Everyday Accidents Candid Portrait Prompt Pack

Scrolled past a GPT image 2 prompt from DeepBlue深藍 (@DeepBlueX0) built on one rule: don't pose the subject, give her an accident.

The premise is "life's little accidents" (生活小意外), and the constraints are the interesting part. Her attention stays on whatever is happening in front of her, never on the camera. The pose has to be caused by something going wrong. And it's explicitly fine — encouraged, even — for her to look clumsy, off-balance, too slow to react, or briefly frozen. No deliberate posing.

What makes it reusable is the scenario pool: ten of them. Soda spraying the moment the cap comes off. Instant noodle soup tipping as she goes to sit down. A shopping bag splitting at the bottom just inside the door. Getting rolled up in a bedsheet while making the bed, only head and limbs showing. A takeout bag catching on the door handle and yanking her back around. A balloon escaping mid-knot. A robot vacuum bumping into her foot while she's crouching. A drying rack blown over on the balcony. An earbud cable tangled on a bag strap, pulling her body to one side. A garbage bag tearing open in the hallway.

The imaging spec is short: iPhone snapshot rendering, natural exposure, white balance and ambient light, slight motion blur or missed focus allowed, keep real traces of daily life and leave some negative space. Every image also gets a signature at the bottom left — a solid #0B3D91 deep blue dot followed by "DeepBlue" in white handwriting.

One last thing: the prompt ends with an instruction to the model rather than about it — confirm the scene with me before generating.

9:16 vertical, Korean-style amateur look, fair skin, exaggerated figure; real living space. The subject's attention is on whatever is happening right in front of her, not on the camera. The pose is triggered naturally by an unexpected event — allow clumsiness, loss of balance, being too slow to react, or a brief freeze. Nothing deliberately posed.

[Scene pool]
1. Shaken soda suddenly sprays: the cap has just been opened, foam rushes out, the subject instinctively reaches to cover the bottle mouth or dodge.
2. Instant noodle soup spills: carrying the bowl and about to sit down, it suddenly tilts, she reaches to steady it, but the soup has already hit the table.
3. A shopping bag suddenly tears: just inside the door, the bottom splits open, fruit and drinks roll out, she hurries to crouch and grab them.
4. Wrapped up in the bedsheet: while straightening it, her body gets tangled, only head and limbs showing, struggling to get free.
5. Takeout bag caught on the door handle: already walking into the room, the bag behind her snags on the handle and pulls her body back around.
6. A balloon slips away: mid-knot, it suddenly flies off, she looks up and reaches after it on instinct.
7. A robot vacuum hits her foot: crouching to tidy things, the robot bumps into her, she turns and puts out a hand to stop it.
8. The drying rack blows over: on the balcony the rack tips suddenly, she reaches to catch it as the clothes slide off.
9. Earbud cable tangled on the bag strap: about to head out, she finds the cable caught on the strap, lowers her head to untangle it, body pulled to one side.
10. Garbage bag suddenly splits: carrying it through the hallway, the bottom tears open, everything scatters across the floor, she freezes for a moment then gets ready to crouch and pick it up.

[Image]
iPhone casual snapshot rendering; natural exposure, white balance and ambient light; allow natural slight motion blur or focus error; keep real traces of daily life and leave some negative space.

Signature at bottom left: "● DeepBlue" — the dot is a solid #0B3D91 deep blue circle, "DeepBlue" in white handwriting.

Before generating, confirm the scene with me.

A small note on the original: this tweet quotes an earlier one from the same author — a companion prompt that randomly picks one of ten facial-expression and hand-gesture pairs (😏 + fingertip at the corner of the mouth, 😳 + both hands over half the face, and so on), with the same iPhone candid framing and the same signature. That earlier post did far better: 1,689 likes, 120 retweets and 166,789 views. Same idea, two entries — one fixes the pose, this one causes it. Four images came with this tweet, all 941 x 1672. At the time of archiving it had 83 likes, 6 retweets, 51 bookmarks, 5,239 views and 4 replies.

Quoted prompt: DeepBlue on X · This post: DeepBlue on X