Upload 4 files

Dockerfile

@@ -21,7 +21,7 @@ FROM python:3.10-slim
 
 WORKDIR /app
 
-RUN apt-get update && apt-get install -y --no-install-recommends libgl1 libglib2.0-0 wget git \
+RUN apt-get update && apt-get install -y --no-install-recommends libgl1 libglib2.0-0 wget \
     && rm -rf /var/lib/apt/lists/*
 
 COPY --from=tgbotapi-builder /usr/local/bin/telegram-bot-api /usr/local/bin/telegram-bot-api

app.py

@@ -9,7 +9,6 @@ import easyocr
 import torch
 import requests
 import vtracer
-import pydiffvg
 from PIL import Image
 from fastapi import FastAPI, File, UploadFile, HTTPException, Form
 from fastapi.middleware.cors import CORSMiddleware
@@ -21,8 +20,6 @@ from basicsr.archs.rrdbnet_arch import RRDBNet
 from realesrgan import RealESRGANer
 from gfpgan import GFPGANer
 
-pydiffvg.set_use_gpu(False)
-
 app = FastAPI(title="ZGrafic API")
 
 # Permite que tu página en GitHub Pages (u otro origen) llame a este servidor.
@@ -711,113 +708,3 @@ async def vectorize(file: UploadFile = File(...)):
             "X-Vectorize-Type": "ilustracion" if is_flat_illustration else "foto-detallada",
         },
     )
-
-
-# ---- /vectorize-ai (experimental: optimización tipo LIVE con pydiffvg) ----
-
-def random_path(num_segments, canvas_w, canvas_h, seed_point=None):
-    """Una forma cerrada de curvas Bézier, arrancando en un punto al
-    azar (o uno dado) — la optimización después la va a ir moviendo y
-    deformando hasta que se parezca a una región real de la imagen."""
-    num_control_points = torch.zeros(num_segments, dtype=torch.int32) + 2
-    points = []
-    p0 = seed_point if seed_point is not None else torch.rand(2) * torch.tensor([canvas_w, canvas_h])
-    points.append(p0)
-    radius = max(canvas_w, canvas_h) * 0.08
-    for _ in range(num_segments):
-        p1 = p0 + (torch.rand(2) - 0.5) * radius
-        p2 = p1 + (torch.rand(2) - 0.5) * radius
-        p3 = p2 + (torch.rand(2) - 0.5) * radius
-        points += [p1, p2, p3]
-        p0 = p3
-    points_t = torch.stack(points)
-    return pydiffvg.Path(
-        num_control_points=num_control_points,
-        points=points_t,
-        is_closed=True,
-        stroke_width=torch.tensor(0.0),
-    )
-
-
-def vectorize_with_live(pil_img, num_paths=80, num_iter=250):
-    """Versión simplificada del enfoque de LIVE: arranca con formas al
-    azar y las va ajustando con descenso de gradiente hasta que, todas
-    juntas, se parezcan a la imagen original."""
-    canvas_w, canvas_h = pil_img.size
-    target = torch.from_numpy(np.array(pil_img).astype(np.float32) / 255.0)
-    if target.shape[2] == 3:
-        alpha = torch.ones((canvas_h, canvas_w, 1))
-        target = torch.cat([target, alpha], dim=2)
-
-    # Colores de partida: tomamos colores reales de la imagen (en vez de
-    # colores al azar), para que arranque más cerca del resultado final.
-    small = pil_img.copy()
-    small.thumbnail((64, 64))
-    palette_source = np.array(small).reshape(-1, 3)
-
-    shapes, shape_groups = [], []
-    for i in range(num_paths):
-        seed_point = torch.rand(2) * torch.tensor([canvas_w, canvas_h])
-        path = random_path(num_segments=3, canvas_w=canvas_w, canvas_h=canvas_h, seed_point=seed_point)
-        shapes.append(path)
-
-        color_rgb = palette_source[np.random.randint(0, len(palette_source))] / 255.0
-        fill_color = torch.tensor([color_rgb[0], color_rgb[1], color_rgb[2], 1.0], requires_grad=True)
-        group = pydiffvg.ShapeGroup(
-            shape_ids=torch.tensor([i]),
-            fill_color=fill_color,
-        )
-        shape_groups.append(group)
-
-    points_vars = [s.points.requires_grad_() for s in shapes]
-    color_vars = [g.fill_color.requires_grad_() for g in shape_groups]
-
-    points_optim = torch.optim.Adam(points_vars, lr=1.0)
-    color_optim = torch.optim.Adam(color_vars, lr=0.02)
-
-    for _ in range(num_iter):
-        points_optim.zero_grad()
-        color_optim.zero_grad()
-
-        scene_args = pydiffvg.RenderFunction.serialize_scene(canvas_w, canvas_h, shapes, shape_groups)
-        img = pydiffvg.RenderFunction.apply(canvas_w, canvas_h, 2, 2, 0, None, *scene_args)
-
-        loss = (img - target).pow(2).mean()
-        loss.backward()
-
-        points_optim.step()
-        color_optim.step()
-
-        for g in shape_groups:
-            g.fill_color.data.clamp_(0.0, 1.0)
-
-    svg_path = tempfile.NamedTemporaryFile(suffix=".svg", delete=False).name
-    pydiffvg.save_svg(svg_path, canvas_w, canvas_h, shapes, shape_groups)
-    with open(svg_path, "r") as f:
-        return f.read()
-
-
-@app.post("/vectorize-ai")
-async def vectorize_ai(file: UploadFile = File(...)):
-    if not file.content_type or not file.content_type.startswith("image/"):
-        raise HTTPException(status_code=400, detail="El archivo debe ser una imagen.")
-
-    input_bytes = await file.read()
-
-    try:
-        pil_img = Image.open(io.BytesIO(input_bytes)).convert("RGB")
-
-        # Achicamos bastante: esto optimiza pixel por pixel en CPU, una
-        # imagen grande sería demasiado lenta.
-        MAX_SIDE = 200
-        w, h = pil_img.size
-        if max(w, h) > MAX_SIDE:
-            ratio = MAX_SIDE / max(w, h)
-            pil_img = pil_img.resize((int(w * ratio), int(h * ratio)), Image.LANCZOS)
-
-        svg_content = vectorize_with_live(pil_img)
-
-    except Exception as e:
-        raise HTTPException(status_code=500, detail=f"No se pudo vectorizar la imagen: {e}")
-
-    return Response(content=svg_content, media_type="image/svg+xml")

bot.py

@@ -27,7 +27,7 @@ BOT_TOKEN = os.environ.get("TELEGRAM_BOT_TOKEN", "")
 REMOVE_BG_URL = "https://zgrafic-remove-bg-api.hf.space/remove-bg"
 ENHANCE_URL = "https://zgrafic-remove-bg-api.hf.space/enhance"
 VECTORIZE_URL = "https://zgrafic-remove-bg-api.hf.space/vectorize"
-VECTORIZE_AI_URL = "https://zgrafic-remove-bg-api.hf.space/vectorize-ai"
+VECTORIZE_AI_URL = "https://zgrafic-enhance-photo-api.hf.space/vectorize-ai"
 
 FREE_USES_PER_DAY = 10
 STARS_PER_IMAGE = 10

requirements.txt

@@ -18,4 +18,3 @@ realesrgan==0.3.0
 gfpgan==1.3.8
 firebase-admin==6.5.0
 vtracer==0.6.15
-git+https://github.com/ANaka/easydiffvg.git