TASK 10 / HTML · V1 · Creative tool

Pixel-Art Editor

像素画编辑器

4 runs · 10.A.01 · HTML

Compare works →
TASK 10 · 10.A.01 · HTMLPixel-Art Editor
Open independent copy ↗
Pixel-Art Editor

Preview note: Embedded storage and downloads can be restricted by browser sandboxing; use the standalone view for full behavior.

ORIGINAL PROMPT

Original prompt

PROMPT/Phase1_TEST_PROMPTS_BILINGUAL.md · The original prompt is kept verbatim

Source ↗
ORIGINAL PROMPT

Original prompt

Build a small pixel-art editor.
It should support:
drawing,
erasing,
color selection,
brush size,
undo/redo,
zoom,
and exporting the result as PNG.
TRANSLATION

Chinese translation

构建一个小型像素画编辑器。它应支持绘制、擦除、颜色选择、画笔大小、撤销/重做、缩放以及将结果导出为 PNG。

Expected outcome

The deliverable should be a lightweight but complete pixel-art editor. Its workspace should use a discrete pixel grid, with drawing and erasing operating precisely on logical pixels and continuous strokes avoiding unintended gaps. Users should be able to choose color and brush size. Undo and redo must restore complete canvas states and correctly handle a new edit branch. Zooming should preserve nearest-neighbor hard edges without changing artwork data. PNG export should produce a valid file with the configured dimensions and transparency, excluding editor grids and cursor overlays.

Additional verification guidance

  • Verify pixel data for clicks, slow and fast drags, multiple brush sizes, and strokes at canvas boundaries.
  • Confirm erasing creates transparent rather than white pixels and that color changes affect only future strokes.
  • Perform a multi-step edit, undo, redo, undo, and new edit; confirm correct history order and redo-branch invalidation.
  • Check that multiple zoom levels remain crisp and data-invariant; export PNG and verify signature, dimensions, pixel colors, alpha, and scale behavior.
THREE INTERPRETATIONS

Same task, three model interpretations

All three results use the same width and the same preview ratio; no model is marked as best.

COMPARISON METRICS

Comparison metrics

Cross-model comparison must note differences in harness, budget and isolation.

MetricDeepSeek-V4.1-Flashrun-deepseek-v4-1-flash-exp-0910-task-10-r1Muse Spark 1.3run-muse-spark-1-3-task-10-r1GPT-5.6 Solrun-gpt-5-6-sol-task-10-r1K3run-k3-task-10-r1
Runtime7:170:466:146:23
API calls34779
Tool calls40668
Failures / retries2001
Token throughput1.8M93.8K220.6K336.7K
AI review91/10093/10091/10094/100
RUN RECORDS

Run records

Two runs of the same task version can be compared side by side.

10.A.01 · run-deepseek-v4-1-flash-exp-0910-task-10-r1DeepSeek-V4.1-Flash● Archived
Runtime
7:17
Token throughput
1.8M
API calls
34
View archive record →
10.B.01 · run-muse-spark-1-3-task-10-r1Muse Spark 1.3● Archived
Runtime
0:46
Token throughput
93.8K
API calls
7
View archive record →
10.C.01 · run-gpt-5-6-sol-task-10-r1GPT-5.6 Sol● Archived
Runtime
6:14
Token throughput
220.6K
API calls
7
View archive record →
RELATED TASKS

Related tasks

View all tasks →