Brand and UI design: Logo and light/dark design templates (concept artwork, not product screenshots).
Edit photos, video and audio locally, with selected on-device AI tools.
Language: English · Deutsch
Explore: Features · Source setup · Docs · Roadmap · Contribute
Interface preview — view the committed English UI test capture:
This is a development-source capture, not evidence of an approved distributable Windows build.
Edivora is an open-source desktop suite with English and German interfaces. Editing happens locally, without an account or cloud AI. Models must be prepared separately before use; the application does not download them at runtime.
Development source: 0.4.0a1. No Windows package of the current suite has been approved for public redistribution. Release approval requires, among other things, Qt/FFmpeg/model-license reviews and Windows 11 acceptance. Historical photo-only test packages do not include the current launcher or the video/audio editors.
| Editor | Implemented in the current source | Main limitations |
|---|---|---|
| Photo | JPEG/PNG/WebP, color and light adjustments, looks, crop/rotate, before/after, undo/redo, local U2NETP subject masking and .laip projects |
No layers, RAW/HDR or prompt-based generative editing |
| Video | Arrange, trim, split and duplicate up to 24 clips in one lane; picture adjustments, .edvp projects and MP4 export; local still-frame subject masking and short-clip DnCNN denoising |
Only pre-rendered full-sequence previews up to 2 minutes; no realtime multitrack preview or transitions |
| Audio | Up to 24 clips in one lane; trimming, levels, fades, WAV/MP3 export, .edap projects and local Silero VAD speech-region suggestions |
No multitrack mixing, transcription or neural noise reduction |
The launcher can open Photo, Video and Audio in separate windows while remaining open. Export creates a separate file rather than intentionally overwriting the original.
AI boundaries: On CPU, DnCNN processes selected video sections of up to 30 seconds, 720p and 30 fps. Processing may remove detail or cause flicker. Baked-in filters and lost image detail cannot be reliably reversed. GPU/CUDA/DirectML operation has not been verified by this project. .edvp and .edap project files reference source media and may contain absolute local paths.
The current development instructions use Python 3.12. From the repository root on Windows:
py -3.12 -m venv .venv
.venv\Scripts\python -m pip install -e .
.venv\Scripts\python tools\prepare_model.py
.venv\Scripts\python tools\prepare_voice_model.py
.venv\Scripts\python -m pip install torch==2.14.1 --index-url https://download.pytorch.org/whl/cpu
.venv\Scripts\python -m pip install onnx==1.23.2
.venv\Scripts\python tools\prepare_restoration_model.py
.venv\Scripts\python -m local_ai_photo_editorModel preparation requires network access and checks pinned model artifacts. The final command launches the editor selection window. Use --tool photo, --tool video or --tool audio to launch one editor directly.
The Windows development build is still called LocalAiPhotoEditor.exe and needs its accompanying directory. An approved downloadable Edivora.exe is not claimed. See Getting started.
The main stack is Python 3.12, PySide6/Qt, Pillow/NumPy, ONNX Runtime (CPU) and FFmpeg for media export.
.venv\Scripts\python -m unittest discover -s tests -v
.venv\Scripts\python -m compileall -q src tests toolsAutomated checks cover specific cases; they do not establish model quality, safe media decoding or approval for redistributed Windows binaries. Treat incoming media as untrusted.
Getting started · Media projects and AI · Models · Video restoration · Architecture · Roadmap · Contributing · Security policy
Application source code is licensed under Apache-2.0. Qt, FFmpeg, AI models and other third-party components have separate terms. Public binary distribution requires a further compliance review. See third-party notices and the public release checklist.