Who is Stuubzzz?
Stuubzzz Creative Studios is a one-person AI studio run by Stuubzzz. It builds self-hosted image and video pipelines in ComfyUI, releases free workflows and node packs under the name Stubelius, and teaches how all of it works in the ComfyUI From Zero course and on YouTube.
What does the studio do?
Three things, for two kinds of people.
- Teaching. ComfyUI From Zero is a three-part video course. Part 1, the beginner part, is free. Parts 2 and 3 are a paid bundle that goes from LoRAs and VRAM fit tables to training your own LoRA and writing a custom node.
- Free tools. The workflows built for the studio's own videos are released for free on GitHub, each with a walkthrough on YouTube. Patrons get them first.
- Builds. For brands and studios: the Brand Character System, fixed-price commissions and 1-on-1 sessions.
What is Stubelius?
Stubelius is the name of the ComfyUI node packs and workflows made by Stuubzzz. There are three public ones: Stubelius Ultimate H3 for MiniMax H3 (Ultimate H3 guide), Stubelius Ultimate LTX 2.5 for LTX 2.5 (Ultimate LTX 2.5 guide), and the first one, Stubelius Director (Director guide), a fork of Muse Collective's MiniMax H3 Director V1.2 that Stubelius Ultimate H3 replaces for new installs.
Who is behind it?
That's me. I started with ComfyUI about a year ago on a PC with 8 GB of VRAM and 16 GB of RAM, because I wanted to make animation for myself. The tools I built along the way turned out to be useful to other people, which led to the YouTube channel, the node packs, the LoRAs and the course. Before that came twelve years in business development, which is why the client work is built to ship and pay back.
What has Stuubzzz released so far?
Three free Stubelius node packs and workflows on GitHub, a 2.5D style LoRA for MiniMax H3, and test videos on YouTube, each with a write-up on this site. Dates are YouTube upload dates, except the first, which is the day the Stubelius Director repo was created on GitHub.
- : Stubelius Director, the first Stubelius node pack, on GitHub.
- : the Stubelius Director video on YouTube (Director guide).
- : a speed test video, FastH3 against the 8-step Turbo LoRA on MiniMax H3 (FastH3 vs Turbo LoRA results).
- : the video on version 1.0 of the 2.5D style LoRA for MiniMax H3 (2.5D style LoRA write-up).
- : a MiniMax H3 checkpoint test video (checkpoint and turbo LoRA pairing results).
- : a blind test of five MiniMax H3 workflows (sampler and scheduler results).
- : the Stubelius Ultimate H3 video, with the pack on GitHub (Ultimate H3 guide).
- : the Stubelius Ultimate LTX 2.5 video, with the pack on GitHub (Ultimate LTX 2.5 guide).
- The course, ComfyUI From Zero: Part 1 is free on YouTube. Parts 2 and 3 are a paid bundle.
How are the tests on this site run?
On one PC. Each test post has a caption that says when the test ran, on what hardware, and how many seeds or runs are behind the figures. Here is what the numbers on the blog mean.
- Hardware. One desktop PC with an RTX 5090 (32 GB of VRAM) and 96 GB of system RAM, on Windows 11. Every video, image and benchmark on this site was rendered locally on that machine, in ComfyUI. Timings on the blog are from that card unless a post says otherwise.
- Times. The Ultimate H3 and Ultimate LTX 2.5 guides report ComfyUI's own execution time per job, with queue wait excluded. The MiniMax H3 sampler test reports wall-clock time for the whole run, post-processing included. The definitions differ, so compare times within one post, not across posts.
- VRAM. A peak VRAM figure is the whole card's memory in use, read with
nvidia-smiabout every 5 seconds, so a short spike can be missed. The VAE decode post logged Windows performance counters every 3 seconds instead, and says so. A peak on a 32 GB card is not a minimum requirement for a smaller one. - Seeds and runs. Unless the caption says otherwise, each setting ran once, on one seed. A figure from one run is one sample, not an average.
- By eye. "By eye" or "by my eye" in a post means a visual verdict with no measurement behind it.
- Captions. Each test caption gives the date of the test and the machine, and the ComfyUI version where it was noted.
- Corrections. A correction goes into the post itself, which says what changed, as the LTX 2.5 sampler post and the Ultimate LTX 2.5 guide do. The post header then shows an Updated date.
How is the name spelled?
The name is Stuubzzz: two U's, three Z's. The GitHub account is stuubszzz, with an S before the Z's. The node packs are Stubelius. The website is stuubzzz.studio.
How do I get in touch?
Email hello@stuubzzz.studio for project work. For a problem on your own ComfyUI setup, book a 1-on-1. Commissions and client projects are brand-safe and SFW.
Learn it, or have it built.
Part 1 of the course is free. Project work starts with one email.
STUUBZZZ