Layer Pixel Art

layerai/skills/skills/layer-pixel-art

作者 layerai315d06db6f76MIT4 個星標收錄於 2026年10月8日更新於 2026年10月8日儲存庫8 天前更新

Use when making pixel art with Layer: sprites, retro game characters, pixel icons, pixel tiles, or a sprite sheet and walk cycle. Also when generated pixel art comes back blurry, anti-aliased, or at the wrong grid size, and when packing separate frames into a sheet or cutting an existing sheet into frames. Keywords: pixel art, sprite, sprite sheet, spritesheet, 16-bit, 32x32, walk cycle, animation frames, palette, dithering, pack, split.

僅含說明Design & Creative
AI 產生的概覽

指導使用 Layer 產生像素藝術:網格、調色盤、精靈影格、走路循環、磚塊與精靈表封裝。

功能
此技能提供使用 Layer 製作像素藝術的說明,涵蓋如何指定網格尺寸、調色盤與年代風格,以及在提示詞中應排除哪些預設效果,例如反鋸齒、模糊與漸層。它描述精靈與走路循環的逐格工作流程,以已確認的第一格作為後續影格的參考,並涉及無縫磚塊。它也說明如何使用 pack_sprite_sheet 與 split_sprite_sheet 工具,將各影格組裝成統一儲存格的精靈表,或將現有精靈表切分成儲存格。
適用情境
適用於製作精靈、復古遊戲角色、像素圖示或磚塊等像素藝術,或當產生的像素藝術出現模糊、反鋸齒或網格尺寸錯誤時。也適用於將個別影格封裝成精靈表,或將現有精靈表拆分為各影格。
執行需求
僅為說明文件,不附帶指令碼。需要可存取 Layer 產生工具,以及用於產生機制的 layer 技能,還需要 pack_sprite_sheet 與 split_sprite_sheet 工具;缺少的同類技能應由使用者安裝。

Layer Pixel Art

Overview

Pixel art is the generation task most likely to come back subtly wrong: a picture of pixel art, soft-edged and anti-aliased at an arbitrary grid, rather than pixel art. Getting it right is mostly about excluding the defaults.

Layer ships two deterministic sprite-sheet utilities alongside generation. pack_sprite_sheet assembles separate frame images into one uniform-cell sheet on a transparent background, and split_sprite_sheet cuts a sheet into its cells in reading order. These are named directly rather than discovered because they are first-party tools rather than models, with no competing alternative in the catalogue.

Generation mechanics are in the layer skill. If a sibling skill named here is missing from your available skills, ask the user to install it (npx skills add layerai/skills --skill <name>); unattended, proceed from tool schemas and flag the gap.

Name the grid and the palette

Both, every time, as numbers.

  • Grid: "32x32 sprite", "16x16 tile". A model given no grid produces something pixel-ish at no particular resolution, which cannot be cleaned up afterwards.
  • Palette: a count and a family. "Limited 16-colour palette, muted earth tones" constrains the output in a way "retro colours" does not.
  • Era, when the user has one in mind: 8-bit NES-era flat colour, 16-bit SNES-era with shading ramps, or modern high-resolution pixel art. These look nothing alike.

Exclude the defaults, in the prompt

The defects are predictable, so name them as exclusions: no anti-aliasing, no blur, no gradients, no soft shadows, hard-edged pixels only, one pixel per cell. Where the model declares negative_prompt, put them there instead. Also state the background: "isolated on transparent background" or "flat single-colour background", since a generated backdrop has to be cut out and pixel art does not survive a sloppy cut.

Sprite sheets and walk cycles

Do not ask one generation for a sheet of frames. Models produce a grid of loosely related drawings, and the character changes between cells, which is exactly what a sheet must not do.

The working sequence:

  1. Generate and approve one frame: the idle or contact pose, isolated on transparent.
  2. Generate each remaining frame separately, with the approved frame attached as reference_image, and describe only what changed: "same character, same palette, same 32x32 grid, left leg forward mid-stride, right arm back."
  3. Review the frames as a set and reroll the ones that drifted.
  4. pack_sprite_sheet to assemble them into a uniform-cell sheet. Its input schema is the authority on the argument shape, so read it rather than assuming the parameter name.

