Repository navigation
Releases
Featherbox 0.3.5
Featherbox 0.3.5: you can now see what Featherbox is doing, and pause or cancel downloads. Most of this release is the first outside contribution to Featherbox, from @jumppack. Thank you!
Downloads
- One-click setup shows its progress: overall percent, size, speed and time left, plus a bar for every engine and model.
- Pause, resume and cancel: pause one download or all of them, and resume later from where they stopped. Cancel setup deletes the partial downloads after asking you, and things already installed stay installed.
- Delete a single partial download with the bin button next to a paused model.
While things are being made
- Background jobs on every page: the sidebar shows what's running and what's waiting, and the window title shows the progress. When something finishes while you're on another page, a note says it's ready; click it to open.
- "Tell me when it's done" can show a Windows notification. It only asks for permission when you click it.
- Scene pictures appear while a video renders, one by one as they're painted.
- Time estimates from your own PC: after your first video, Featherbox knows how long pictures, clips and editing take on your PC, and shows "about 2 min left". It shows no estimate rather than guess.
Smaller improvements
- The Video page works without a video model: pick one and download it right there.
- Readable voice names: all 54 voices now have names like "Alloy — US female" instead of
af_alloy. - No more flashes of empty pages while a page loads.
- "← Library" link on a finished video, and Ctrl+Enter to start from the prompt box on Image, Video and Voice.
- A short launch animation when the app opens. Click or press any key to skip it.
Fixes
- A script or video that finished while you were on another page now shows up in Studio.
- Deleting a partial download no longer fails when another running download shares one of its files (for example, the Wan 2.2 text encoder).
Updating
If you have 0.3.0 or later, click Update now on the banner (or Models → Check now). From 0.2.x, download Featherbox-Setup-0.3.5.exe and run it once. Your models and creations are kept either way.
On Linux, update with git pull and start it again with ./start.sh.
The installer isn't code-signed yet. If Windows SmartScreen says "Windows protected your PC", choose More info → Run anyway.
Featherbox 0.3.4
Featherbox 0.3.4: Featherbox now runs on Linux too.
Linux (new)
Featherbox runs on Linux from source. It's tested end to end on Ubuntu 24.04 LTS: chat, Studio scripts, pictures, Hindi captions, voice and transcription all work.
sudo apt install python3 python3-venv nodejs npm ffmpeg fonts-noto-core
git clone https://github.com/AshishYadavStudio/Featherbox.git
cd Featherbox && ./start.sh
- Runs on Ubuntu 24.04 or newer, or any distro of the same age. There's no installer or tray app on Linux yet; update with
git pull. - For graphics cards, also install
mesa-vulkan-drivers libvulkan1. - Featherbox downloads the official Linux builds of its engines by itself.
- NVIDIA: text runs on CUDA (needs driver 570 or newer). Pictures and video use Vulkan, and transcription uses the processor, because those engines have no Linux CUDA build.
- AMD and Intel graphics: these use Vulkan, and built-in graphics are detected the same way as on Windows.
Linux with a real graphics card hasn't been tested yet; the tests so far ran on the processor only. If something goes wrong on your Linux PC, please open an issue on GitHub. The README has a Linux section with the details.
Windows
Works the same as before. This release mainly brings the code shared with Linux, plus GitHub now runs the tests on both systems for every change.
Updating
If you have 0.3.0 or later, click Update now on the banner (or Models → Check now). From 0.2.x, download Featherbox-Setup-0.3.4.exe and run it once. Your models and creations are kept either way.
The installer isn't code-signed yet. If Windows SmartScreen says "Windows protected your PC", choose More info → Run anyway.
Featherbox 0.3.3
Featherbox 0.3.3: this year's newest AI models, and steadier scripts with small text models.
New models (marked "New" on the Models page)
- Text: Qwen3.5 2B, 4B and 9B; Gemma 4 E2B, E4B and 12B (strong in Hindi and other Indian languages); Qwen3.6 35B-A3B for PCs with 32 GB of RAM or more. All Apache-2.0.
- Image: FLUX.3 Image Klein (October 2026): sharp pictures that follow the prompt closely, best with a 6–8 GB GPU. Qwen-Image 2.1 Turbo for 10–12 GB GPUs. Check their licenses before commercial use: FLUX.3 Klein uses the FLUX Kommunity License, and Qwen-Image 2.1 Turbo is research-only.
- Video: Wan 2.2 5B Turbo, the same model distilled to 4 steps, so clips take a fraction of the time. If you already have Wan 2.2 5B, it's a 3.2 GB extra download.
These are new to Featherbox and haven't been tested on many PCs yet. If one misbehaves on yours, please open an issue on GitHub.
Fixes
- Small text models write scripts again: with Gemma 3 1B, English scripts sometimes failed. Featherbox now repairs the small mistakes these models make, tries again more carefully, and if it still can't, suggests a bigger model you already have.
Updating
If you have 0.3.0 or later, click Update now on the banner (or Models → Check now). From 0.2.x, download Featherbox-Setup-0.3.3.exe and run it once. Your models and creations are kept either way.
The installer isn't code-signed yet. If Windows SmartScreen says "Windows protected your PC", choose More info → Run anyway.
Featherbox 0.3.2
Featherbox 0.3.2: a small maintenance release, and the first one most people can install with one click.
