Upload 6 files

.dockerignore

@@ -0,0 +1,15 @@
+__pycache__/
+*.pyc
+*.pyo
+*.pyd
+.Python
+*.so
+*.egg
+*.egg-info/
+dist/
+build/
+.git/
+.gitignore
+outputs/*.mp4
+outputs/*.jpg
+.DS_Store

Dockerfile

@@ -1,24 +1,44 @@
-FROM python:3.11-slim
+FROM nvidia/cuda:12.4.1-cudnn-devel-ubuntu22.04
 
 ENV DEBIAN_FRONTEND=noninteractive
 ENV PYTHONUNBUFFERED=1
-ENV HF_HOME=/tmp/huggingface
-ENV TRANSFORMERS_CACHE=/tmp/huggingface
-ENV TORCH_HOME=/tmp/torch
+ENV HF_HOME=/app/hf_cache
+ENV CUDA_HOME=/usr/local/cuda
+ENV PATH=${CUDA_HOME}/bin:${PATH}
+ENV LD_LIBRARY_PATH=${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}
 
-RUN apt-get update && apt-get install -y --no-install-recommends \
-    build-essential git curl wget libgomp1 \
+# Install system dependencies
+RUN apt-get update && apt-get install -y \
+    python3.11 python3.11-pip python3.11-venv \
+    git wget curl ffmpeg libsm6 libxext6 \
+    libgl1 libglib2.0-0 \
     && rm -rf /var/lib/apt/lists/*
 
+RUN update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.11 1
+RUN update-alternatives --install /usr/bin/python python /usr/bin/python3.11 1
+
 WORKDIR /app
 
-RUN pip install --no-cache-dir torch==2.5.1 --index-url https://download.pytorch.org/whl/cu121
-RUN pip install --no-cache-dir transformers==4.46.0 accelerate==1.0.0 bitsandbytes==0.44.0
-RUN pip install --no-cache-dir protobuf sentencepiece
-RUN pip install --no-cache-dir flask==3.0.3 flask-cors==4.0.0 requests==2.31.0
+# Upgrade pip
+RUN python -m pip install --no-cache-dir --upgrade pip setuptools wheel
+
+# Install PyTorch with CUDA
+RUN pip install --no-cache-dir torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124
+
+# Copy requirements and install
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
 
-COPY server.py .
-COPY chat.html .
+# Pre-download model components (optional - can also download at runtime)
+# This speeds up first run but increases image size
+# RUN python -c "from diffusers import AutoencoderKLWan, WanImageToVideoPipeline; from transformers import CLIPVisionModel; print('Diffusers ready')"
+
+# Copy app code
+COPY . .
+
+# Create directories
+RUN mkdir -p /app/outputs /app/hf_cache /app/static
 
 EXPOSE 7860
-CMD ["python3", "server.py"]
+
+CMD ["python", "-m", "uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860", "--workers", "1"]

README.md

@@ -1,28 +1,78 @@
----
-title: Titan Group Chat
-emoji: 🧠
-colorFrom: indigo
-colorTo: purple
-sdk: docker
-app_port: 7860
----
-
-# Titan Group Chat
-
-نظام دردشة جماعية ذكي بـ 3 نماذج AI (Favor, Brain, Cell).
-
-## المميزات
-- 3 نماذج AI مختلفة
-- أفكار داخلية (Internal Thoughts)
-- حالات داخلية (Internal State)
-- صلاحيات (Permissions)
-- مؤشر "يكتب الآن"
-- ذاكرة طويلة المدى (Turso DB)
+# Wan 2.1 I2V 14B - Image to Video Generator
+
+أداة تحويل الصور إلى فيديو باستخدام نموذج **Wan 2.1 I2V 14B 720P** على NVIDIA A100.
+
+## المتطلبات
+
+- NVIDIA GPU مع دعم CUDA (موصى: A100 80GB)
+- Docker + Docker Compose
+- NVIDIA Docker Runtime
 
