Turn a natural-language idea or an uploaded photo into a restrained 8-bit pixel editorial. 8ドルbit-pixel-art distills the source into a small set of coarse pixel masses, preserves the visual relationship that matters, and gives the result intentional blank space instead of applying a uniform pixel filter.
The default look is a vertical 3:5 composition with a clean paper field and a soft-edged pixel cluster. When the source depends on continuous scenery, it can instead use an intentional edge-to-edge pixel composition.
8bit-pixel-art is an image-generation skill, not merely a prompt-writing helper. It can create an original pixel editorial from an idea or reinterpret an uploaded image while keeping its important structure recognizable.
- Personal images - wallpapers, journal visuals, social posts, and quiet visual experiments.
- Editorial and publishing - article covers, newsletters, zines, cultural content, and campaign imagery.
- Concept studies - a single visual relationship distilled from a mood, scene, object, or memory.
- Photo transformation - landscapes, architecture, people, and travel images translated into coarse, source-aware pixel compositions.
| Invoke with | 8ドルbit-pixel-art |
| Modes | create for an idea; fuse for an uploaded image |
| Accepts | Natural-language briefs, exact copy, placement directions, source photos, and visual references |
| Returns | Generated bitmap, final prompt, and a compact production note |
| Default format | Portrait 3:5, adapted when another ratio is explicitly requested |
| Default behavior | Generates a finished image and returns the final prompt |
Describe an original visual in ordinary language:
8ドルbit-pixel-art create a quiet blue study of a boat crossing a lake at dusk
Upload an image and ask for a source-aware transformation:
8ドルbit-pixel-art fuse this photo into a sparse 8-bit pixel editorial
Attach a photo with exact copy and placement requirements:
8ドルbit-pixel-art fuse this photo into a pixel style image.
Add the exact small text "Under The Tree" in the blank area.
create turns a textual art direction into a finished pixel editorial. These examples show how a subject, its position, the blank field, and exact typography can be specified in the same request.
User prompt:
创作一张夜行列车的稀疏像素编辑视觉,生成在左侧(可以是居中或左下角),然后在右侧空白处用一条像素轨道(一条细线)承载这一段文字:"Midnight Train".
02 · Twilight JungleUser prompt:
创作一张日暮森林像素图,生成右下角和下方,然后在所生成图片地块的左侧(距离右侧的树要有一定的距离)有一个小篝火冒着像素粒子烟雾,烟雾上升之后承接着一段小字:"Twilight Jungle".
The normal conversion preserves a small number of large forms and the source palette, then turns them into coarse pixel blocks on a clean field.
Original photo -> pixel result. The water, long ridgeline, warm shore, and small boat are reduced to broad value bands and stepped shapes. 02 · Riverside city
Original photo -> pixel result. The river corridor, bridge, tree masses, and distant mountains remain readable without reproducing the full photograph.
Text is added only when requested. Keep the wording exact and specify that it belongs in the blank area; the composition reserves space for it rather than treating it as an afterthought.
Original photo -> pixel result. User input:
将图片转换成像素风,并在空白处添加一段小字"Under The Tree".
04 · Sunlight Hitting The Golden MountainOriginal photo -> pixel result. User input:
将图片转换成像素风,并在空白处添加一段小字"Sunlight Hitting The Golden Mountain".
- Source-first abstraction - retain the anchor, largest masses, axes, rhythm, and color roles; discard incidental surface detail.
- Coarse pixel marks - use crisp, enlarged blocks, stepped silhouettes, segmented bands, and irregular clusters rather than fine tracing or anti-aliasing.
- Deliberate blank space - soft-edged paper mode normally leaves 70-90% of the canvas empty, with the pixel cluster usually occupying 8-18%.
- Adaptive composition - follow an explicit placement request first; otherwise use the source's visual weight and useful negative space.
- Two boundary modes - use a soft-edged, inset cluster for isolatable subjects, or full bleed for landscapes and scenes that require continuous space.
- Controlled text - themes do not become titles automatically. Exact copy, placement, dates, labels, and other visible details are used only when supplied.
- Read the intent - identify the subject, source relationship, requested copy, placement, palette direction, and intended use.
- Choose a boundary mode - select soft-edged paper mode for an isolated anchor or justified full bleed for continuous scenery.
- Distill and compose - keep one dominant relationship and no more than a few supporting cues; choose the blank field before rendering.
- Compile the prompt - state retained features, pixel mark system, color roles, text allowlist, composition, and exclusions against filters or hard panels.
- Generate and inspect - check source fidelity, pixel scale, hierarchy, blank space, readable requested text, and unrequested visual clutter.
- Return the working recipe - provide the image, final prompt, and concise production note.
Clone the repository into the Codex skills directory:
git clone https://github.com/TwentyfiveBTea/8bit-pixel-art.git \
~/.codex/skills/8bit-pixel-artRestart Codex if the skill does not appear immediately. To update an existing installation:
git -C ~/.codex/skills/8bit-pixel-art pullCreate a new image from a short brief:
Use 8ドルbit-pixel-art to create a sparse pixel editorial of a night train.
Transform an image while preserving its recognizable relationship:
Use 8ドルbit-pixel-art to transform this photo into an 8-bit pixel editorial.
Keep the bridge, river path, and dark tree masses recognizable.
Request exact text in the clean field:
Use 8ドルbit-pixel-art to transform this photo into a pixel style image.
Place the exact small text "Sunlight Hitting The Golden Mountain" in the blank area.
The skill separates a theme from approved in-image copy. A request about a mountain does not automatically authorize a MOUNTAIN title, a translation, a date, a signature, or decorative characters. Supply exact wording and a location whenever typography matters.
Image models can still distort lettering. The skill checks visible text and makes one focused correction when it finds malformed or unapproved copy. For long, legally sensitive, or brand-critical text, add final typography in a dedicated layout tool after generating the artwork.
8bit-pixel-art/
├── SKILL.md # Complete workflow and visual rules
├── agents/openai.yaml # Codex skill-list metadata
├── examples/ # Images shown in this documentation
├── README.md # English documentation (default)
├── README.zh-CN.md # Simplified Chinese documentation
└── README.ja.md # Japanese documentation