Ecosystem and Tooling

This page lists the ComfyUI node pack, the third-party runtimes, and interoperability.

ComfyUI

I recommend to use ComfyUI with the comfyui-gguf-reboot node pack. The loader nodes give the best performance. The node pack provides GGUF loader nodes for diffusion transformers, text encoders, and VAEs.

External link to Comfy RegistryGGUF Loader Node pack

The loader nodes accept GGUF files directly. Load the model with the GGUF loader instead of the standard checkpoint loader for the best performance. See Load GGUF models in ComfyUI for install instructions.

Interoperability philosophy

I do not benefit from locking you into using models converted by me and forcing you to use tools maintained by me, so I put in the effort to stay compatible with other tools in the ecosystem.

• Backward and cross-compatibility with other GGUF providers, including city96 and Unsloth.
• Workflows, quant formats, and node graphs stay open and interchangeable between providers.
• Standard formats such as Q4_0, Q4_K_M, Q8_0, and BF16/FP8 variants load natively.

Other recommended engines

stable-diffusion.cpp

A C/C++ GGUF runtime for diffusion models. It reads the same standard GGUF quant formats, so models published by the maintainer load without conversion. (Compatible types: Q4_0, Q4_1, Q5_0, Q5_1 and Q8_0)

External link to GitHubDownload stable-diffusion.cpp

Unsloth Studio

Unsloth Studio and Unsloth's dynamic quantization engines produce and execute GGUF-compatible weights. Models quantized with Unsloth load in the ComfyUI node pack, and the reverse also works. (Compatible types: Q4_0, Q4_1, Q5_0, Q5_1 and Q8_0)

External link to Unsloth.aiDownload Unsloth Studio