Vixy
f5e37bc5b2
When IP-Adapter FaceID is initialized, it modifies the pipeline's UNet cross-attention layers. Calling raw pipeline() without face embeddings leaves these layers in a broken state, causing corrupted output. Solution: When IP-Adapter is loaded but no face_image provided, call ip_model.generate() with s_scale=0.0 and zero embeddings to properly disable face conditioning while satisfying the modified layers.
41 lines
946 B
Plaintext
Executable File
41 lines
946 B
Plaintext
Executable File
# DreamTail Dependencies
|
|
|
|
# Web framework
|
|
fastapi==0.109.0
|
|
uvicorn[standard]==0.27.0
|
|
python-multipart==0.0.6
|
|
|
|
# SDXL / Stable Diffusion (upgraded for compatibility)
|
|
diffusers==0.27.0
|
|
transformers==4.38.0
|
|
accelerate==0.27.0
|
|
safetensors==0.4.2
|
|
huggingface_hub==0.21.0
|
|
omegaconf==2.3.0
|
|
|
|
# PyTorch (installed in Jetson container, but listed for reference)
|
|
# torch==2.1.0+cu121 (from Jetson L4T)
|
|
# torchvision==0.16.0+cu121
|
|
|
|
# Image processing
|
|
Pillow==10.2.0
|
|
opencv-python==4.9.0.80
|
|
numpy<2 # IMPORTANT: PyTorch in Jetson base image requires numpy 1.x
|
|
|
|
# IP-Adapter FaceID (for consistent face generation)
|
|
insightface==0.7.3
|
|
# onnxruntime-gpu installed separately for Jetson (from NVIDIA wheels)
|
|
|
|
# IP-Adapter from GitHub
|
|
git+https://github.com/tencent-ailab/IP-Adapter.git
|
|
einops # Required by IP-Adapter
|
|
|
|
# Utilities
|
|
pydantic==2.6.0
|
|
pydantic-settings==2.1.0
|
|
python-dotenv==1.0.1
|
|
aiofiles==23.2.1
|
|
|
|
# Monitoring (optional)
|
|
prometheus-client==0.19.0
|