It loads the official DiffiT checkpoint, creates latent noise from seed, samples with the official pipeline path, decodes with the selected VAE, and outputs IMAGE.

Class IDs ITS A LIE dont use 0

class_id is zero-based:

1 = tench
2 = goldfish
3 = great white shark

Install

C:\Users\xcool\AppData\Roaming\StabilityMatrix\Packages\ComfyUI-rect\venv\Scripts\python.exe -m pip install -r C:\Users\xcool\AppData\Roaming\StabilityMatrix\Packages\ComfyUI-rect\custom_nodes\ComfyUI-DiffiT\requirements.txt

Restart ComfyUI after install.

Weights

Preferred Stability Matrix path:

C:\Users\xcool\AppData\Roaming\StabilityMatrix\Models\DiffusionModels

Fallback ComfyUI path:

C:\Users\xcool\AppData\Roaming\StabilityMatrix\Packages\ComfyUI-rect\models\diffusion_models

Required filenames:

diffit_256.safetensors
diffit_512.safetensors

Defaults

steps=250
scale_pow=4.0
cfg_scale=4.4 for diffit_256.safetensors
cfg_scale=1.49 for diffit_512.safetensors
cfg_cond=True
use_ddim=False
fp32 only

Notes

Image size is inferred automatically from the selected checkpoint.
checkpoint_path override was removed.
use_fp16 was removed.
Upstream code remains non-commercial. See LICENSE_UPSTREAM_NOTICES.md.