 ## التشغيل
-1. اضبط Secrets:
-   - `HF_TOKEN`
-   - `TURSO_DB_URL`
-   - `TURSO_DB_TOKEN`
-   - `ACCESS_PASSWORD`
-2. Factory Rebuild
+
+### 1. تعديل المفتاح السري
+
+عدّل ملف `.env` أو اضبط متغير البيئة `PASS_KEY`:
+
+```bash
+PASS_KEY=your-secret-key-here
+```
+
+### 2. البناء والتشغيل
+
+```bash
+cd wan-i2v-tool
+
+# باستخدام Docker Compose (موصى)
+docker-compose up --build -d
+
+# أو باستخدام Docker مباشرة
+docker build -t wan-i2v .
+docker run -d --gpus all -p 7860:7860   -e PASS_KEY=your-secret-key   -e MODEL_ID=Wan-AI/Wan2.1-I2V-14B-720P-Diffusers   -v $(pwd)/outputs:/app/outputs   -v hf_cache:/app/hf_cache   --shm-size=16gb   wan-i2v
+```
+
+### 3. الوصول
+
+افتح المتصفح على: `http://localhost:7860`
+
+أدخل مفتاح المرور في صفحة `pass.html` للوصول إلى الأداة.
+
+## الميزات
+
+- ✅ **Wan 2.1 I2V 14B 720P** - أحدث نموذج لتحريك الصور
+- ✅ **جودة عالية** - Negative prompt محسّن ضد التشوهات
+- ✅ **A100 محسّن** - torch.compile + VAE slicing/tiling
+- ✅ **واجهة عربية** - Dark mode, mobile first, responsive
+- ✅ **مصادقة آمنة** - مفتاح سري في `.env`
+- ✅ **لا قيود** - Safety: Middle, No restrictions
+- ✅ **تحكم كامل** - Resolution, FPS, Frames, Guidance, Steps, Seed
+- ✅ **معاينة مباشرة** - تشغيل الفيديو في المتصفح
+- ✅ **حفظ/تحميل** - زر حفظ الفيديو
+- ✅ **تتبع المهام** - قائمة المهام مع التقدم الحي
+
+## API Endpoints
+
+| Endpoint | Method | Description |
+|----------|--------|-------------|
+| `/api/auth` | POST | التحقق من المفتاح |
+| `/api/generate` | POST | توليد فيديو |
+| `/api/status/{job_id}` | GET | حالة المهمة |
+| `/api/download/{job_id}` | GET | تحميل الفيديو |
+| `/api/jobs` | GET | قائمة المهام |
+| `/api/jobs/{job_id}` | DELETE | حذف مهمة |
+| `/health` | GET | حالة النظام |
+
+## الإعدادات الافتراضية
+
+- **الدقة:** 720×1280
+- **الإطارات:** 81 (~10 ثواني بـ 8fps)
+- **خطوات الاستنتاج:** 30
+- **Guidance Scale:** 5.0
+- **Negative Prompt:** محسّن تلقائياً ضد التشوهات
+
+## ملاحظات
+
+- أول تشغيل يحمل النموذج من HuggingFace (~30GB)
+- يُفضّل كتابة البرومبت بالإنجليزية
+- وقت التوليد: 3-8 دقائق حسب الإعدادات
+- يدعم رفع صور JPG/PNG/WEBP

