Hugging Face's logo

julius119
/
verseflow_studio

Model card Files
xet
 Community
verseflow_studio
File size: 3,615 Bytes
8ab53f1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
<div align="center">

# 🎬 VerseFlow Studio 
### Next-Gen Enterprise Autonomous AI-Video Operating System & PaaS Infrastructure

[![License: Dual/NC](https://img.shields.io/badge/License-Apache_2.0_%2B_NC-red.svg)](LICENSE)
[![Python Version](https://img.shields.io/badge/Python-3.11%2B-blue.svg)](https://www.python.org/)
[![CUDA Version](https://img.shields.io/badge/CUDA-12.2%2B-green.svg)](https://developer.nvidia.com/)
[![Docker Support](https://img.shields.io/badge/Docker-Ready-blue.svg)](docker/docker-compose.yml)
[![Hugging Face Spaces](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-orange)](https://huggingface.co/)
[![Author](https://img.shields.io/badge/Author-XiaoZhe-purple.svg)](#-author--license-notice)

[Key Features](#-key-architectural-breakthroughs) •
[Quick Start](#-quick-start) •
[Architecture](#-system-architecture) •
[API & SDK](#-api--sdk-usage) •
[License](#-author--license-notice)

</div>

---

## 📜 版权与作者声明 (Author Declaration & Notice)

- **作者**：XiaoZhe (XiaoZhe AIGC工作室)
- **非商业用途**：允许免费学习、研究和非商业化使用，但**必须保留作者信息**（“XiaoZhe创作”）。
- **商业化使用**：**未经授权不得擅自商业化使用**。如需商业化部署、商用授权或项目定制，请联系作者：
  - **Email**: `janejulius119@gmail.com`
  - **WeChat**: `julius119`
  - **QQ**: `290187006`

---

## 💡 项目简介

**VerseFlow Studio** 是一套企业级 AI 视频操作系统。对比传统生成工具，在底层计算内核、状态管理、导演编排及工业化流转上完成了全面跃迁：


---

## 🌟 Key Architectural Breakthroughs

### 1. 🧠 vLLM-Inspired PagedVideoLatent Memory Pool
- Implements page-based allocation for 3D spatial-temporal Latent Tensors.
- Completely eliminates memory fragmentation during long-sequence video generation.
- Enables continuous batching for high-concurrency enterprise video requests.

### 2. ⚡ ComfyUI-Style Incremental Latent DAG Caching
- Hashes sub-graph nodes based on parameters and upstream dependencies.
- Re-renders only modified shots while reusing cached Latents for unchanged shots.
- Slashes iteration and trial costs by up to **90%**.

### 3. ⏸️ LangGraph-Driven Stateful HITL & Time-Travel Breakpoints
- Introduces Human-In-The-Loop (HITL) execution pause points at keyframes.
- Exposes mask restoration and canvas repainting controls to artists.
- Resumes execution seamlessly from saved state snapshots without restarting the DAG.

### 4. 🎯 Agentic Director & RLHF/DPO Self-Evolution Loop
- Automatically parses text scripts into camera movements, shot sizes, and lighting parameters.
- Ingests real-world video analytics (CTR, 5s retention, completion rate) to construct Direct Preference Optimization (DPO) datasets.
- Continuously fine-tunes director prompt strategies over time.

### 5. 🌐 Edge Heterogeneous Swarm & SaaS Infrastructure
- Features multi-tenant RBAC isolation and precise GPU-Second billing.
- Supports heterogeneous Edge rendering nodes (Vast.ai, RunPod, RTX 4090/Spot instances).
- Auto-handles SIGTERM preemptions with zero-data-loss checkpoint recovery.

---

## 🚀 快速启动

```bash
# 1. 克隆仓库
git clone [https://github.com/julius119/verseflow-studio.git](https://github.com/julius119/verseflow-studio.git)
cd verseflow-studio

# 2. Docker 容器化启动 (推荐)
docker compose -f docker/docker-compose.yml up -d

# 3. 运行端到端测试
python3 scripts/run_pipeline.py

# 4. 运行硬件性能基准测试
python3 scripts/benchmark.py