Detailed description of ComfyUI server configuration options
Server Config
settings menu only exists in the Desktop version, and this settings menu item does not exist in other versions127.0.0.1
means only local access is allowed. If you need LAN access, you can set it to 0.0.0.0
auto
: Automatically selects the most suitable precisionfp64
: 64-bit floating point precision, highest precision but largest VRAM usagefp32
: 32-bit floating point precision, standard precisionfp16
: 16-bit floating point precision, can save VRAMbf16
: 16-bit brain floating point precision, between fp16 and fp32fp8_e4m3fn
: 8-bit floating point precision (e4m3), minimal VRAM usagefp8_e5m2
: 8-bit floating point precision (e5m2), minimal VRAM usageauto
: Automatically selects the most suitable precision, recommended for users with 8-12GB VRAMfp16
: 16-bit floating point precision, recommended for users with 6GB or less VRAM, can save VRAM but may affect qualityfp32
: 32-bit floating point precision, recommended for users with 16GB or more VRAM who pursue the best qualitybf16
: 16-bit brain floating point precision, recommended for newer graphics cards that support this format, can achieve better performance balanceauto
: Automatically selects the most suitable precisionfp8_e4m3fn
: 8-bit floating point precision (e4m3), minimal VRAM usagefp8_e5m2
: 8-bit floating point precision (e5m2), minimal VRAM usagefp16
: 16-bit floating point precision, can save VRAMfp32
: 32-bit floating point precision, standard precisionauto
: Automatically manages VRAM, allocating VRAM based on model size and requirementslowvram
: Low VRAM mode, uses minimal VRAM, may affect generation qualitynormalvram
: Standard VRAM mode, balances VRAM usage and performancehighvram
: High VRAM mode, uses more VRAM for better performancenovram
: No VRAM usage, runs entirely on system memorycpu
: CPU-only mode, doesn’t use graphics cardnone
: No preview images displayed, only shows progress bar during generationauto
: Automatically selects the most suitable preview method, dynamically adjusts based on system performance and VRAMlatent2rgb
: Directly converts latent space data to RGB images for preview, faster but average qualitytaesd
: Uses lightweight TAESD model for preview, balances speed and qualityauto
: Automatically selects the most suitable attention computation methodsplit
: Block-wise attention computation, can save VRAM but slower speedquad
: Uses quad attention algorithm, balances speed and VRAM usagepytorch
: Uses PyTorch native attention computation, faster but higher VRAM usagesha256
: Uses SHA-256 algorithm for hash verification, high security but slower computationsha1
: Uses SHA-1 algorithm, faster but slightly lower securitysha512
: Uses SHA-512 algorithm, provides highest security but slowest computationmd5
: Uses MD5 algorithm, fastest but lowest securityCRITICAL
: Only outputs critical error information that may cause the program to stop runningERROR
: Outputs error information indicating some functions cannot work properlyWARNING
: Outputs warning information indicating possible issues that don’t affect main functionalityINFO
: Outputs general information including system runtime status and important operation recordsDEBUG
: Outputs the most detailed debugging information including system internal runtime details