The Diamoji Artistic Process

There is a particular kind of artwork that stops people mid-scroll — a portrait, recognizable and expressive, but made entirely from emoji. Look closely and you see a laughing face, a crescent moon, a red heart, a soccer ball. Step back, and it resolves into a person. This is Diamojism, and the program I use to produce it is not simply a filter. It is a structured creative tool that I wrote that allows me to make meaningful decisions at every stage, in a deliberate sequence.

Diamoji mosaics are built in five sequential phases: physical scale, tile subdivision, facial detail, emoji selection, and tonal refinement. Each phase must be completed before the next can begin, and each one creates the conditions the next depends on.

Phase 1: Deciding What the Work Physically Is

Before any emoji is placed, I establish the physical scale of the final piece — how large it will print, at what resolution, and what the smallest and largest tiles can be. These aren't merely technical choices. Tiles that are too small shrink the emojis into indistinct texture; tiles that are too large sacrifice portrait legibility. Five core parameters govern this phase, and my goal is a scale that supports both emoji visibility and recognizability at the same time.

Phase 2: Shaping How Detail Divides the Canvas

The program divides the source image into a grid of tiles, but not a uniform one. Areas with more visual complexity are split into smaller tiles, allowing more emoji to describe them. I control how sensitive the system is to detail, how deep subdivision can go, and how aggressively the grid responds to any given region — roughly five parameters in total. The goal is a tile structure that feels intentional before a single emoji has been chosen.

Phase 3: Refining the Artistically Important Details

With the grid established, I build a facial hierarchy — a layered importance map that ensures the eyes, mouth, and other critical features receive finer tile resolution than the surrounding image. The program detects faces and optional facial landmarks automatically, but I control how far each region expands, how sharply influence falls off at its edges, and whether non-face areas like hands or clothing should receive comparable attention. About fifteen parameters are active here, and my goal is selective clarity: fine detail where recognition depends on it, consistent structure everywhere else.

The face-detection step runs automatically, but every parameter that shapes it — how far each region expands, how sharply its influence falls off — is set by hand. The program finds the face; the artist decides how much it matters.

Phase 4: Choosing Which Emojis Fill the Tiles

This is where the mosaic takes on its character. For each tile, the program scores candidate emojis against the source image using three signals: color similarity, edge alignment, and the spatial distribution of edges within the tile. I adjust how much weight each signal carries, how many candidates are considered, how strongly the program discourages repetition, and whether rotated emoji variants are eligible. About eight parameters govern this phase. I consider the work here complete when the portrait reads convincingly — without any tonal correction applied yet.

Phase 5: Refining the Light Without Losing the Emojis

Once tile structure and emoji selection are locked, I turn to tone. Each tile can be individually brightened or darkened to match the source image's luminance, but the correction must stay restrained — too much, and the emoji layer begins to feel flat or overprocessed. Seven parameters control overall brightness, per-tile correction strength, and the ceiling and floor on how much any tile can shift. Global contrast and saturation adjustments are available as finishing tools, meant to refine what is already working rather than compensate for earlier problems.

The phases move in a strict sequence for good reason: each one creates the conditions the next depends on. Scale enables subdivision; subdivision enables the importance map; the importance map enables facial clarity; and tone can only be refined once scoring is stable. Diamojism is not a one-click process — it is a composed, iterative practice, and the artwork it produces carries the decisions made at every stage.

Diamojism is not a filter or a one-click process. The five phases move in strict sequence because each one creates the conditions the next depends on — and the finished artwork carries the decisions made at every stage.
Previous
Previous

Diamojis in the Wild

Next
Next

From Pixels to Diamoji: A Layman’s Guide to How it’s Made