app.py

@@ -0,0 +1,409 @@
+#!/usr/bin/env python3
+"""
+Wan 2.1 I2V 14B - Image to Video Generator
+Backend: FastAPI + Diffusers
+Optimized for NVIDIA A100 80GB
+"""
+
+import os
+import io
+import base64
+import uuid
+import json
+import asyncio
+import threading
+from datetime import datetime
+from typing import Optional
+from contextlib import asynccontextmanager
+
+import torch
+import numpy as np
+from PIL import Image
+from fastapi import FastAPI, File, UploadFile, Form, HTTPException, Request
+from fastapi.responses import FileResponse, JSONResponse, HTMLResponse
+from fastapi.staticfiles import StaticFiles
+from fastapi.middleware.cors import CORSMiddleware
+from pydantic import BaseModel
+from dotenv import load_dotenv
+
+# Diffusers imports
+from diffusers import AutoencoderKLWan, WanImageToVideoPipeline
+from diffusers.utils import export_to_video, load_image
+from transformers import CLIPVisionModel
+
+# Load env
+load_dotenv()
+
+# Config
+PASS_KEY = os.getenv("PASS_KEY", "default-key-change-me")
+MODEL_ID = os.getenv("MODEL_ID", "Wan-AI/Wan2.1-I2V-14B-720P-Diffusers")
+DEVICE = os.getenv("DEVICE", "cuda")
+TORCH_DTYPE = torch.bfloat16 if os.getenv("TORCH_DTYPE", "bfloat16") == "bfloat16" else torch.float16
+
+# Paths
+OUTPUT_DIR = "/app/outputs"
+STATIC_DIR = "/app/static"
+os.makedirs(OUTPUT_DIR, exist_ok=True)
+os.makedirs(STATIC_DIR, exist_ok=True)
+
+# Global pipeline
+pipe = None
+pipe_lock = threading.Lock()
+
+# Active jobs
+active_jobs = {}
+
+# ============================================================
+# MODEL LOADING
+# ============================================================
+def load_model():
+    """Load Wan 2.1 I2V 14B 720P model."""
+    global pipe
+    print(f"[INIT] Loading model: {MODEL_ID}")
+    print(f"[INIT] Device: {DEVICE}, Dtype: {TORCH_DTYPE}")
+
+    # Load components
+    image_encoder = CLIPVisionModel.from_pretrained(
+        MODEL_ID, 
+        subfolder="image_encoder", 
+        torch_dtype=torch.float32
+    )
+
+    vae = AutoencoderKLWan.from_pretrained(
+        MODEL_ID, 
+        subfolder="vae", 
+        torch_dtype=torch.float32
+    )
+
+    pipe = WanImageToVideoPipeline.from_pretrained(
+        MODEL_ID,
+        vae=vae,
+        image_encoder=image_encoder,
+        torch_dtype=TORCH_DTYPE,
+    )
+
+    pipe.to(DEVICE)
+
+    # Optimizations for A100
+    if hasattr(pipe, 'enable_vae_slicing'):
+        pipe.enable_vae_slicing()
+    if hasattr(pipe, 'enable_vae_tiling'):
+        pipe.enable_vae_tiling()
+
+    # Optional: torch.compile for transformer (A100 benefits greatly)
+    try:
+        if hasattr(pipe, 'transformer'):
+            pipe.transformer = torch.compile(pipe.transformer, mode="max-autotune", fullgraph=False)
+            print("[INIT] torch.compile applied to transformer")
+    except Exception as e:
+        print(f"[INIT] torch.compile skipped: {e}")
+
+    print("[INIT] Model loaded successfully")
+    return pipe
+
+
+# ============================================================
+# LIFESPAN
+# ============================================================
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    print("[STARTUP] Initializing Wan 2.1 I2V Generator...")
+    load_model()
+    print("[STARTUP] Ready")
+    yield
+    print("[SHUTDOWN] Cleaning up...")
+
+
+app = FastAPI(
+    title="Wan 2.1 I2V Generator",
+    description="Image to Video using Wan 2.1 14B 720P",
+    version="2.1.0",
+    lifespan=lifespan
+)
+
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+# Static files
+app.mount("/static", StaticFiles(directory=STATIC_DIR), name="static")
+
+
+# ============================================================
+# AUTH
+# ============================================================
+class AuthRequest(BaseModel):
+    pass_key: str
+
+@app.post("/api/auth")
+async def authenticate(data: AuthRequest):
+    if data.pass_key != PASS_KEY:
+        raise HTTPException(status_code=401, detail="Invalid pass key")
+    return {"status": "ok", "token": PASS_KEY}
+
+
+# ============================================================
+# GENERATION
+# ============================================================
+class GenerationRequest(BaseModel):
+    prompt: str
+    negative_prompt: Optional[str] = None
+    height: int = 720
+    width: int = 1280
+    num_frames: int = 81
+    guidance_scale: float = 5.0
+    num_inference_steps: int = 30
+    fps: int = 8
+    seed: Optional[int] = None
+
+
+def aspect_ratio_resize(image, max_area=720*1280, mod_value=32):
