VibeShot Candid Photography Skill

Here's the thing.

@MANISH1027512 said earlier that he'd release his photography Skill once a post crossed 100 retweets. It crossed. People showed up for it.

So he did. And he didn't just drop the Skill. He also walked through the thinking behind it, so that anyone who finishes reading could become an AI photographer themselves, or even build a Skill in their own style.

So what's the framework?

A fixed subject first. A Korean Instagram influencer, fair skin, striking figure. Then five slots.

[expression] + [outfit style] + [scene] + [moment] + [photography technique]

That's it?

Yeah, the skeleton really is that simple.

The part that actually matters comes next. He keeps hammering on one point. GPT-Image 2 is not Nano Banana, don't over-control it.

Nail down everything, what she wears, where her hands go, which way her head tilts, the exact degree the light comes from, every object in the background, and GPT starts fighting itself. It wants to satisfy all of it at once, and that's where drawing conflicts show up. Noise, for instance, is born exactly there.

So. Control the key anchors, hand the rest back to the model.

But handing it back doesn't mean typing random and calling it done.

Here's what actually matters. Build thick enum lists for the key anchors.

Tell the model random expression and it will circle between a handful of the most common faces. Give it blank and distant, eyes gently narrowed, looking down spaced out, holding back a laugh, sudden look back, brief dissociation, squinting naturally into the sun, the laziness of just waking up, and the boundary of random becomes yours to define.

That, in his view, is a big part of what separates a Skill from an ordinary prompt.

What's a Skill doing, then?

Nothing mystical. It keeps thickening those key variables into pools. Material banks, if you like.

Expression pool. Outfit style pool. Scene pool. Moment pool. Shot size pool. Focal length pool. Camera position pool. Composition pool. Foreground pool. Light pool.

Generation just recombines from them.

Think of it as bounded randomness. Not leaving the model to guess freely, not pre-deciding every pixel either. Skeleton plus variable pools, which is pretty much the underlying logic behind most image Skills going around right now.

A version you can paste straight into ChatGPT

n=10, 9:16 real-life candid photography portraits. Fixed subject, a highly photogenic Korean Instagram influencer, fair skin, striking figure. Random expression: blank and distant, eyes gently narrowed, natural squint from sunlight, half-smile, looking down spaced out, eyes closed feeling the wind, sudden look back, looking up into the distance, holding back a laugh, just-woke-up laziness, natural pause when caught on camera, slight frown, thoughtful, brief dissociation. Random outfit style: Brandy Melville, soft-innocent, elegant stepmother chic, mature feminine, Chanel-inspired tweed, loungewear at home, Korean soft-mature, French lazy ease, minimal cold, old money, vacation resort, urban commute, sweet-cool, Y2K, K-pop idol off-duty, plain-water understated luxury, desaturated luxury casualwear, relaxed-fit casual. Random scene: midsummer lotus pond, old wooden boat, lakeside boardwalk, shallow country stream, bamboo by the water, white-wall courtyard, window of an old southern house, modern glass greenhouse, stone steps by the lake, mountain road, white seaside building, city rooftop, old bus, vintage train carriage, minimal homestay, convenience store freezer, washing fruit by the river, bench under trees, rainy alley, old pier, summer poolside, reed lakeside, country orchard, white gazebo, under lotus leaves. Random moment: just sat down, about to stand up, looking down adjusting the hem, raising a hand to the hat, hair lifted by wind, peeking out from behind leaves, bending to touch the water, barefoot in the water, walking slowly along the boardwalk, daydreaming at the stern, turning sideways to look into the distance, fixing hair messed by wind, washing fruit head down, napping against the window, sudden look back, eyes closed in the sun, stepping from shade into light, hand on the railing, sitting on stone steps swinging legs, shielding the sun with a giant lotus leaf, crouching by the water, leaning slightly back, resting with hands on the ground, caught by chance passing the lens. Random shot size: extreme close-up of part of the face, shoulder close-up, chest-up portrait, waist-up medium close, above-knee medium, half-body environmental portrait, full-body environmental portrait, small-figure large-environment composition. Random focal length: 24mm ultra-wide close, 28mm wide, 35mm environmental portrait, 50mm natural perspective, 85mm compressed portrait, 105mm distant voyeuristic telephoto. Random camera position: extremely low near the water surface, looking up from under a giant lotus leaf, ground level, below waist, eye-level candid, slight high angle, obvious high angle, shooting down from stairs, through plants from the side, from behind a doorframe, from outside a window into the room, from the front of a train carriage at distance, from behind the subject, over-the-shoulder, tilted Dutch angle, subject completely off-center, photographer hidden behind environmental objects for a sneaky shot. Random composition: extreme negative space, large sky, large water surface, subject pushed to the bottom third, subject at the far edge, centered but blocked by foreground, asymmetrical, diagonal, perspective leading lines, oversized natural object overwhelming the subject, small-figure environmental portrait, foreground taking one third, foreground taking half, peeking through gaps in plants, doorframe framing, window framing, lotus leaf forming a natural circular frame, architectural geometry cutting the frame. Random foreground: fully out-of-focus lotus flowers, giant lotus leaves, tree leaves, bamboo leaves, white curtains, glass reflection, water reflection, petals, reeds, doorframe, window frame, car glass, bus seat, transparent plastic curtain, railing, edge of a white wall, wet branches. The foreground must intrude naturally into the frame and create visible occlusion; do not display every element completely. Random light: harsh midday midsummer light, dappled shade, afternoon side backlight, low-angle evening backlight, clear diffused light after rain, high-contrast window light, fill light from water reflection, bounce from a white wall, broken light through leaves, locally blown-out highlights, subject at the edge of light and shadow, bright background with slightly darker subject, only a small patch of sunlight on the face. Random but restrained color: blue sky + emerald + white, deep green + cream + skin tone, water blue + white + gray, teal + off-white + a few pink flowers, deep blue night + cold white light, post-rain gray-blue + wet green. Keep at most 3 to 4 main color blocks, avoid a riot of colors. Random photography state: natural candid, unposed, captured by chance by the photographer, subject completely unaware of the camera, slight motion blur, partial defocus, glass reflection ghosting, real lens flare, edge highlight bloom, slight overexposure, shallow depth of field, long-distance compression, close-range wide-angle distortion, natural grain, slight digital noise, the immediacy of a point-and-shoot camera. Overall: real-life photography, strong photographic quality, modern East Asian summer portrait, clean frame, clear large-shape relationships, defined spatial layers, no studio look, no commercial strobe setup, no standard influencer posing, not everyone staring into the lens, no conventional centered portrait, no complex props, no cluttered background, no over-smoothing, no plastic skin. Each image must randomly combine different scenes, outfit styles, moments, shot sizes, focal lengths, camera positions, compositions, foregrounds and lighting. Prioritize unconventional camera positions and a sense of accidental capture. Avoid repetition within the same batch.

The Skill version

Open source repo, github.com/vibeshotclub/vsc-skills

No black magic in there either. Same framework, just thicker pools, so the variables can combine in more ways. His words were, if you're sharp you can already take it from here.

He closes by mentioning VibeShotClub, the community he runs. A bunch of people who've been grinding away at AIGC visuals for a long time, sharing prompts, skills, workflows, model tricks, and the good work people in the community produce. If you're serious about AI images or AI video, or you're building your own creative workflow, you're welcome to drop by.

At the time of writing, the post sits at 1,031 likes, 2,096 bookmarks and 375K views. Original article here, x.com/MANISH1027512/status/2095347397509652954