AI fantasy photoshoot — turn your photo into an elf, sorceress or knight

How to turn your own photo into a fantasy portrait

Until recently a fantasy portrait meant one of two things: a cosplay costume plus a location shoot, or a commission from an illustrator — several days of waiting, and a face that ended up merely similar to yours rather than actually yours. Generation removes both. You hand over one selfie; the costume, the scene and the light are painted from scratch.

In Twin AI it is a single screen. Open /create/photo, attach the shot as a face reference and describe the character in words: "elf ranger, silver circlet, green hooded cloak with fur trim, misty pine forest, cool diffused light". First variants land in about a minute, then you refine with small instructions — "add a bow across the back", "make the armour worn", "drop the fog".

The fantasy looks people ask for most

You do not need to invent a character from nothing: 2026 models know genre archetypes natively, so naming one is enough. Below are the ones requested most often, with the details worth adding to the prompt so the frame does not come out generic.

• Elf or elf ranger — pointed ears, silver circlet or ear cuff, hooded cloak, forest backdrop
• Sorceress, witch, dark mage — high collar, thorned iron crown, candlelight, stone hall
• Knight or warrior — worn pauldron, leather gorget, hard side light, castle wall
• Dragon rider — scaled bronze armour, leather riding harness, warm sandstone background
• Dwarf, orc, halfling — carried by silhouette and texture: beard, scars, forged metal
• King and queen — heavy brocade, fur mantle, throne room in warm light
• Fae and dryad — thin wings, a wreath of leaves, light filtered through foliage

One face, three characters: elf ranger, dark sorceress, dragon rider

Fantasy and dark fantasy avatars

The most common request is not a full shoot but a single avatar — for Discord, a forum, a game profile. That frame has its own rules, and they belong in the prompt from the start, otherwise a beautiful wide portrait becomes a circle showing nothing but a chin.

Ask for a square, shoulders-up crop: "portrait from the shoulders up, face centred, even margins on all sides, plain single-tone background". Keep the background calm — a detailed forest turns to mush at avatar size, while a flat one recedes and leaves the face readable.

Dark fantasy is its own branch: cold palette, near-black background, metal instead of fabric, almost no colour. Models handle it without long explanations — say "dark fantasy" and name two or three objects, such as an iron crown, raven feathers and a single candle to the side.

• Format — 1:1 square, shoulders-up, face centred, equal margins
• Background — flat or heavily blurred, or everything merges at avatar size
• Dark fantasy — cold palette, black background, forged metal, one side light source
• Matching pair avatars — two references plus a shared palette in the prompt
• Game profiles — ask for a tight crop, fine armour detail is invisible at 128 px anyway

From a plain selfie to a fantasy frame

The source needs no preparation: a phone shot under normal room light works. What matters is that the face is large in frame, without dark glasses or deep shadow — that is what the model reads before carrying your features into the costume.

Likeness and drama pull against each other, and it helps to know that in advance. The more you ask the face itself to change — heavy makeup, scars, non-human features — the less of you survives in the frame. If recognisability matters more, add two or three face references and keep the transformation at the level of costume and scene.

A source like this works

This is the output

Left — an ordinary phone selfie, right — the fantasy portrait generated from it

Which model suits fantasy best

The genre demands two things at once that models deliver unevenly: a recognisable face and convincing texture — metal, leather, velvet, fur. That is why it pays to run the same prompt through several models and choose with your eyes; Twin AI has a compare mode that returns the results next to each other in one screen.

• Nano Banana Pro — best likeness through a full costume change; the default when the point is that people recognise you
• Flux 2 Pro — cinematic light and fabric texture; take it for cloaks, velvet, fur, fog and rim light
• GPT Image 2 — follows a long prompt most precisely; take it when specific armour details and a complex scene matter
• Seedream 4 — rich book-illustration stylisation when photorealism is not the goal
• Kling and Veo 3 — not for the still, but for animating a finished portrait into a short clip

Building a series instead of a single frame

One portrait is an avatar. A photoshoot is a series where the same character appears in different shots and scenes, and it is built by pinning things down and then changing exactly one.

Generate an anchor frame first and get the look right. Then in every following prompt repeat the character description word for word — the same hair, the same armour, the same colours — and change only one thing: the crop (close portrait to full length), the scene (forest, throne room, battlefield at dawn) or the light (sunrise, torches, storm). Change several at once and the character drifts, and the series falls apart.

The shortcut to the same result is /photoshoots: ready-made scene sets where you upload a photo once and get a consistent series without assembling prompts by hand.

Animating a fantasy portrait

A finished frame does not have to stay still. On /create/video, Kling and Veo 3 take your image as the first frame and paint the motion around it: a cloak moving in the wind, drifting dust, torch flame, a slow head turn. A five-second clip is made from the portrait you already have.

Describe one specific movement at a time — "wind moves the fur collar, the camera pushes in slowly". A vague "make it move" produces chaotic motion and breaks the face. Vertical clips of this kind travel far better than stills in stories and short-video feeds.

What it costs

One generation in Twin AI is 200 credits. Working through 6–10 looks with refinements takes 1,200–2,000 credits, so a whole session of trying characters still costs a fraction of a commission. Free credits on sign-up cover the first frames, and the Start subscription includes 2,600 credits. Current pricing is on /pricing.

For comparison: a commissioned fantasy portrait starts around $40 and takes days, and a cosplay shoot with costume and location runs into the hundreds. Here the whole exploration fits into one evening.