D 3.5 Large: With 8 billion parameters, it performs excellently in image quality and prompt accuracy, making it the most powerful model in the SD series, suitable for professional use cases with 1-megapixel resolution.

First, download the text encoder files (if you haven't already downloaded them from SD3, Flux, or other models):

(clip_l.safetensors, clip_g.safetensors, and t5xxl), if they are not already in your ComfyUI/models/clip/ folder. For t5xxl, if your RAM exceeds 32GB, I recommend using t5xxl_fp16.safetensors; if not, I recommend using t5xxl_fp8_e4m3fn_scaled.safetensors.

sd3.5_large.safetensors needs to be placed in your ComfyUI/models/checkpoints/ directory

It does not include text encoder/CLIP weights, so you must load them separately to use this file.

Workflow usage notes:

1. The Large model is suitable for users with 16GB or more VRAM.
2. The SD 3.5 Large model itself does not include text encoder/CLIP weights, so they need to be loaded separately. Before officially generating, make sure to check whether you have installed and correctly selected these models.
3. The image resolution needs to be a multiple of 64; CFG parameter recommendation: 4.5–5.5; sampler: DPM 2M; generation steps: 30–40.