+    """Smart resize maintaining aspect ratio."""
+    aspect_ratio = image.height / image.width
+    height = round(np.sqrt(max_area * aspect_ratio)) // mod_value * mod_value
+    width = round(np.sqrt(max_area / aspect_ratio)) // mod_value * mod_value
+    if height < 256: height = 256
+    if width < 256: width = 256
+    image = image.resize((width, height), Image.LANCZOS)
+    return image, height, width
+
+
+def generate_video_task(job_id: str, image_path: str, params: dict):
+    """Background video generation task."""
+    try:
+        active_jobs[job_id] = {"status": "processing", "progress": 0, "message": "Loading image..."}
+
+        # Load and preprocess image
+        image = Image.open(image_path).convert("RGB")
+        max_area = params.get("height", 720) * params.get("width", 1280)
+        image, height, width = aspect_ratio_resize(image, max_area=max_area)
+
+        active_jobs[job_id]["message"] = f"Generating {height}x{width} video..."
+        active_jobs[job_id]["progress"] = 10
+
+        # Default negative prompt optimized for quality
+        default_negative = (
+            "Bright tones, overexposed, static, blurred details, subtitles, style, works, paintings, "
+            "images, static, overall gray, worst quality, low quality, JPEG compression residue, ugly, "
+            "incomplete, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, "
+            "misshapen limbs, fused fingers, still picture, messy background, three legs, many people "
+            "in the background, walking backwards, distorted face, mutated hands, extra limbs, "
+            "malformed anatomy, bad anatomy, watermark, text, logo, cropped, out of frame"
+        )
+
+        negative_prompt = params.get("negative_prompt", "") or default_negative
+
+        # Seed
+        seed = params.get("seed")
+        generator = torch.Generator(device=DEVICE).manual_seed(seed) if seed else None
+
+        # Progress callback
+        total_steps = params.get("num_inference_steps", 30)
+        def progress_callback(step, timestep, latents):
+            progress = int((step / total_steps) * 80) + 10
+            active_jobs[job_id]["progress"] = min(progress, 90)
+            active_jobs[job_id]["message"] = f"Denoising step {step}/{total_steps}..."
+
+        active_jobs[job_id]["message"] = "Running inference (this may take 3-8 minutes)..."
+
+        with pipe_lock:
+            output = pipe(
+                image=image,
+                prompt=params["prompt"],
+                negative_prompt=negative_prompt,
+                height=height,
+                width=width,
+                num_frames=params.get("num_frames", 81),
+                guidance_scale=params.get("guidance_scale", 5.0),
+                num_inference_steps=total_steps,
+                generator=generator,
+                callback_on_step_end=progress_callback,
+            ).frames[0]
+
+        # Export video
+        fps = params.get("fps", 8)
+        output_filename = f"{job_id}.mp4"
+        output_path = os.path.join(OUTPUT_DIR, output_filename)
+        export_to_video(output, output_path, fps=fps)
+
+        active_jobs[job_id] = {
+            "status": "completed",
+            "progress": 100,
+            "message": "Video generated successfully",
+            "output_file": output_filename,
+            "output_path": output_path,
+            "height": height,
+            "width": width,
+            "fps": fps,
+            "duration": round(params.get("num_frames", 81) / fps, 1),
+            "created_at": datetime.now().isoformat()
+        }
+
+    except Exception as e:
+        active_jobs[job_id] = {
+            "status": "failed",
+            "progress": 0,
+            "message": str(e)
+        }
+        print(f"[ERROR] Job {job_id} failed: {e}")
+
+
+@app.post("/api/generate")
+async def generate_video(
+    request: Request,
+    image: UploadFile = File(...),
+    prompt: str = Form(...),
+    negative_prompt: Optional[str] = Form(None),
+    height: int = Form(720),
+    width: int = Form(1280),
+    num_frames: int = Form(81),
+    guidance_scale: float = Form(5.0),
+    num_inference_steps: int = Form(30),
+    fps: int = Form(8),
+    seed: Optional[int] = Form(None),
+    pass_key: str = Form(...)
+):
+    # Auth check
+    if pass_key != PASS_KEY:
+        raise HTTPException(status_code=401, detail="Unauthorized")
+
+    # Validate image
+    if not image.content_type or not image.content_type.startswith("image/"):
