Gregory Shaffer
c0b7f5d9d2
GPU-native, NSFW-friendly image captioner for ComfyUI. Drop-in replacement for GGUF JoyCaption nodes using transformers + bitsandbytes 4-bit. - Three nodes under AITocha/Vision category - Three system presets (Scene-only, Full, Booru tags) + Custom - Four user-prompt presets including SD-prompt format - Module-level model cache, never returns None - Apache 2.0 licensed
12 lines
456 B
Python
12 lines
456 B
Python
"""AITocha Vision — NSFW-friendly image captioning for ComfyUI.
|
|
|
|
Wraps fancyfeast/llama-joycaption-* (and pin-compatible LLaVA variants) via
|
|
HuggingFace transformers so the model runs natively on GPU (no llama-cpp
|
|
compilation required) and survives pod restarts via the standard
|
|
HF_HOME=/workspace/huggingface cache.
|
|
"""
|
|
|
|
from .nodes import NODE_CLASS_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS
|
|
|
|
__all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS"]
|