ComfyUI text encoder weights derived from an uncensored fine-tune of Qwen/Qwen3.5-9B, packaged as single-file safetensors for direct use in ComfyUI/models/text_encoders/.
Files
File
Precision
qwen3.5_9b_uncensored_bf16.safetensors
BF16
qwen3.5_9b_uncensored_fp8_e4m3fn.safetensors
FP8 (E4M3FN)
qwen3.5_9b_uncensored_int8_convrot.safetensors
INT8 (ConvRot)
Usage
Place the file you want into:
ComfyUI/models/text_encoders/
Select it from the Qwen3.5 text-encoder loader node.
Notes
BF16 is the full-precision reference; FP8 and INT8 are quantized for lower VRAM use.
INT8 uses row-wise scaling with Hadamard-rotation outlier suppression (ConvRot).
This is an uncensored variant. Output content is unfiltered — use accordingly.
