v0.2.6 Release

[zidage]

更新日志

[0.2.6] (8399448..c4cf768) — 2026-06-11 ~ 2026-06-24

新功能

• 胶片颗粒与 Halation 效果：新增胶片颗粒密度/粒度控制和 Halation 算子，接入编辑历史，并将效果流水线扩展到 CPU/CUDA/OpenCL/Metal 路径，补齐多后端一致性。(514a667, 5803529, e943264, c0b3043, 6c19966, 15d9228, 6f24a66, f614dae, c84016b, bac945c, 97778e7)
• 语义 AI 生成与搜索：新增图像内容识别、AI 标签生成、多模型支持、多语言/Jina CLIP 路径、标签搜索/过滤、语义搜索分发、VSS/HNSW 存储、标签提示词优化、标签分配逻辑优化，以及未使用模型标签清理。(0304551, ee94c48, 84ad809, 9d13093, c152e86, 529e587, 7a3b3ea, 3745126, 4e06e46, 3baaad7, 6b24957)
• 模型下载与激活流程：新增模型下载器 UI、基于 aria2c 的下载服务、生成 ETA 显示、下载/激活体验优化、模型激活提示，以及专用的激活对话框。(a20a1ba, 100a4a8, 157eb5, 70af466, 882ce4b, 0444a80, b0029c)
• RAW 与相机兼容性：通过 patched LibRaw 路径加入 Nikon HE/HE* 支持，更新 LibRaw submodule，并为 macOS 加入原生 CoreML 模型支持。(419fb74, 3880f2c, 5f7e288, 631fb11, b7d50a1)
• 编辑器效率改进：新增 LUT 收藏、可折叠编辑分区，以及应用 About 面板。(9570898, 0d41c03, 81ea7d6)

界面

• AI 工作流打磨：改进 AI 标签生成界面、标签展示、文件信息展示、模型切换说明、模型激活提示，以及生成/下载反馈。(aa53477, 223c24d, 3bd23ce, c41b24a, 70af466, 0444a80, b0029c)
• 网站更新：更新网站设计、文案、截图和 0.2.6 发布信息。(0028db6, a0d4c01)

性能优化

• 语义数据与模型流水线提速：改进异步模型执行，优化数据库插入，移除冗余数据复制，并重构后端路径以更好支持多模型。(bce8b9e, 9c800f9, f1ec59f, 99be227)
• GPU 处理优化：优化 Metal 高光/阴影调整性能，并改进胶片颗粒算法。(7206724, 97778e7)

缺陷修复

• 发布与打包稳定性：修复 macOS 安装依赖解析与校验、包含 HNSW 的项目包保存、CI 编译失败、VSS 配置缺失、Windows/macOS 模型打包范围、Windows 安装脚本准备，以及进程退出码上报问题。(9cc06a3, 60faf7e, 8a688cd, 66c306e, 8493f02, 4878464, c4cf768)
• RAW、编辑器与 GPU 稳定性：修复 CUDA 高光恢复运行时切换失败、编辑器死锁/卡死、随机非法内存访问、LibRaw 相关段错误，以及部分相机裁切偏移错误。(9cc0da9, 48bf884, 7cf756c, 5aea74b, 2d4ca5a, 3880f2c, 4262737)
• AI/模型正确性：修复 AI 面板生成时机、aria2c 重试逻辑、模型名称显示不一致、标签数量过多、数据库连接所有权，以及语义分支合并冲突引入的回归。(60255ae, 622c06d, 102e4d5, 2936946, cf5a823, 6a756c5, bb0852f)

文档

• 新增语义生成/搜索规划、CUDA Halation/胶片颗粒合并交接说明、网站文案更新、赞助元数据，以及 macOS/Windows 安装树校验相关发布工程脚本。(7b045c2, 8f0004a, 0028db6, a0d4c01, 5ef4433, 4878464)

Changelog

[0.2.6] (8399448..c4cf768) — 2026-06-11 ~ 2026-06-24

Features

• Film grain and halation effects: Added film grain density/granularity controls and halation operators, wired them into edit history, and brought the effect pipeline across CPU/CUDA/OpenCL/Metal paths with backend alignment work. (514a667, 5803529, e943264, c0b3043, 6c19966, 15d9228, 6f24a66, f614dae, c84016b, bac945c, 97778e7)
• Semantic AI generation and search: Added semantic image content detection, AI label generation with multiple model support, multilingual/Jina CLIP handling, label search/filter, semantic search dispatching, VSS/HNSW storage, better label prompts, label assignment improvements, and pruning for unused model labels. (0304551, ee94c48, 84ad809, 9d13093, c152e86, 529e587, 7a3b3ea, 3745126, 4e06e46, 3baaad7, 6b24957)
• Model download and activation workflow: Added the model downloader UI, aria2c-based download service, ETA display during generation, model download/activation UX improvements, activation prompts, and a dedicated activation dialog. (a20a1ba, 100a4a8, 157eb5, 70af466, 882ce4b, 0444a80, b0029c)
• RAW and camera compatibility: Added Nikon HE/HE* support through a patched LibRaw path, updated the LibRaw submodule, and added native CoreML model support on macOS. (419fb74, 3880f2c, 5f7e288, 631fb11, b7d50a1)
• Editor productivity polish: Added LUT favorites, collapsible edit sections, and an About panel for the application. (9570898, 0d41c03, 81ea7d6)

UI

• AI workflow polish: Improved AI label generation UI, tag display, file information display, model switching clarity, model activation prompts, and generation/download feedback. (aa53477, 223c24d, 3bd23ce, c41b24a, 70af466, 0444a80, b0029c)
• Website refresh: Updated website design, copy, screenshots, and release information for the 0.2.6 cycle. (0028db6, a0d4c01)

Performance

• Semantic data and model pipeline speedups: Improved async model execution, optimized database insertion, removed redundant data copies, and refactored backend paths for better multimodel support. (bce8b9e, 9c800f9, f1ec59f, 99be227)
• GPU processing optimizations: Optimized Metal Highlights/Shadows adjustment performance and improved the film grain algorithm. (7206724, 97778e7)

Bug Fixes

• Release and packaging stability: Fixed macOS install dependency resolution and verification, HNSW project packaging, CI compile failures, missing VSS setup, Windows/macOS model packaging scope, Windows install script preparation, and process exit-code reporting. (9cc06a3, 60faf7e, 8a688cd, 66c306e, 8493f02, 4878464, c4cf768)
• RAW, editor, and GPU stability: Fixed CUDA highlight reconstruction runtime-switch failures, editor deadlocks/freezes, random illegal memory access, LibRaw-related segmentation faults, and incorrect crop offsets for some cameras. (9cc0da9, 48bf884, 7cf756c, 5aea74b, 2d4ca5a, 3880f2c, 4262737)
• AI/model correctness: Fixed AI panel generation timing, aria2c retry handling, model-name display inconsistencies, excessive label counts, DB connection ownership, and merge-conflict regressions in the semantic branch. (60255ae, 622c06d, 102e4d5, 2936946, cf5a823, 6a756c5, bb0852f)

Documentation

• Added semantic generation/search planning, CUDA halation/film-grain merge handoff notes, website text updates, sponsorship metadata, and release engineering scripts for macOS/Windows install-tree verification. (7b045c2, 8f0004a, 0028db6, a0d4c01, 5ef4433, 4878464)

---

This discussion was created from the release v0.2.6 Release.