Releases: ruzhe85/Komiho
Release list
Komiho 1.3.1
中文 / Chinese
- 新增 图像增强「大图强制 AI 增强」开关:分辨率足够但扫描质量差的图也能靠 AI 补细节
- 新增 远程缓存进度动画
- 优化 增强状态角标显示输出分辨率
- 优化 打开书籍解码期间显示「正在打开」提示,避免黑屏无反馈
- 修复 条漫模式偶发加载卡转圈
- 修复 图片型 EPUB 打开后回退文件浏览器
- 修复 设置-存储「远程缓存」统计漏算 PDF 缓存
English
- New Image enhancement — "Force AI on large images" toggle: images with enough resolution but poor scan quality also get AI detail restoration
- New Remote cache progress animation
- Improved Enhancement badge now shows the output resolution
- Improved "Opening" hint shown while decoding to avoid a black screen with no feedback
- Fixed WebToon mode could intermittently get stuck on the loading spinner
- Fixed Image-type EPUB falls back to the file browser after opening
- Fixed Settings → Storage "remote cache" stats missed the PDF cache
Komiho 1.3.0
中文 / Chinese
- 新增 本地/WebDAV/SMB PDF阅读
- 新增 文件浏览器搜索功能
- 新增 图像增强-降噪
- 优化 文件浏览器现在按数字/字母/拼音首字母排序
- 修复 努比亚/红魔无法加载NPU模型
English
- New Local / WebDAV / SMB PDF reading
- New File browser search
- New Image enhancement — denoise
- Improved File browser now sorts by number / letter / pinyin initial
- Fixed Nubia / Red Magic failing to load NPU models
v1.2.0
v1.2.0(2026-09-19)
- 新增 图像增强 AI 超分(GPU / NPU):基础代码移植自 mihon_img_upscale
- 新增 来源管理新增启动区,选择冷启动默认行为
- 新增 同路径文件夹自动加载,散图目录读完自动续下一卷
- 优化 阅读器性能提升,修复了若干 BUG
- 优化 阅读器页面模式和条漫模式缓存数值可调
- 优化 图像增强切换实时生效
- 优化 Komga UI,下载可用了
- 优化 增强状态角标按页准确显示引擎与耗时
- 修复 i18n 全面本地化(简 / 繁 / 英),切换语言后所有界面正确显示
- 修复 冷启动每次都进欢迎页
- 修复 SMB / CIFS 共享浏览偶发读取被系统中断导致加载失败
- 修复 Komga 最近阅读封面误用系列封面
特别说明
Komiho图像增强基于解码端,GPU/NPU首次打开请耐心等待。
AnimeVideo更适合于彩漫,Omni-Mini更适合于黑白漫,Omni-Turbo性能更强耗时更长。
GPU分块大小影响处理速度,建议选择192。为确保使用体验,请选择适合于设备的模型。
NPU模型可能出现轻微偏色,这是速度和性能的取舍,并非BUG。
v1.2.0 (2026-09-19)
- New AI image upscaling (GPU / NPU): base code ported from mihon_img_upscale
- New Source management gains a "Startup" section to pick the default cold-start behavior
- New Same-path folder auto-load; loose-image folder continues to the next volume after finishing one
- Improved Reader performance improved, several bugs fixed
- Improved Reader paged and WebToon mode cache values are now adjustable
- Improved Image enhancement applies in real time
- Improved Komga UI reworked — downloads now work
- Improved Enhancement badge shows engine and timing accurately per page
- Fixed Full i18n (zh / zh-TW / en); all screens display correctly after switching language
- Fixed Cold start always landed on the welcome page
- Fixed SMB / CIFS browsing could intermittently fail when the OS reclaimed the idle connection
- Fixed Komga recents cover used the series cover instead of the book cover
Special notes
Komiho's image enhancement runs at the decode stage; please allow some time on first launch with GPU / NPU.
AnimeVideo suits color comics better, Omni-Mini suits black-and-white comics, and Omni-Turbo is more powerful but slower.
GPU tile size affects processing speed; 192 is recommended.
For the best experience, choose a model that fits your device.
NPU models may show slight color shift — a trade-off for speed and performance, not a bug.
