ComfyUI Extension: Newbie-Teacache-ComfyUI
Run ComfyUI workflows without the setup
No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.
TeaCache optimization nodes for ComfyUI's NewBie implementation, featuring patched loader and coefficient calculator for faster inference with configurable quality. (Description by CC)
Looking for a different extension?
Custom Nodes (0)
README
Newbie-Teacache-ComfyUI
Working with woct0rdho lighter Newbie Implementation (https://github.com/woct0rdho/ComfyUI/tree/newbie) // Use the new node TeaCache (Lumina2)
Didn't test the calculator but it should be working
Major parts of these projects were used as base and are still on the nodes:
https://github.com/spawner1145/CUI-Lumina2-TeaCache // Huge thanks to spawner1145 o/
https://github.com/comfyanonymous/ComfyUI/pull/11284 // Most of the info about how to properly make TeaCache work with NewBie came from here o/
Most of the code here was written with the help of multiple coding assistants (mainly Claude and Gemini). These nodes contain a patched loader for using TeaCache with NewBie and a Coefficient Calculator that I used to get an okayish coefficient.
Just place the "TeaCache (Newbie)" on your workflow and hook up the model to it. By default it uses the best coefficient I could find ([0, 0, 0, 4.11423217, 0.36885889]) and a speed-focused l1 value (-20s gen time on a 3060 12GB using the default models and res_multistep + linear_quadratic). You can set it between 0.3 (slowest, best quality) and 0.8 (fastest, lower quality).
The Coefficient Calculator is more of a testing node (I don't even know if it's optimal or not), mess with it if you want to.
Run ComfyUI workflows without the setup
No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.