ComfyUI Node
MOSS-Image Loader (DiT + TE + VAE)
A ComfyUI node in MOSS-Image with 6 inputs and 3 outputs.
MOSS-Image Loader (DiT + TE + VAE)
- MODEL
- CLIP
- VAE
◄dit▾►
◄weight_dtypedefault►
◄text_encoder▾►
◄vae▾►
◄text_devicedefault►
◄int8_matmultrue►
CategoryMOSS-Image
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| dit | COMBO | 0 options: | |
| weight_dtype | COMBO | default | 6 options: default, bf16, fp32, fp8_e4m3fn, fp8_e4m3fn_fast, fp8_e5m2 |
| text_encoder | COMBO | bf16: a moss_image/<name>/text_encoder dir. INT8: qwen35_9b_int8_convrot.safetensors (moss_image/int8/ or text_encoders/). fp8: qwen35_9b_fp8_e4m3fn.safetensors (moss_image/fp8/ or text_encoders/). | |
| vae | COMBO | 0 options: | |
| text_deviceopt | COMBO | default | 2 options: default, cpu |
| int8_matmulopt | BOOLEAN | true | INT8 TE: int8 activations too (W8A8, faster, as anima_lora encodes). Off: the weight is dequantised for a bf16 matmul (ComfyUI's default, closer to bf16). No effect on a bf16 / fp8 TE. |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |
| CLIP | CLIP | — |
| VAE | VAE | — |