ComfyUI Node: Load Diffusion Model INT8 (W8A8)
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.
Category
loaders
Inputs
unet_name
weight_dtype
- default
- fp8_e4m3fn
- fp16
- bf16
model_type
- anima
- boogu
- chroma
- ernie
- flux2
- flux2_fast_unsafe
- hidream o1
- ideogram4
- krea2
- ltx2
- qwen
- sdxl
- wan
- z-image
on_the_fly_quantization BOOLEAN
outlier_method
- none
- convrot
- quarot
- hadanorm
small_batch_fallback
- only_small_layers
- always
- never
runtime_backend
- torch_int_mm
- triton
- triton_legacy_unsafe
prepack_int8_weights BOOLEAN
Outputs
MODEL
Extension: ComfyUI-INT8-Fast-Fork
Fork of node to load models in INT8 for 1.5~2X Speed gains on 30 series cards. Contains additional fixes and performance improvements.
Authored by SparknightLLC
Looking for a different node?
More nodes in ComfyUI-INT8-Fast-Fork
Other loaders nodes
- Additional Features With Attention
- Anima EasyControl (KSampler)
- Apply Anima ControlNet-LLLite
- Apply Trained LoRA
- Requires Asset
- Load Attention couple
- Auto Model Finder (Experimental)
- Apply CDTuner
- Checkpoint File Loader (Soze)
- Checkpoint Loader Iterator
- Load Checkpoint Mix With Diffusers
- Load FP4 or NF4 Quantized Checkpoint Model
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.