50 Followers
54 Following
265 Posts
Short version: I do VFX, I eat 納豆. Longer version: I am a compositing supervisor currently based in France. I like learning stuff and speaking languages.
Websitehttps://indievfx.com
IMDBhttps://www.imdb.com/name/nm10213273
Vimeohttps://vimeo.com/pitapan
Three episodes into #Pluribus and I’m having a great time with it so far!
Using a #Blackwell 5080 with #Nuke 15/16 on Linux and the open drivers is a disaster — GPU-accelerated nodes (AI+OFX) fill VRAM and never release, causing total GPU crash after a few frames. Yesterday I did some testing in Nuke 17 and 16.1 Beta... the issue is gone! Need to wait for official release but this is amazing news.
If you're reformatting smart vectors, don't forget to multiply the vector values by the reformat scale factor. #nuke
Beeble AI just dropped Switchlight 3.0 and it’s honestly wild. Version 2 was already a massive boost for relighting workflows, but the new model just blows it out of the water.

After more than 5 years I've finally written a post on my old blog because this post is longer. It's about my recent success of repurposing old hardware at the office to build a PC with two graphics cards. It hosts two virtual machines and each one has access to a dedicated Nvidia GPU.

With two monitors, mice and keyboards two people could even work on site at the same machine 😎 But this is primarily meant to provide remote desktops for I/O and editorial.

https://www.comp-fu.com/2025/10/dual-gpu-passthrough-on-proxmox/

#proxmox

Dual GPU Passthrough on Proxmox | Stefan Ihringer's Comp-Fu

How to set up two Linux VMs with Nvidia GPUs on a single Z820 workstation using Proxmox.

When it comes to UI and UX - Apple Pages is the zBrush of the office productivity tools.
The ViTMatte for #Nuke from Rafael Silva is much more memory efficient than VitMatte from Foundry. Unfortunately on the official one I can hardly process 1080p while the older one can easily push 4K.
https://github.com/rafaelperez/ViTMatte-for-Nuke
GitHub - rafaelperez/ViTMatte-for-Nuke: [Information Fusion] Boosting Image Matting with Pretrained Plain Vision Transformers

[Information Fusion] Boosting Image Matting with Pretrained Plain Vision Transformers - rafaelperez/ViTMatte-for-Nuke

GitHub
So now after running PxF_NukeBench the numbers are even crazier. 66% average speed improvement on calculations, 21% on writes and 223% on reads. What a lovely day :)
I’ve got all the tools and OFX I use working, and the only missing parts aren’t very important: Parsec still has no Linux hosting (plus I can’t activate hardware decoding on Rocky), and I still haven’t figured out how to reprogram my extra mouse keys. (3/3)
It finally happened - I managed to pull off a fully functional workstation on #Rocky #Linux 9.6. I’d tried before a few times but gave up quickly after hitting different roadblocks - dependencies that, no matter how much I searched, I couldn’t find (or maybe spot) the right answer for. Today ChatGPT was a big help - I guess many system admins are using it, so it was highly effective in fixing the errors. (1/3)