Descriptions:
ComfyUI now includes native support for LongCat Image and LongCat Image Edit, and this tutorial from Nerdy Rodent makes the case that the model is worth serious attention for users with mid-range hardware. LongCat BF-16 comes in at just 12.5 GB with 6 billion parameters—roughly a quarter the size of the 40+ GB Qwen model—while reportedly matching it on several benchmarks.
The video walks through complete workflows using the presenter’s color-coded ‘rodent method’ of node grouping in ComfyUI, covering text-to-image generation at resolutions up to 1536×1536, latent upscaling versus image upscaling comparisons, and targeted image editing via masked regions. Specific tested settings include CFG 1.0 at 8 steps for the Turbo variant, CFG 1.5 at 40 steps for the standard model, with a warning against using CFG 4.5 on the non-Turbo model (produces burnt outputs). Generation runs at approximately 15 seconds per image on an RTX 3090. The presenter also compares two encoding methods—masked latent encoding versus the prompt-based Qwen Image Edit node—finding that prompt encoding introduces subtle head-shape shifts absent from the masked latent approach.
For anyone running a local ComfyUI setup who wants capable image generation and editing without loading a 40+ GB model, this tutorial provides ready-to-use workflow configurations and honest performance observations across multiple style types.
📺 Source: Nerdy Rodent · Published March 29, 2026
🏷️ Format: Tutorial Demo
