v2.2.0

   🚀 Features

• Horizontal canvas arrange with oldest, newest, and type order  -  by @saihhold-zhao (c96ce)

    View changes on GitHub

v2.1.0

No significant changes

    View changes on GitHub

v2.0.0

   🚀 Features

• Skill Creator L1 and DeepSeek harness LLM (v2.0.0)  -  by @saihhold-zhao (64817)

   🐞 Bug Fixes

• Keep Skill Edit/Test tabs mounted during Test generation  -  by @saihhold-zhao (58269)
• Allow DeepSeek v4.1 Flash images; mention GrokBot in READMEs  -  by @saihhold-zhao (fc8a5)

    View changes on GitHub

v1.6.0

   🚀 Features

• Image Object Removal skill and canvas toolbar (v1.6.0)  -  by @saihhold-zhao (a0b28)

    View changes on GitHub

v1.5.0

   🚀 Features

• Image asset toolbar skills (v1.5.0)  -  by @saihhold-zhao (6fbdf)

    View changes on GitHub

v1.4.0

   🚀 Features

• Image Text Editor coords, layer-split adjust fix, skill split (v1.4.0)  -  by @saihhold-zhao (d403a)

    View changes on GitHub

v1.3.1

   🚀 Features

• Layer-split boxed overlay + LLM region strip (v1.3.1)  -  by @SaihholdChiu (0efdc)

    View changes on GitHub

PoloX AI v1.3.0 — Asset Libraries

What's new

• Added Asset Libraries: keep shared images, video, and audio on your machine and reuse them across projects
• Import library assets into chat with @ (searchable third column next to project assets)
• Save canvas results into a library from the project page
• Improved Image Layer Splitter: the agent inspects the image, then asks you to Confirm / Adjust before splitting, with one target per line
• Wider canvas zoom range (about 2%–800%)
• Clearer delete copy: removing an asset takes it out of the project, not a library
• Miscellaneous small fixes and polish

更新内容

• 新增 资源库（Asset Libraries）：在本地整理可跨项目复用的图片 / 视频 / 音频
• 在 Agent 输入框用 @ 搜索并导入资源库素材（独立第三列，与项目资产并列）
• 项目画布可将生成结果一键保存到资源库
• 优化 图层拆分 流程：拆分前先由 Agent 检视画面，再确认 / 调整（Confirm / Adjust），目标逐行列出更清晰
• 画布缩放范围放宽至约 2%–800%
• 删除文案更明确：删除的是项目中的资产，不会从资源库移除
• 其他小修复与体验优化

v1.2.0

   🚀 Features

• Expose long-form video in public Skills catalog  -  by saihhold (6aa17)
• Sync Enter-send, Annotated Image Edit, and sketch agent open  -  by saihhold (f2b60)
• Sync App Store Graphics skill + truncate/expand UI  -  by saihhold (9e11b)

   🐞 Bug Fixes

• Use English for upcoming skills placeholder  -  by saihhold (fd2c0)
• Open chat media lightbox before canvas focus  -  by saihhold (b23c6)
• Default image edits to Sunburst, chat file drops, body-scroll unlock  -  by @saihhold-zhao (57d7a)
• Map generate_image input_urls to GPT Image 2.5 Sunburst  -  by @saihhold-zhao (578dd)
• AgentJobs default i2i to GPT Image 2.5 Sunburst  -  by @saihhold-zhao (c9e42)
• Close MediaLightbox on route change  -  by @saihhold-zhao (5c83d)
• AgentSlots default i2i to GPT Image 2.5 Sunburst  -  by @saihhold-zhao (77036)
• Image-editing clear-edit path defaults to Sunburst  -  by @saihhold-zhao (fd194)
• Single-generator defers annotate card when edit request is clear  -  by @saihhold-zhao (41a4d)
• Add chat file drops and Sunburst confirmation labels  -  by saihhold (d2aa5)
• Stop homepage remount from reopening sketch agent  -  by saihhold (0ff18)
• Confirm-card param expand uses Thinking-style chevron  -  by saihhold (11911)
• Result-evaluation asks before regenerate  -  by saihhold (d8952)
• Keep OSS Product Hunt skill reference-only sync  -  by saihhold (65bbd)
• Keep agent history cards and media during thinking/confirm  -  by @saihhold-zhao (846f0)
• Keep agent history cards and media during thinking/confirm  -  by @Saihhold (6efee)

    View changes on GitHub

v1.2.0-preview.1

PoloX AI v1.2.0-preview.1

Preview release — this build has not gone through full QA. Use for early tryouts and feedback; expect rough edges.

English

Skills

• Added Annotated Image Edit — mark points on an image and describe each change for precise edits
• Added App Store Graphics — turn app screenshots into a matched set of App Store–style frames

Agent & canvas

• Chat composer accepts drag-and-drop file attachments
• Clicking a chat attachment opens the media lightbox directly (previously it only focused the asset on the canvas)

Image editing defaults

• Default image-to-image path now uses GPT Image 2.5 Sunburst (generate_image, agent jobs/slots, image-editing / single-generator flows)
• Tightened the image-editing skill guidance and related confirmation copy

Long-text UI

• Collapsible long content in Thinking, choice cards, and confirmation cards when the model writes too much at once

Result evaluation

• result-evaluation still checks whether a generated image matches the brief, but no longer auto-retries on a suspected mismatch
• The agent now flags a possible mismatch, proposes a fix, and asks you whether to regenerate or accept

Other

• Assorted small fixes (lightbox routing, sketch remount, README showcase gallery, and more)

---

中文

Skills

• 新增 Annotated Image Edit：在图上标注点位并描述修改，做更精确的编辑
• 新增 App Store Graphics：把应用截图做成风格统一的 App Store 展示图

Agent 与画布

• 聊天区支持 拖拽添加附件
• 点击聊天附件会直接打开 lightbox 预览（此前只会聚焦到画布中央）

图像编辑默认模型

• 默认 image-to-image 路径改为 GPT Image 2.5 Sunburst（覆盖 generate_image、agentJobs / agentSlots、image-editing / single-generator 等）
• 优化了图像编辑 skill 说明与部分确认文案

长文本折叠

• 当模型输出过长时，Thinking、选项卡、确认卡支持折叠/展开，减少刷屏

结果评估策略

• result-evaluation 仍用于判断生成图是否符合任务意图
• 不再在疑似不匹配时自动重试；改为先提示可能不匹配、给出修改建议，再 询问你是否重生成或接受当前结果

其他

• 修复若干次要问题（lightbox 路由、sketch 重挂载、README 展示图库等）