave of new views, you are late. But a ten-second crying drama clip still works as a reaction, a storytime opener or a joke about your own week, because the format does not depend on a specific song or creator. Remakes are still being posted. If you like the look, make it for the fun of it, not as a growth plan. If you want the next wave early, watch the trend page and the other trend pages on videosi rather than waiting for a template to go viral.
What does one attempt cost?
There is no free generation and no free trial. A standard 10-second video (Standard model, 720p) is 9 credits, and better models and higher resolutions cost more. Credit packs start at $6.99 for 10 credits and never expire, and plans start at $29.90 a month for 100 credits. A failed, timed-out or blocked render returns its credits automatically.
How do I make it without hurting anyone?
The trend is a piece of theatre, and it should stay that way. Make it of yourself, of your pet for a funny version, or of an adult friend who has agreed and is in on the joke. Do not use it to make someone look miserable as a prank, to embarrass a person, or to imply that a real person is upset. Do not upload photos of minors or public figures. videosi does not accept them, and users must be 18 or older.
Keep the context light as well. The scene is a fictional goodbye on a night street, so avoid attaching it to real sad events or to real people who did not ask to be part of it. If a friend appears in the clip, show it to them before you post it and take it down if they hesitate. Photos are deleted within 30 days.
When you are ready, open the crying AI trend page, upload one clear photo, pick your setting, and let the model do the rest.