Remove aDetailer processing

app.py

@@ -9,7 +9,6 @@ import time
 import urllib.parse
 import urllib.request
 import uuid
-from functools import lru_cache
 import warnings
 from asyncio.base_events import BaseEventLoop
 from concurrent.futures import ThreadPoolExecutor
@@ -45,7 +44,7 @@ from fastapi.responses import FileResponse, Response
 from gradio_client import Client
 from gradio.routes import App
 from huggingface_hub import download_bucket_files, list_bucket_tree
-from PIL import Image, ImageDraw, ImageFilter, ImageFont
+from PIL import Image, ImageDraw, ImageFont
 from PIL.PngImagePlugin import PngInfo
 from pydantic import BaseModel, ConfigDict, Field
 from starlette.background import BackgroundTask
@@ -89,19 +88,6 @@ MODEL_KINDS = ("checkpoints", "vae", "loras")
 MODEL_SUFFIXES = (
     ".bin", ".ckpt", ".pkl", ".pt", ".pt2", ".pth", ".safetensors", ".sft",
 )
-AD_PASSES = (
-    (
-        "Eyes.pt",
-        "masterpiece, best quality, eyes, detailed eyes",
-    ),
-)
-AD_MODELS = tuple(name for name, _ in AD_PASSES)
-AD_CONF = 0.3
-AD_PADDING = 32
-AD_BLUR = 4
-AD_DENOISE = 0.3
-AD_STEPS = 22
-AD_CFG = 5
 ASSETS = {
     "checkpoint": (
         "52_novaAnimeXL_ilV190.safetensors",
@@ -377,10 +363,6 @@ def init_comfy():
         folder_paths.add_model_folder_path(kind, str(DATA_DIR / kind))
         folder_paths.add_model_folder_path(kind, str(local), is_default=True)
 
-    ad_dir = DATA_DIR / "ultralytics"
-    ad_dir.mkdir(parents=True, exist_ok=True)
-    folder_paths.add_model_folder_path("ultralytics", str(ad_dir))
-
     state.update(
         apply_lora=comfy.sd.load_lora_for_models,
         chains={},
@@ -521,20 +503,6 @@ def generate_core(body):
             samples=samples,
         )[0]
 
-        sampler = request.second_sampler if request.upscale else request.sampler
-        scheduler = request.second_scheduler if request.upscale else request.scheduler
-        model_name = (
-            request.second_model
-            if request.upscale and request.second_model
-            else request.model
-        )
-        model, clip = load_model(model_name)
-        negative = state["encode"].encode(clip=clip, text=request.negative)[0]
-        image = run_adetailer(
-            image, model, clip, negative, sampler, scheduler,
-            secrets.randbits(64),
-        )
-
         return image[0], first_seed, second_seed
 
 
@@ -653,74 +621,6 @@ def stage_model(kind, name):
     return target
 
 
-@lru_cache(maxsize=len(AD_MODELS))
-def load_ad_model(name):
-    from ultralytics import YOLO
-    path = state["folders"].get_full_path("ultralytics", name)
-    if path is None:
-        raise ValueError(f"aDetailer model not found: {name}")
-    log(f"Loading aDetailer model {name}")
-    return YOLO(path)
-
-
-def ad_detect(image_np, name):
-    h, w = image_np.shape[:2]
-    pil = Image.fromarray(np.clip(image_np * 255, 0, 255).astype(np.uint8))
-    masks = []
-    yolo = load_ad_model(name)
-    results = yolo(pil, conf=AD_CONF, verbose=False)
-    for result in results:
-        if result.boxes is None:
-            continue
-        for box in result.boxes.xyxy.cpu().numpy():
-            x1 = max(0, int(box[0]) - AD_PADDING)
-            y1 = max(0, int(box[1]) - AD_PADDING)
-            x2 = min(w, int(box[2]) + AD_PADDING)
-            y2 = min(h, int(box[3]) + AD_PADDING)
-            mask = Image.new("L", (w, h), 0)
-            ImageDraw.Draw(mask).rectangle([x1, y1, x2, y2], fill=255)
-            mask = mask.filter(ImageFilter.GaussianBlur(AD_BLUR))
-            mask_np = np.array(mask).astype(np.float32) / 255.0
-            masks.append(torch.from_numpy(mask_np))
-    return masks
-
-
-def run_adetailer(image, model, clip, negative, sampler, scheduler, seed):
-    for name, prompt in AD_PASSES:
-        image_np = image[0].detach().numpy()
-        masks = ad_detect(image_np, name)
-        if not masks:
-            continue
-        positive = state["encode"].encode(clip=clip, text=prompt)[0]
-
-        mask = torch.stack(masks).amax(dim=0).unsqueeze(0)
-        latent = state["vae_encode"].encode(
-            vae=state["vae"],
-            pixels=image,
-        )[0]
-        latent["noise_mask"] = mask
-        samples = run_sampler(
-            model,
-            seed,
-            AD_STEPS,
-            AD_CFG,
-            sampler,
-            scheduler,
-            positive,
-            negative,
-            latent,
-            AD_DENOISE,
-        )
-        detailed = state["decode"].decode(
-            vae=state["vae"],
-            samples=samples,
-        )[0]
-        blend = mask.unsqueeze(-1).to(image)
-        image = image * (1 - blend) + detailed * blend
-
-    return image
-
-
 def stored_image_tensor(path):
     with Image.open(BytesIO(stored_bytes(path))) as image:
         pixels = np.asarray(image.convert("RGB"), dtype=np.float32) / 255
@@ -797,15 +697,6 @@ def generate_comparison_cell(request, output):
         vae=state["vae"],
         samples=samples,
     )[0]
-    image = run_adetailer(
-        image,
-        second_model,
-        second_clip,
-        negative,
-        request.sampler,
-        request.scheduler,
-        MATRIX_SEED,
-    )
     save_named_image(tensor_image(image), output)
     return request.output
 
@@ -902,15 +793,6 @@ def generate_matrix_cell_core(body):
             vae=state["vae"],
             samples=samples,
         )[0]
-        image = run_adetailer(
-            image,
-            model,
-            clip,
-            negative,
-            MATRIX_SAMPLER,
-            MATRIX_SCHEDULER,
-            MATRIX_SEED,
-        )
         save_named_image(tensor_image(image), output)
         return f"{first_id}-{second_id}"
 

requirements.txt

@@ -5,6 +5,5 @@ huggingface_hub>=1.6
 numpy
 Pillow
 py7zr==1.1.2
-ultralytics
 uvicorn
 -r https://raw.githubusercontent.com/Comfy-Org/ComfyUI/master/requirements.txt