37 lines
1.1 KiB
Python
37 lines
1.1 KiB
Python
import numpy as np
|
|
import torch
|
|
from PIL import Image
|
|
from transformers import pipeline as hf_pipeline
|
|
import json
|
|
|
|
_pipeline_cache = {}
|
|
|
|
|
|
def get_pipeline(task, model_name, **kwargs):
|
|
"""Cached pipeline loader - avoids reloading models on every call."""
|
|
key = (task, model_name, frozenset(kwargs.items()))
|
|
if key not in _pipeline_cache:
|
|
_pipeline_cache[key] = hf_pipeline(task, model=model_name, **kwargs)
|
|
return _pipeline_cache[key]
|
|
|
|
|
|
def comfyui_to_pil(image_tensor):
|
|
"""ComfyUI IMAGE tensor (B,H,W,C) [0,1] -> PIL Image (first frame)."""
|
|
img = 255.0 * image_tensor[0].cpu().numpy()
|
|
return Image.fromarray(np.clip(img, 0, 255).astype(np.uint8))
|
|
|
|
|
|
def pil_to_comfyui(pil_image):
|
|
"""PIL Image -> ComfyUI IMAGE tensor (1,H,W,C) [0,1]."""
|
|
arr = np.array(pil_image).astype(np.float32) / 255.0
|
|
if len(arr.shape) == 2:
|
|
arr = np.stack([arr, arr, arr], axis=-1)
|
|
elif arr.shape[-1] == 4:
|
|
arr = arr[:, :, :3]
|
|
return torch.from_numpy(arr)[None,]
|
|
|
|
|
|
def format_output(data):
|
|
"""Convert pipeline output to formatted JSON string."""
|
|
return json.dumps(data, indent=2, default=str)
|