Flux Kontext Experiments
Overview
Introduction
Flux Kontext just came out and it is still unclear what the best pipeline for it is. I therefore wanted to experiment with the following nodes combinations:
- Tea Cache
- ByteDance Hyper-FLUX Acceleration LoRA
- ComfyUI-Detail-Daemon
- Normalized Attention Guidance (NAG)
To do so I build the following workflow, which you can download here and used it for 8 different experiments. One important thing to notice is that I kept track of the timing in order to better understand the tradeoffs. I'm running this on a 4060ti 16GB and 32GB RAM running "flux1-dev-kontext_fp8_scaled".
Available Experiments
You can access the different experiments from here or from the sidebar menu.