amnt docs
Concepts

Picture modes

Whether the creator's own picture is the style, an example, or both.

This page is about image agents

Slots and picture modes are features of the image connector, not of every agent. Most agents — text, API-backed, MCP-tool-backed — simply declare which boxes a buyer fills in. Start at Concepts → Connectors for the whole picture.

Slots decide what words a buyer can change. Picture mode decides what happens with images — specifically, whose picture ends up mattering: the creator's, the buyer's, or both.

The four modes

ModeCreator's picture is…Buyer's pictureExample
✍️ Words onlyNot usedNot usedIcon set, pattern, wordmark
🔒 Style lockedThe style itself❌ Can't changeA logo style, an NFT collection's look
📸 Subject swapJust a sample✅ RequiredHeadshot, pet portrait, product shot
🎭 Style + subjectThe style✅ Optional"Cartoon me," "restyle my room"

A creator picks exactly one, at publish time:

   How does your agent work?

   ○ ✍️  Just words — no pictures involved
   ○ 🔒  My picture is the style — buyers can't change it
   ○ 📸  Buyers upload their own — mine is just an example
   ● 🎭  My picture is the style + buyers can add theirs

Why mode 4 is the money-maker

"Turn my photo into your style" is the single most sellable idea in this list — and it only works because the creator's picture stays locked as the style while the buyer's picture becomes the subject. Cartoon-me generators, NFT-collection restyles, "make my room look like this" — all mode 4, all built on the same mechanism.

The reason it's not the default: it needs the creator's reference picture to genuinely define a look, not just be a nice image. Style-locked mode (3) is the simpler starting point for most first agents.

How each mode reads at a glance

Words only — no reference picture anywhere in the flow. Best for things whose whole value is the words: a pattern, an icon set, a wordmark. Cheapest to build and to run.

Style locked — the creator's picture defines the entire visual language. A buyer can never override it, because overriding it is the recipe leaking. This is the shape of a logo-style agent, or an agent that lets strangers generate art matching an already-minted NFT collection.

Subject swap — the creator's picture is a sample, shown so a buyer understands the kind of output to expect, but the buyer's own upload is what actually gets processed. Headshots, pet portraits, product photos — anything where the subject is the point and the style is secondary or uniform.

Style + subject — both matter. The creator's picture still locks the style; the buyer's picture becomes the input the style is applied to.

Output size — a purpose, not a ratio

Creators don't think in aspect ratios; they think in what it's for. amnt maps a purpose to a size rather than asking for one directly:

PurposeRatioHow
💎 NFT · avatar · logo · app icon1:1Native
🖼️ Photo · blog header3:2Native
📕 Poster · book cover · wallpaper2:3Native
📺 Thumbnail · slide16:9Native
📱 Story · reel · TikTok9:16Native
🐦 X header / wide banner3:1Cropped from 3:2
📷 Instagram feed post4:5Cropped from 2:3

For sizes the image model can't make natively, amnt generates the nearest native shape and crops. The crop-safe presets automatically add "centred with margin" to the recipe so the crop never cuts anything that matters — the creator never has to think about this.

The exact list of natively supported ratios is read from the model at build time and can change; treat the table above as the common set, not a guarantee for every future model.

On this page