+        raise HTTPException(status_code=400, detail="File must be an image")
+
+    # Clamp values for safety
+    height = max(256, min(1280, height))
+    width = max(256, min(1920, width))
+    num_frames = max(16, min(161, num_frames))
+    guidance_scale = max(1.0, min(20.0, guidance_scale))
+    num_inference_steps = max(10, min(100, num_inference_steps))
+    fps = max(1, min(60, fps))
+
+    job_id = str(uuid.uuid4())
+
+    # Save uploaded image
+    image_bytes = await image.read()
+    image_path = os.path.join(OUTPUT_DIR, f"{job_id}_input.jpg")
+    with open(image_path, "wb") as f:
+        f.write(image_bytes)
+
+    params = {
+        "prompt": prompt,
+        "negative_prompt": negative_prompt,
+        "height": height,
+        "width": width,
+        "num_frames": num_frames,
+        "guidance_scale": guidance_scale,
+        "num_inference_steps": num_inference_steps,
+        "fps": fps,
+        "seed": seed,
+    }
+
+    # Start background task
+    active_jobs[job_id] = {"status": "queued", "progress": 0, "message": "Queued..."}
+    thread = threading.Thread(target=generate_video_task, args=(job_id, image_path, params))
+    thread.start()
+
+    return {"job_id": job_id, "status": "queued"}
+
+
+@app.get("/api/status/{job_id}")
+async def get_status(job_id: str, pass_key: str):
+    if pass_key != PASS_KEY:
+        raise HTTPException(status_code=401, detail="Unauthorized")
+
+    if job_id not in active_jobs:
+        raise HTTPException(status_code=404, detail="Job not found")
+
+    return active_jobs[job_id]
+
+
+@app.get("/api/download/{job_id}")
+async def download_video(job_id: str, pass_key: str):
+    if pass_key != PASS_KEY:
+        raise HTTPException(status_code=401, detail="Unauthorized")
+
+    if job_id not in active_jobs or active_jobs[job_id].get("status") != "completed":
+        raise HTTPException(status_code=404, detail="Video not ready")
+
+    output_path = active_jobs[job_id]["output_path"]
+    if not os.path.exists(output_path):
+        raise HTTPException(status_code=404, detail="File not found")
+
+    return FileResponse(
+        output_path,
+        media_type="video/mp4",
+        filename=f"wan_i2v_{job_id}.mp4"
+    )
+
+
+@app.get("/api/jobs")
+async def list_jobs(pass_key: str):
+    if pass_key != PASS_KEY:
+        raise HTTPException(status_code=401, detail="Unauthorized")
+
+    jobs = []
+    for jid, info in active_jobs.items():
+        jobs.append({
+            "job_id": jid,
+            "status": info.get("status"),
+            "progress": info.get("progress"),
+            "message": info.get("message"),
+            "created_at": info.get("created_at")
+        })
+    return {"jobs": sorted(jobs, key=lambda x: x.get("created_at", ""), reverse=True)}
+
+
+@app.delete("/api/jobs/{job_id}")
+async def delete_job(job_id: str, pass_key: str):
+    if pass_key != PASS_KEY:
+        raise HTTPException(status_code=401, detail="Unauthorized")
+
+    if job_id in active_jobs:
+        info = active_jobs[job_id]
+        # Cleanup files
+        for ext in ["_input.jpg", ".mp4"]:
+            fpath = os.path.join(OUTPUT_DIR, f"{job_id}{ext}")
+            if os.path.exists(fpath):
+                os.remove(fpath)
+        del active_jobs[job_id]
+
+    return {"status": "deleted"}
+
+
+# ============================================================
+# ROOT & PAGES
+# ============================================================
+@app.get("/", response_class=HTMLResponse)
+async def root():
+    with open(os.path.join(STATIC_DIR, "pass.html"), "r", encoding="utf-8") as f:
+        return HTMLResponse(content=f.read())
+
+
+@app.get("/app", response_class=HTMLResponse)
+async def app_page():
+    with open(os.path.join(STATIC_DIR, "index.html"), "r", encoding="utf-8") as f:
+        return HTMLResponse(content=f.read())
+
+
+@app.get("/health")
+async def health():
+    return {
+        "status": "ok",
+        "model_loaded": pipe is not None,
+        "model_id": MODEL_ID,
+        "device": str(DEVICE),
+        "torch_version": torch.__version__,
+        "cuda_available": torch.cuda.is_available(),
+        "cuda_device": torch.cuda.get_device_name(0) if torch.cuda.is_available() else None,
+        "vram_gb": round(torch.cuda.get_device_properties(0).total_memory / 1024**3, 1) if torch.cuda.is_available() else 0
+    }
+
+
+if __name__ == "__main__":
+    import uvicorn
+    uvicorn.run(app, host="0.0.0.0", port=7860)

