This release reworks the generation history into a queryable database and adds several quality-of-life features around sharing, batch operations, and tag autocomplete. Generation history moved to a Room database. The legacy per-record JSON sidecar files are replaced with a Room-backed table that records additional fields (generation mode, denoise strength, upscaler id, etc.). On first launch after upgrading, a one-shot migration screen imports the existing on-disk history into the database — please wait for it to finish before exiting. Filterable history view. The third pager tab now exposes top-level current / all chips plus a bottom-sheet filter (modes, models, sizes, time range, schedulers, devices, prompt substring, sort order). Upscaled images are also persisted as first-class history entries with their upscaler id recorded. Batch save history images. Multi-select in the history view now has a save action with a confirmation dialog and modal progress UI. The screen is kept on while ModelRunScreen is active so long generations and batch operations aren't interrupted by display timeout. Share generation parameters via clipboard. A share icon on the result, history, and advanced settings parameter dialogs serializes the selected fields to the clipboard (optionally wrapped with an LDPARAMS: base64 marker). When the model finishes loading, the clipboard is parsed once and an apply dialog lets you pick which fields to import, with an option to clear the clipboard afterwards. Both toggles persist in settings. Embeddings in tag autocomplete. Imported textual-inversion embeddings are pinned to the top of the prompt suggestion list with their own badge. The suggestion popup is now IME-aware (no longer hides behind the keyboard under edge-to-edge), uses a LazyColumn for smoother scrolling, and the suggestion cap is raised to 100. Show mode and denoise strength in both the result and history parameter detail dialogs. Fixed inpaint stitch incorrectly applying to past generations — it is now scoped to the most recent generation with a real source image. Cleaned up unused string entries and fixed several translation typos. Variants : LocalDream_xxx_with_filter: Block NSFW Results. Same as Google Play version. LocalDream_xxx: No filter. Co-Authored-By: Claude Opus 4.7

本地运行文生图模型应用 ———————————— 使用 Snapdragon NPU 加速在 Android 上运行 Stable Diffusion。还支持CPU/gpu推理。 NPU 支持的设备： - 骁龙 8gen1 - 骁龙8+第一代 - 骁龙 8gen2 - 骁龙 8gen3 - 骁龙 8gen4 CPU支持的设备： - 全部 技术实施 NPU加速 SDK：利用 Hexagon NPU 的高通 QNN SDK 量化：W8A16静态量化以实现最佳性能 分辨率：固定 512×512 模型形状 性能：极快的推理速度 CPU/GPU 推理 框架：由 MNN 框架提供支持 量化：W8 动态量化 分辨率：灵活尺寸（128×128、256×256、384×384、512×512） 性能：速度适中，兼容性高 NPU 高分辨率支持 下载 512 分辨率模型后，您可以下载补丁以启用 768×768 和 1024×1024 图像生成。请注意，量化高分辨率模型可能会产生布局不佳的图像。我们建议首先以 512 分辨率生成，然后使用 img2img 的高分辨率模型（本质上是 Highres.fix）。建议的 img2img denoise_strength约为 0.75。 设备兼容性 NPU 加速支持 兼容具有以下特点的设备： 骁龙 8 Gen 1 骁龙 8+ 第 1 代 骁龙 8 Gen 2 骁龙 8 第 3 代 骁龙8精英 注意：其他设备无法下载 NPU 模型 CPU/GPU 支持 RAM 要求：~2GB 可用内存 兼容性：近年来的大多数 Android 设备