Komiho/MihonSY NPU 模型包
Komiho/MihonSY NPU 模型包
Qualcomm HTP(NPU)图像超分模型包,每个 SoC 代次一个 APK。
需先安装 Komiho 或 MihonSY 本体。这些包是纯数据插件:不含主程序,也不含任何可执行代码。
1. 按代次选择
| HTP | Snapdragon | SoC | 模型包 |
|---|---|---|---|
| V69 | 8 Gen 1 / 8+ Gen 1 | SM8450 / SM8475 | komiho-model-v69.apk 7.9 MB |
| V73 | 8 Gen 2 / 8s Gen 3 | SM8550 / SM8635 | komiho-model-v73.apk 6.9 MB |
| V75 | 8 Gen 3 | SM8650 | komiho-model-v75.apk 6.9 MB |
| V79 | 8 Elite | SM8750 | komiho-model-v79.apk 7.0 MB |
| V81 | 8 Gen 5 / 8 Elite Gen 5 | SM8845 / SM8850 | komiho-model-v81.apk 7.1 MB |
不确定属于哪一代?五个包都装上也没关系 —— App 只显示与本机 HTP 代次匹配的模型,其余自动忽略。
2. 安装与使用
- 纯数据插件,与 Komiho / MihonSY 本体同签名,可直接覆盖安装(
versionCode随每次构建递增,不需要先卸载)。 - 安装后无需任何配置:App 在启动、以及每次返回设置页时自动扫描,模型出现在 阅读器设置 → 图像增强 → NPU。
- 列表里同系列折叠为一组(如
W2xEX),点组头展开成员;组头会显示当前选中的模型。
3. 包内模型
8 个模型,**全部为 2× 超分、INT8(W8A16:int8 权重 + 16-bit 激活)**量化。
| 系列 | 文件 | 显示名 | 倍率 | 精度 | Padding |
|---|---|---|---|---|---|
| W2xEX | w2xex-omni-smallv2-x2-int8 |
W2xEX Omni Small | 2× | INT8 | 18 |
w2xex-omni-turbo-v15-x2-int8 |
W2xEX Omni Turbo | 2× | INT8 | 10 | |
universal-fast-v2-int8 |
W2xEX Universal Fast | 2× | INT8 | 18 | |
w2xex-photo-small-x2-int8 |
W2xEX Photo Small | 2× | INT8 | 18 | |
| Real-CUGAN | realcugan-pro-x2-denoise3x-int8 |
Real-CUGAN Pro Denoise3x | 2× | INT8 | 18 |
realcugan-se-x2-denoise3x-int8 |
Real-CUGAN SE Denoise3x | 2× | INT8 | 18 | |
| NomosUni | nomosuni-span-x2-int8 |
NomosUni Span | 2× | INT8 | 21 |
| Real-ESRGAN | realesrgan-general-x4v3-x2-int8 |
Real-ESRGAN | 2× | INT8 | 35 |
4. 运行前提
- 设备必须启用 CDSP。 推理走 Qualcomm HTP(CDSP/DSP)。部分掌机或厂商 ROM 在固件层禁用了 CDSP —— 此时 App 会自动隐藏 NPU 分组,请改用 GPU 模型。
- 仅限 Qualcomm 平台:非高通设备上 NPU 分组不会出现。
- HTP 代次不匹配的包不会生效(App 按本机代次过滤,不会列出跑不起来的模型)。
5. 相关
- 模型包不需要随本体一起更新:本体升级后已装模型依然可用。
- 更换/新增模型只需重装对应代次的包;卸载包即可让该组模型从列表消失。
Komiho/MihonSY NPU Model Packages
Super-resolution model packages for the Qualcomm HTP (NPU) — one APK per SoC generation.
Install Komiho or MihonSY first. These are data-only plugins: no main app, no executable code.
1. Pick Your Generation
| HTP | Snapdragon | SoC | Package |
|---|---|---|---|
| V69 | 8 Gen 1 / 8+ Gen 1 | SM8450 / SM8475 | komiho-model-v69.apk 7.9 MB |
| V73 | 8 Gen 2 / 8s Gen 3 | SM8550 / SM8635 | komiho-model-v73.apk 6.9 MB |
| V75 | 8 Gen 3 | SM8650 | komiho-model-v75.apk 6.9 MB |
| V79 | 8 Elite | SM8750 | komiho-model-v79.apk 7.0 MB |
| V81 | 8 Gen 5 / 8 Elite Gen 5 | SM8845 / SM8850 | komiho-model-v81.apk 7.1 MB |
Not sure which generation? Installing all five is fine — the app surfaces only the models matching this device's HTP and silently ignores the rest.