docker-compose.yml

@@ -1,16 +1,22 @@
-version: "3.8"
+version: '3.8'
 
 services:
-  titan-chat:
+  wan-i2v:
     build: .
-    container_name: titan-chat
+    image: wan-i2v:latest
+    container_name: wan-i2v-generator
+    runtime: nvidia
+    environment:
+      - NVIDIA_VISIBLE_DEVICES=all
+      - PASS_KEY=${PASS_KEY:-wan2i2v-secret-2026}
+      - MODEL_ID=Wan-AI/Wan2.1-I2V-14B-720P-Diffusers
+      - TORCH_DTYPE=bfloat16
+      - HF_HOME=/app/hf_cache
+    volumes:
+      - ./outputs:/app/outputs
+      - hf_cache:/app/hf_cache
     ports:
       - "7860:7860"
-    environment:
-      - HF_TOKEN=${HF_TOKEN}
-      - TURSO_DB_URL=${TURSO_DB_URL}
-      - TURSO_DB_TOKEN=${TURSO_DB_TOKEN}
-      - ACCESS_PASSWORD=${ACCESS_PASSWORD}
     deploy:
       resources:
         reservations:
@@ -18,9 +24,8 @@ services:
             - driver: nvidia
               count: 1
               capabilities: [gpu]
-    volumes:
-      - huggingface-cache:/tmp/huggingface
     restart: unless-stopped
+    shm_size: '16gb'
 
 volumes:
-  huggingface-cache:
+  hf_cache:

requirements.txt

@@ -1,9 +1,12 @@
-torch==2.5.1
-transformers==4.46.0
-accelerate==1.0.0
-bitsandbytes==0.44.0
-protobuf
-sentencepiece
-flask==3.0.3
-flask-cors==4.0.0
-requests==2.31.0
+torch>=2.6.0
+torchvision>=0.21.0
+diffusers>=0.33.0
+transformers>=4.49.0
+accelerate>=1.6.0
+fastapi>=0.115.0
+uvicorn[standard]>=0.34.0
+python-multipart>=0.0.20
+pillow>=11.0.0
+numpy>=2.0.0
+python-dotenv>=1.0.0
+aiofiles>=24.0.0