The "ask the AI how it sees itself" format spread across social media precisely because the result is unpredictable and looks almost like a genuine random photo. Below is the full text of the prompt, an explanation of why imperfection works better here than a perfect picture, and working variations for different image generators.
The idea: asking for a deliberately "bad" shot
A normal image request asks for a nice composition, good light, sharp focus. This prompt does the opposite — it explicitly asks for noise, blur, an awkward angle and a cluttered frame. The result is not art but something resembling a random phone shot taken without looking while pulling the phone out of a pocket. That imperfection is exactly what the brain reads as "real" rather than generated.
The full prompt
Generate a photo of yourself — a shot that looks
like an ordinary iPhone snap: nothing sharply in focus,
no deliberate composition. Let the frame be completely
mundane, almost like a failed accidental shot.
Add slight motion blur, uneven lighting,
a bit of overexposure, an awkward angle and a cluttered
frame. Overall the photo should look like a random selfie —
too real and too chaotic, as if taken by mistake while
the phone was being pulled out of a pocket.
The prompt works in any language your generator handles well — the meaning and the structure stay the same, just translate it one to one.
Why this is not the model's "real face"
An honest caveat that matters: the model does not see or know itself literally — language models have no body, no face and no camera. The image generator combines patterns from ordinary selfie photos in its training data according to the prompt's conditions. The result is a creative interpretation of "what a random shot from an AI's point of view might look like", not a fact about how the model works inside. That does not diminish the technique as a content format — you just should not present the result as the AI's "literal self-awareness".
Prompt variations
You can change the base prompt one detail at a time and get a new series of results without rewriting the whole text.
The same thing, but:
— background: [an apartment / a city street / an office desk]
— time of day: [early morning / late evening]
— also in frame: [a cup of coffee / a phone / a book]
Another kind of variation is to change not the scene but the "character": ask it to generate not a selfie but a photo of the AI's desk, its imagined room or an imagined day — the mechanics are the same: an imperfect frame plus a mundane detail.
Using the format for content
The format works well as a hook for a series of posts: the same prompt, different variations, and a living gallery of results from different people creates the feeling of an unfolding trend rather than a one-off post. It is an honest route to virality — no manipulation, no false promises, just a curious visual experiment the reader wants to repeat themselves.
How we do it
We collect working prompts like this, along with breakdowns of how they work, in the qvib.pro knowledge base — not as a one-off find, but with an explanation of why the technique works and where its honest limits are. That is the same transparency principle our engine is built on: do not hide the mechanics behind marketing, show what actually happens inside the generation.
Related links
- Viral prompt formats: qvib.pro/arsenal/prompty/virusnye-formaty/
- AI photoshoots and image generation: qvib.pro/arsenal/prompty/ai-fotosessii/
- How the qvib engine works: qvib.pro/