Y2K aesthetic photos and PFPs from one selfie

How to make a Y2K photo from your selfie

A Y2K frame is not a filter laid over a finished picture. The model takes your face from the reference and rebuilds the light, the outfit and the background around it, so the prompt should describe shooting conditions rather than ask for something pretty.

In Twin AI that is one screen. Open /create/photo, add a selfie as the face reference and describe the frame: hard direct on-camera flash, a sharp shadow on a pale wall, blown highlights on the cheekbones, glossy lips, a thin-strap satin top, butterfly hair clips. The first variants come back in about a minute, and from there you edit detail by detail.

1. Upload one front-facing selfie to /create/photo — the whole face visible, no sunglasses, no deep shadow.
2. Pick the model: Nano Banana 2 keeps the likeness through a full change of clothes and light.
3. Describe the light first — hard direct flash, sharp wall shadow, slight overexposure. That is the style.
4. Add two or three styling details: butterfly clips, satin top, lip gloss, frosted eyeshadow.
5. Set the format: 1:1 for a PFP, 3:4 for a vertical frame, 9:16 for stories.
6. Change one thing per edit, so you can see which change broke the frame.

Five traits that make a frame read as Y2K

The Y2K aesthetic covers products and styling of the late 1990s and early 2000s: lime, orange and hot pink next to white and chrome, glossy and metallic surfaces, chunky rounded type — en.wikipedia.org/wiki/Y2K_aesthetic. Photography keeps five of those traits. Leave two of them out of the prompt and the result is a modern portrait in retro clothes, which reads as a miss immediately.

• Hard direct flash straight at the face — the defining trait, named in 425 of our 457 prompts.
• A sharp dark shadow on the wall right behind the shoulder, the kind only a built-in flash makes.
• Blown highlights on cheekbones, forehead and shoulders, so the skin looks lit from the front.
• Oversaturated colour and digital sharpening — the texture of early-2000s compact cameras.
• Glossy styling: lip gloss, frosted eyeshadow, butterfly clips, satin tops, velour.

What format to ask for a Y2K PFP

The profile picture is the most common use of the style: it goes on Telegram, Discord and social profiles, where the frame gets cropped to a circle. That crop has its own requirements, and they belong in the prompt from the start.

Ask for a 1:1 square, a shoulders-up portrait, the face centred and even margins on all sides. Keep the background flat — a white, hot pink or near-black wall stays readable at small sizes while a detailed room turns to mush. Keep the flash: without the shadow it reads as an ordinary portrait on a coloured background.

• Format — 1:1 square, shoulders-up, face centred in the frame.
• Background — one flat colour: white, hot pink, lime or near-black.
• Light — direct flash with a visible shadow, or the frame is not Y2K.
• Details — one or two only; more of them merge into noise inside a circle.
• Matching pair PFPs — two references and a shared palette in the prompt.

Why the three-frame strip is the most copied Y2K format

The most reproduced version of the style is not a single portrait but a vertical strip of three frames. In our data that is 137 of 457 prompts, roughly every third Y2K request. In the public Twin AI feed the same format leads on reuse: the top post was repeated 206 times with 211 likes, and across the style there are 61 posts, 2,760 likes and 1,715 repeats.

It works because three frames give what one cannot: a change of angle under identical light. The classic layout is a tight colour close-up into the lens, a high-contrast black-and-white three-quarter turn, then another colour close-up with hair covering one eye. Ask for all three in one prompt, not three, so the light and the hair stay consistent.

Filter app or generation: which one you need

The two solve different problems, and the answer depends on whether you already have the right source frame. Dazz Cam and Prequel lay grain, glow and borders over a picture you have already taken — enough if you shot yourself with flash against a white wall. What a filter cannot add is the wall shadow, the satin top or the frosted eyeshadow: either they are in the shot or they are not.

Generation works from the other end, taking only the face and building a new scene around it. The rule is simple: colour and grain over an existing frame — use a filter; different light, clothes and background — generate.

A prompt template that works for Y2K

Prompts in this style follow a predictable order: light first, then background, then styling, then texture. That is the order the authors of the most repeated frames in our feed use.

A base to start from: a Y2K-style portrait, hard direct on-camera flash, a sharp shadow on the pale wall behind, slight overexposure, neutral light background, glossy lips, defined lashes, butterfly hair clips, a thin-strap satin top, oversaturated colour, vintage digital texture, soft focus, glamorous mood.

Then change one block at a time. Club version: near-black background plus a blurred motion trail at the edge of the frame. Pink version: a hot pink wall and a zip-up velour hoodie. Strip version: open with a vertical strip of three frames and describe each frame in its own sentence.

Which model to use and what it costs

Model choice in this style is lopsided: two thirds of Y2K generations in Twin AI run on Nano Banana 2, which holds the face through a full change of clothes and light — exactly what Y2K changes. GPT Image 2 gets picked for long prompts where the order of details matters, and the image-to-image variant when the original frame has to survive and only the light should change.

Reliability is not the constraint: of 142 Y2K generations in the last 30 days, 141 finished successfully. Pricing is flat across the photo catalogue at 200 credits per image; full plans are on /pricing.

What will not work on the first try

Three things are worth knowing before you spend credits on them.

Text. Sparkly lettering over a photo was a real part of the era, but generative models still misspell letters. Getting a clean frame and adding the caption in any editor is more reliable.

Copying a specific outfit. The model understands a thin-strap satin top, but it will not reproduce one particular garment from a photo of someone else unless you attach that photo as a separate reference.

Motion. A Y2K frame is still by nature, but it can be animated: on /create/video, Kling 3.0 and Veo 3.1 Fast take a finished image as the first frame and add movement. Five seconds costs 600 credits, 900 on Veo 3.1.