What's new
- Open source under the MIT License. The license now ships with the app (
app\LICENSE.txtin the install folder). The AI engines and models keep their own licenses, listed on the Models page. - Benchmark tool (for testers): the graphics name in reports is no longer cut short on AMD ("AMD Radeon" instead of "AMD Radeon(TM) 860M Graphics"), and the RTX 3050 reference now has correct GPU-use numbers.
- Behind the scenes: automatic tests now run on every change, covering the model advice, the script fallback for small models, update checks and the benchmark.
Updating
If you have 0.3.0 or 0.3.1, click Update now on the banner (or Models → Check now). Featherbox downloads the new version, checks it, installs it and restarts by itself in about 15 seconds. From 0.2.x, download Featherbox-Setup-0.3.2.exe and run it once. Your models and creations are kept either way.
The installer isn't code-signed yet. If Windows SmartScreen says "Windows protected your PC", choose More info → Run anyway.
Featherbox 0.3.1
Featherbox 0.3.1: fixes for laptops with built-in graphics, and a steadier Studio script.
Fixes
- Studio scripts with small text models: with Gemma 3 1B, writing an English script could fail with "500 Internal Server Error". The model's answer was fine, but it came wrapped in a code block that the text engine refused. Featherbox now asks again and reads the answer itself.
- Advice for built-in graphics: on laptops without a graphics card (such as Ryzen AI with Radeon 860M, or Intel Core Ultra with Arc graphics), the Models page used to say "No suitable graphics card". It now counts the shared memory these chips use and says how each model will run on them. The sidebar and Models page show the built-in graphics too.
- AMD and Intel graphics cards: their own memory is now detected, so model advice and image settings fit the card.
Benchmark tool (for testers)
- GPU use is now measured correctly on NVIDIA, AMD and Intel. Before, it showed close to 0% because the AI engines start after the benchmark does.
Updating
If you have 0.3.0, click Update now on the banner (or Models → Check now). From 0.2.x, download Featherbox-Setup-0.3.1.exe and run it once. Your models and creations are kept either way.
The installer isn't code-signed yet. If Windows SmartScreen says "Windows protected your PC", choose More info → Run anyway.
Featherbox 0.3.0
Featherbox 0.3.0: Featherbox now keeps itself up to date.
New: update notifications and one-click updates
- When a new version comes out, a banner tells you, with What's new and an Update now button.
- Update now downloads the new installer, checks it against GitHub's published checksum, and installs it. Featherbox then restarts by itself with your models and creations untouched. It takes about 15 seconds.
- It won't update while something is being made or a model is downloading.
- Featherbox looks for updates about once a day. Only that request goes to GitHub; nothing about you or your files is sent. Turn it off, or check by hand, under Models → Featherbox 0.3.0.
Updating to 0.3.0
Version 0.2.x has no updater yet, so install this version by hand once: download Featherbox-Setup-0.3.0.exe and run it. Your models and creations are kept. From now on, updates are one click.
The installer isn't code-signed yet. If Windows SmartScreen says "Windows protected your PC", choose More info → Run anyway.
Featherbox 0.2.1
Featherbox 0.2.1: a bug-fix update. Recommended for everyone on 0.2.0.
Fixed
- The app no longer goes blank after leaving Chat. On recent versions of Microsoft Edge, opening another page after Chat could leave an empty window until Featherbox was restarted.
- Leaving a page no longer hides its progress. The work always carried on in the background, but the page forgot about it. Now, if you switch pages while something is being made and come back, the page shows the progress again and the result appears when it's done. This covers:
- Image, Video and Transcribe
- Studio: writing a script or making a video
- fixing a scene or the music on a finished video
- Chat answers keep coming while you're on another page, and are complete when you return.
Install or update
Download Featherbox-Setup-0.2.1.exe and run it. Updating keeps all your models and creations.
The installer isn't code-signed yet. If Windows SmartScreen says "Windows protected your PC", choose More info → Run anyway.
Featherbox 0.2.0
Featherbox 0.2.0: local AI on your own PC, now with speech to text.
New: Transcribe
- Turn a recording, a voice note or a video (MP3, WAV, M4A, MP4, MKV…) or a microphone recording into text with timestamps.
- Click any line to play from there. Copy the text, or download it as TXT, SRT or VTT subtitles.
- Uses Whisper in 99 languages, including Hindi and other Indian languages. It can also translate to English (with the Small and Large v3 models).
- Models: Whisper Base (57 MB), Small (181 MB), Large v3 Turbo (547 MB, recommended) and Large v3 (1 GB).
- Runs on any PC's processor. On NVIDIA PCs an optional GPU engine (258 MB) is about 8× faster: 107 s of speech took 8 s on an RTX 3050 4 GB.
Also
- Shared downloads are smarter: files another model or engine already installed aren't downloaded again.
- After an update, the app always loads the new interface (no stale cached page).
Install
Download Featherbox-Setup-0.2.0.exe and run it. No admin rights are needed, and Python is bundled. Then open Models → One-click setup.
Updating from 0.1.0 keeps your downloaded models and creations.
The installer isn't code-signed yet. If Windows SmartScreen says "Windows protected your PC", choose More info → Run anyway.
So far this has only been tested on one Windows 11 laptop (Ryzen 7 5800H, RTX 3050 4 GB). Reports from other PCs, especially AMD/Intel graphics or no GPU, are very welcome.