Warn: The v0.8 executables are old (Oct 2024) and not working, prefer wheels until a new release is made.

Sorry for the inconvenience.

✅ Full change-log: https://brokensrc.dev/about/changelog/#0.8.0

New release with lots of changes and improvements, most notably:

• 📦 Batch export: Input folders of images and animate all of them 🤯 (experimental™)
• ♻️ Management: Unused versions of the software are prompted to be deleted after one week
• 💫 Version check: Show a message when newer versions of the projects are available
• 📚 Documentation: Much improved website and proper examples sections <- Start here!
• ✨ New estimator: Added Apple's DepthPro as an depth estimator option
• 🚀 Top-most CLI to manage pytorch, single binary for all platforms

And important internal changes:

• Extended support of TurboPipe to intel macs
• Move away from pyimgui to imgui-bundle:
  • Python 3.12 is officially supported now
  • Lowest Python version bumped to 3.10
• Added an option to disable TurboPipe on the WebUI
• Linux/macOS releases are now .tar.gz to preseve executable xattr
• Fix wrong time calculation for the first 3 frames of looped animations

_{Made in this GitHub Actions run}

🐛 Hotfix release with a few bug fixes:

• Fixed black video exports on Linux llvmpipe
• Fixed black video exports on macOS
• Fixed readme link to the website

^{MacOS: Executables are completely untested, feedback is welcome, use at your own risk!}

📦 Urgent release due the high influx of users, welcome! 🙂

Major changes and improvements:

• New WebUI rewrite as seen on the Readme
• Single binary for the WebUI and CLI
• Many internal bug fixes and improvements
• Many others I forgot, see commit history

I'll try keeping track of the changelog for the future!

^{MacOS: Executables are completely untested, feedback is welcome, use at your own risk!}

⚠️ Experimental releases powered by PyApp - Portable self-installing binaries

Please get in touch on Discord for errors on releases, as they're in testing stage :) !

Major changes and improvements:

• New CLI-only initial implementation of a animation/preset system, see each of the command's --help for more customization, eventually they'll come to the WebUI

• Binaries are now split between WebUI and normal CLI version. The WebUI now uses threading than multiprocessing, as macOS was happier with it than forking current process resources of gradio

• Experimental fix of #33, where there's a top-most CLI that provides the REPL with the only command being initializing the DepthScene, avoiding re-running the CLI on an instance of a scene itself, but recreating as needed

• Initial support for macOS, by fixing segfaults on shader compilation (bad non-constant global variable initialization), and torch installation flavor; might need to brew install ffmpeg for rendering

• ShaderFlow uses a new TurboPipe version that allows for waiting until a buffer is no more pending of a pipe, as the previous implementations could override the Nth frame with the Nth+1, causing duplication or tearing (serious bug)

MacOS: Executables are completely untested, feedback is welcome, use at your own risk
WebUI: I know the animations aren't configurable, use the CLI for now 👍🏻

⚠️ Experimental releases powered by PyApp - Portable self-installing binaries

Releases are now made with GitHub Actions, deployed on the main repository

• Major speed improvements on the Shader (~10x), DepthAnything v2 support by default, easy pipeline for Realesr/Waifu2x upscalers, post processing options of Depth of Field and Vignette, and an experimental WebUI, all built-in

• Windows releases should keep the terminal alive when double clicking, showing basic/recommended information for usage and a very simple REPL mode (shell, just feed back input() to the typer cli) [experimental, first implementation]

Note: Configurable animations are still WIP, but there's at least VERTICAL=1 depthflow (...) and HORIZONTAL, ORGANIC
MacOS: Executables are completely untested, feedback is welcome, use at your own risk

⚠️ Experimental release with newer architecture of single wheel distribution. Powered by PyApp

Note: The binaries are small as are bootstrapped at runtime. See PyApp for more details.

Note: To render files, select inputs, open a terminal where the executable is, and run with the same "from source code" instructions, for example: ./depthflow-cpu-windows-amd64-0.3.1.exe input -i ./image.png main -q 90 -s 2 -o ./video.mp4

Note: Configurable animations and presets are WIP, for that, run from source and edit the .update function

Proof of Concept:

PyApp release of DepthFlow for Windows and Linux in many Flavors

• ⚠️ Running directly from the Source Code is a better experience for now
• Same warnings as previous releases

Fixes

• Fixed order of building Scenes on ShaderFlow as running the input command crashed

Proof of Concept:

PyApp release of DepthFlow for Windows and Linux in many Flavors

• ⚠️ Running directly from the Source Code is a better experience for now
• The binaries are small as dependencies are installed at runtime

Different PyTorch versions installed at runtime, once: CPU, CUDA 12.1 for both or ROCm on Linux

There might be version incompatibilities, I haven't exhaustively tested but should work

PyApp release of DepthFlow for Windows and Linux. Proof of concept

• ⚠️ Running directly from the Source Code is a better experience for now
• The binaries are small as dependencies are installed at runtime
• Doesn't use GPU acceleration for Depth Estimation. Thinking on it

Fixes:

• Writing old texture data on recreation could raise errors when formats change mid way

PyApp release for Windows and Linux. Proof of concept

• ⚠️ Running directly from the Source Code is a better experience for now
• The binaries are small as dependencies are installed at runtime
• Doesn't use GPU acceleration for Depth Estimation. Thinking on it