Every frame after the first references frame one, not the frame before it. Chaining each frame off the last one compounds drift, so by frame four the knight is a different knight.

When a trained pixel style is in play, attach it as reference_sets and omit base_model_id entirely, so the server picks a model that can apply it. Use the same set, the same weight, and the same reference frame on every frame of the cycle.

A four-frame cycle (contact, down, passing, up) reads as motion and is far more reliable than eight. Start there and add frames only if the motion needs them.

Going the other way, split_sprite_sheet cuts a user's existing sheet into individual frames, which is the way to edit one frame of a sheet without touching the others.

Tiles

A pixel tile that repeats seamlessly is a tiling problem before it is a pixel problem, so filter for the tileability capability and read layer-textures for how to specify scale and avoid visible repetition. Everything on this page about grid, palette, and hard edges still applies.

Worked example

"A 32x32 walk cycle for our knight sprite."

  1. Generate the contact pose. list_base_models with filter.use_case: "text_to_image", take the first result, then prompt: "32x32 pixel art sprite of an armoured knight, side view, contact pose with left foot forward, limited 16-colour palette, 16-bit era shading, hard-edged pixels, no anti-aliasing, no blur, no gradients, isolated on transparent background."
  2. Approve one. This frame is now the reference for everything else.
  3. Three more runs, one per remaining frame, each attaching the approved frame as reference_image and naming only the pose change.
  4. Check the four together for drift in palette, height, and armour detail. Reroll any that moved.
  5. pack_sprite_sheet with the four file_id values, which returns the sheet on a transparent background.

Common mistakes

  • Omitting the grid size, then trying to downsample a soft image into pixels.
  • Asking one generation for a whole sprite sheet.
  • Forgetting the anti-aliasing and blur exclusions, which are the defaults being excluded.
  • Generating frames independently with no reference frame, so the character drifts.
  • Generating an eight-frame cycle before a four-frame one works.
  • Accepting a generated background instead of asking for transparent.
  • Packing a sheet by hand when pack_sprite_sheet produces uniform cells deterministically.

來源與署名

來源:layerai/skills位於skills/layer-pixel-art提交315d06d

授權條款: MIT

內容歸原作者所有。SourceWeft 從公開儲存庫中收錄這些內容。

檢舉或申請下架

更多來自 layerai/skills 的技能

Layer Workflows

layerai

Use when running a saved Layer Blueprint workflow rather than a single generation: discovering what workflows a workspace has, reading a workflow's input schema, estimating and executing a run, polling its steps, or cancelling it. Also when a repeatable multi-step pipeline exists for a task, when the user names a workflow, or when a workflow input expects a style or a file. Keywords: workflow, blueprint, pipeline, app, multi-step, run, node graph.

待分類48 天前更新

Layer Workflow Import

layerai

將其他工具中的工作流程重建為 Layer Blueprint 圖,並透過 import_workflow 匯入。

AI & Agents48 天前更新

Layer Video Timeline

layerai

將現有片段、圖像與音訊組合成一支完整影片時間軸,包含轉場、疊加層與字幕。

Design & Creative48 天前更新

Layer Video

layerai

Use when generating video with Layer: text-to-video, animating a still image, extending a clip, adding camera motion, generating native audio or lip sync, looping animations, or planning a multi-shot ad, trailer, or cutscene. Also when a video prompt produces the wrong motion or the shot drifts off the source image. Keywords: txt2vid, img2vid, image to video, camera motion, video effects, loop, seamless, trailer, cutscene, lipsync.

待分類48 天前更新

Layer Textures

layerai

指導使用 Layer 影像模型產生無縫可平鋪的紋理,涵蓋尺度、光照與重複檢查。

Design & Creative48 天前更新

Layer Reference Sets

layerai

Use when a look must hold across many Layer generations: training a custom style or LoRA on a studio's own artwork, curating the images that go into a reference set, choosing its training category, tuning reference-set weight on a run, or deciding whether to train at all rather than attach a style reference. Also when a trained style produces weak, inconsistent, or silently ignored results. Keywords: LoRA, custom model, trained style, reference set, consistency, on-model, art direction, dataset.

待分類48 天前更新