2. Install & Use
- A data-only plugin signed with the same key as Komiho / MihonSY, so it installs directly over an existing version (
versionCodeincreases with every build — no uninstall needed). - No configuration required: the app scans on launch and whenever you return to the settings screen. Models appear under Reader settings → Image enhancement → NPU.
- Models of one series are folded into a single chip (e.g.
W2xEX) — tap it to expand. The chip names the model currently selected inside it.
3. Included Models
Eight models, all 2× upscale, INT8 (W8A16: int8 weights + 16-bit activations).
| Series | File | Display name | Scale | Precision | Padding |
|---|---|---|---|---|---|
| W2xEX | w2xex-omni-smallv2-x2-int8 |
W2xEX Omni Small | 2× | INT8 | 18 |
w2xex-omni-turbo-v15-x2-int8 |
W2xEX Omni Turbo | 2× | INT8 | 10 | |
universal-fast-v2-int8 |
W2xEX Universal Fast | 2× | INT8 | 18 | |
w2xex-photo-small-x2-int8 |
W2xEX Photo Small | 2× | INT8 | 18 | |
| Real-CUGAN | realcugan-pro-x2-denoise3x-int8 |
Real-CUGAN Pro Denoise3x | 2× | INT8 | 18 |
realcugan-se-x2-denoise3x-int8 |
Real-CUGAN SE Denoise3x | 2× | INT8 | 18 | |
| NomosUni | nomosuni-span-x2-int8 |
NomosUni Span | 2× | INT8 | 21 |
| Real-ESRGAN | realesrgan-general-x4v3-x2-int8 |
Real-ESRGAN | 2× | INT8 | 35 |
4. Requirements
- CDSP must be enabled. Inference runs on the Qualcomm HTP (CDSP/DSP). Some handhelds or vendor ROMs disable it in firmware — the app then hides the NPU group entirely; use a GPU model instead.
- Qualcomm hardware only — the NPU group never appears elsewhere.
- Packages of another HTP generation are filtered out, never offered.
5. Related
- Model packages do not need to be reinstalled when the host app updates.
- To add or change models, just reinstall that generation's package; uninstalling it removes those models from the picker.
v1.1.0
v1.0.6
首个对外发布的 Komiho 版本,包含以下能力:
服务器连接
- 通过 URL + API-Key(
X-API-Key)连接自托管 Komga 服务器,连接前自动校验。 - 多 Komga 服务器:可添加多个服务器,设置内一键切换激活;每条可编辑 / 删除(删除确认)。
- 旧版单连接配置首次启动自动迁移为单条连接,配置不丢。
- 修:「重新配置连接」点击后误重进主页的 bug;「清除连接」改为仅清连接数据,不再误清显示 / 主题 / 语言等偏好。
书库浏览
- 库 / 系列 / 书三层模型,对齐 Komga(Series = 书,Book = 章节)。
- 库列表、系列网格(封面来自 Komga thumbnail)、系列详情(书按序排列)。
- Home 聚合 Keep Reading / On Deck,语义对齐 Komga Web Dashboard。
- 顶栏库选择器(DropdownMenu)快速切换当前库。
阅读器
- 复用 MihonSY 完整阅读器:翻页(LTR / RTL / 垂直)、条漫、双击捏合缩放、双页、阅读设置、进度记忆。
- 进度回写:读到末页置
completed,book 级 + series 级(/api/v2/series/{id}/read-progress/tachiyomi)节流回写。 - 修:阅读器 / 详情页误显示「在 WebView 打开 / 在浏览器打开 / 分享」按钮(Komga 为内部 scheme,无真实网页,已对 Komga 隐藏);设置-阅读器底部按钮同步屏蔽这三项。
图像增强
- Lanczos3(1.5x / 2x / 2.5x / 3x),轻量无大模型。
搜索与筛选
- 全局关键词搜索。
- 按标签 / 作者筛选:主页搜索范围为所有库,库内选择为当前库;切 tab 自动重拉。
- 搜索结果复用库的显示模式(不再固定平铺 1 行 3 列)。
设置与关于
- 外观、阅读器、服务器连接(多服务器管理器)。
- 关于页:显示版本号、检查更新、GitHub 源码链接(更新检查目标仓库
ruzhe85/Komiho)。