👻 GhostPack

Veo 3-Level Video Sorcery Haunts Laptops with 8GB RAM! 😈🚀

Unleash photorealistic videos with ghostai1’s math wizardry on GPUs ≥6GB VRAM.

GhostPack Preview

Preview: ghostgradio.py in action

Summon Now 🛠️ GitHub 🌟

✨ Spectral Features

📹 Veo 3-Level AI

Hyper-realistic videos via next-frame prediction, running on 8GB RAM laptops.

⚡️ Phantom Speed

Teacache cuts ~40% off generation time (~12–15s/frame on RTX 3060).

🛠️ Laptop Domination

13B models run on GPUs with ≥6GB VRAM, tuned for RTX 3060 and beyond.

🧙‍♂️ Ghostai1’s Math Sorcery

GhostPack’s SDK is a spectral masterpiece, conjured with ghostai1’s math optimizations to haunt any laptop with 8GB RAM.

🔮 Teacache

Caches diffusion states, slashing ~40% off frame time.

Boost: 40%
Stat: ~12s/frame on RTX 3060
Math: \( T_{\text{total}} \approx 0.6 \cdot T_{\text{base}} \)

🧙‍♂️ Sage-Attention

Streamlines attention layers for ~5–10% speed-up.

Boost: 5–10%
Stat: ~1–2s/frame saved
Math: \( T_{\text{attn}} \approx 0.9 \cdot T_{\text{base_attn}} \)

⚡ CUDA Tweaks

Optimized allocation for ~10–15% latency cut.

Boost: 10–15%
Stat: ~10–15% faster
Math: \( L_{\text{CUDA}} \approx 0.85 \cdot L_{\text{base}} \)

🧬 Context Packing

Compresses contexts, saving ~50% VRAM.

Boost: 50%
Stat: ~2–3GB saved
Math: \( M_{\text{VRAM}} \propto O(1) \)

💾 tcmalloc

Cuts memory overhead by ~5–20%.

Boost: 5–20%
Stat: ~15% CPU relief
Math: \( O_{\text{mem}} \approx 0.8 \cdot O_{\text{glibc}} \)

⚡ Dynamic Batching

Adapts batches for ~30–50% throughput gain.

Boost: 30–50%
Stat: ~1.5x FPS
Math: \( \text{Throughput} \propto B \cdot \text{FPS}_{\text{base}} \)

Optimization Breakdown

  • 🔮 Compressed Context Packing: Collapses frame contexts into a fixed-size matrix, slashing VRAM by ~50% (\( M_{\text{VRAM}} \propto O(1) \)), enabling 60s videos on 6GB VRAM GPUs.
  • 🧬 Dynamic Batching: Adapts batches (2–4 frames), boosting throughput by ~30–50% (\( \text{Throughput} \propto B \)), ideal for RTX 3050.
  • ⚡️ Teacache Efficiency: Caches diffusion states, cutting ~40% off compute time (\( T_{\text{total}} \approx 0.6T_{\text{base}} \)), hitting ~10–15s/frame on RTX 3060.
  • 🧙‍♂️ Sage-Attention: Streamlines attention layers, saving ~5–10% time (\( T_{\text{attn}} \approx 0.9T_{\text{base_attn}} \)).
  • 💾 tcmalloc: Reduces memory overhead by ~5–20% (\( O_{\text{mem}} \approx 0.8O_{\text{glibc}} \)), easing CPU load by ~15%.
  • ⚡ CUDA Tweaks: Cuts latency by ~10–15% (\( L_{\text{CUDA}} \approx 0.85L_{\text{base}} \)) with optimized allocation.

VRAM Requirements 🖥️

GPU VRAM Performance Notes
GTX 1650 6GB ~18–25s/frame Minimum, teacache recommended.
RTX 3050 8GB ~15–20s/frame Balanced, batch size 2.
RTX 3060 12GB ~10–15s/frame Optimal, batch size 2–4.
RTX 4090 24GB ~1.5–2.5s/frame High-end, max batch size.

🎥 Demo Video Showcase

Witness GhostPack’s spectral video generation in action!

Sample output from ghostgradio.py, showcasing a neon-lit spectral figure.

🌌 Future Spectral Phantoms

GhostPack is brewing epic upgrades:

🗣️ Voice Generation

AI-driven voice synthesis to narrate videos with ghostly tones, coming to ghostgradio.py.

🖼️ AI Images

Generate spectral starting frames with AI-crafted images for cinematic pipelines.

🛠️ Unleash GhostPack

Summon from GitHub. Requires >30GB disk space.

🐧 Ubuntu

git clone https://github.com/ghostai1/GhostPack
cd GhostPack
chmod +x install_ubuntu.sh
./install_ubuntu.sh

NVIDIA GPU, 8GB RAM, Python 3.10, CUDA 12.6.

🪟 Windows

git clone https://github.com/ghostai1/GhostPack
cd GhostPack
install.bat

Auto-downloads models (>30GB).

🍎 macOS

git clone https://github.com/ghostai1/GhostPack
cd GhostPack
chmod +x install_macos.sh
./install_macos.sh

M1/M2 support (slower).

📹 Weave Ghostly Videos

  1. Launch Phantom GUI: 
    source ~/ghostpack_venv/venv/bin/activate
cd ~/ghostpack_venv
python ghostgradio.py --port 5666 --server 0.0.0.0
  2. Craft Spells:
    • Upload an image.
    • Prompt: “A spectral figure glides through a neon-lit city.”
    • Enable teacache (⚡️ speed).
    • Set 15fps, 5s video (~75 frames).
  3. Summon Spirits: Run nvidia-smi (~80–100% GPU).

📂 Latest Conjurations

Explore the latest files and logs from the GhostPack repo.

📜 Latest File

Check out ghostgradio.py, the heart of GhostPack’s GUI.

View File

📋 Logs

Dive into render logs for debugging and insights.

View Logs
Refresh Repo 🔄 Refresh Logs 📥