Face clear and sharp
Upload two clear photos in order: photo 1 is the person who stands up, photo 2 is the date who stays. Soft, tiny, or cropped-out subjects make identity drift.
Viral template
A short Obsession restaurant walk-out: your photo keeps that motion. Genjutsu AI holds the timing. Stills change the look.
About 15 seconds
On this page
The panel below is the homepage studio, already on this template. The sample is the Obsession restaurant walk-out motion reference.
Reference video

I Thought We Were Having a Nice Date AI
15s · 1,140 credits · Seedance 2.5
I Thought We Were Having a Nice Date AI is a Genjutsu AI photo-to-video template on genjutsuai.app. The reference clip is one person at a restaurant booth smiling, then crying, standing up and walking away, while the date stays seated and stunned. Your still changes who is on screen; the sample keeps the staging.
Anyone who already has permission to use both faces for this restaurant walk-out — yourself and a date, a friend, or a two-person joke take.
The sample holds one person at a restaurant booth smiling, then crying, standing up and walking away, while the date stays seated and stunned. A photo changes who is on screen; it should not rewrite the staging.
Background
I Thought We Were Having a Nice Date is the restaurant beat from the 2026 horror film Obsession: a smile across the booth, a sudden cry, a stand-up, and the line “I thought we were having a nice date,” while the other person sits frozen.
Obsession opened in US theaters on 15 May 2026. On 21 May, Universal Pictures posted the restaurant scene on TikTok; by 6 November that post had about 75.8 million views and 5.4 million likes. Green-screen meme templates followed — @greenscreens4memes’ 29 May take reached about 2.7 million views — and fan remakes, including a LEGO short, kept the same walk-out. AI sites later offered it as a two-photo template.
This page is an independent Genjutsu AI photo-to-video remake of that restaurant walk-out. Genjutsu AI is not affiliated with MakeViral, Universal, CapCut, or Higgsfield.
How it works
01
This page opens with the I Thought We Were Having a Nice Date AI clip already chosen. That file is the motion, the staging, and the timing.
02
Photos are optional. With none, the template still generates from the sample. To remap the scene, upload two clear photos: first the person who cries and walks out, then the date who stays. Order decides who is who.
03
Press Generate. The button shows credits for this length and quality first. Keep source audio on by default unless you want a silent export.
Photo tips
The set, the timing, and the camera are already in the sample. The still only has to show the subject clearly.
Upload two clear photos in order: photo 1 is the person who stands up, photo 2 is the date who stays. Soft, tiny, or cropped-out subjects make identity drift.
Ordinary daylight or clean indoor light is enough. Soft focus, heavy beauty filters, and extreme HDR make identity drift.
The sample already has the Obsession restaurant walk-out staging. A photo with extra people or clutter makes the remap harder.
Use a photo of yourself, or of a friend who already agreed (or a child/pet you have permission for). Some templates ask you to skip children’s photos — follow that note when it appears.
Credits
The Generate button shows the credit price before you start. It follows the reference length and the quality you pick. A longer sample costs more, and 1080p costs more than 720p or 480p.
Sample length is the default. If a duration is missing, the studio measures the file, then shows credits. Your own reference video is priced from that file. Clips that are too long are rejected before a job starts.
Leave mute off for a soundtrack from the model, or check mute and add music later. Mute does not change the camera or the timing. A failed job returns the credits.
Use cases
The motion and the timing stay. The look changes.
Upload two photos in order (the person who leaves, then the date) and leave the sample. You get a short clip of the smile, the cry, and the walk-away.
One clip per subject who already said yes (or whose owner/guardian said yes). Separate photos keep each identity clean.
Run two photos through the same sample when you want to see which still holds the beat. Staging and timing stay fixed.
Generate with no photo to watch the motion, or to check credits and length. Add a photo on the next run.
Examples
The player at the top is the I Thought We Were Having a Nice Date AI motion reference: a smile in the booth, a cry, a stand, and a stunned date. That timing is what Genjutsu AI keeps when photos change the look.
Other published templates use the same photo-to-video studio. Each one keeps its own sample motion, and you can open it to generate on that template’s page.
A Genjutsu AI photo-to-video template of one person at a restaurant booth smiling, then crying, standing up and walking away, while the date stays seated and stunned. Upload a photo to change who is on screen; keep the sample for motion and timing.
No. The clip generates from the template alone. Upload a photo when you want the look to change — with permission.
The aim is to keep the sample’s one person at a restaurant booth smiling, then crying, standing up and walking away, while the date stays seated and stunned. Extreme crops or busy group shots make remapping harder.
Look at the Generate button. The shipped sample is about ~15s. The price follows that length and the quality you pick.
Yes. Source audio stays on uploads by default for this template. Mute in the studio if you want a silent export.
Two clear photos in order — the person who leaves first, then the date who stays — faces large and sharp. Skip group shots and tiny distant faces. Use only photos of people who already agreed.