Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Multi-model DAG-driven parallel AI film generation — parallel speedup scales with scene independence; Generate film scenes simultaneously instead of one by one; "把影视生成的执行图从拓扑序变成关键路径最优调度" ; 唯一把场景叙事依赖建模为 DAG、以 CPM 算法驱动并行调度的影视生成引擎
| Date | Stars |
|---|---|
| 2026-07-24 | 645 |
| 2026-07-25 | 645 |
| 2026-07-28 | 645 |
| 2026-07-30 | 646 |
| 2026-08-07 | 647 |
| 2026-08-15 | 580 |
| 2026-08-18 | 534 |
| 2026-08-22 | 533 |
| 2026-08-23 | 534 |
| 2026-08-24 | 533 |
| 2026-08-25 | 534 |
| 2026-08-26 | 535 |
| 2026-08-28 | 536 |
| 2026-08-29 | 536 |
| 2026-09-01 | 535 |
| 2026-09-06 | 536 |
| 2026-09-09 | 537 |
| 2026-09-11 | 537 |
| 2026-09-12 | 538 |
| 2026-09-14 | 538 |
| 2026-09-15 | 537 |
| 2026-09-20 | 537 |
Today
— stars today
This week
-1 stars this week
This month
+3 stars this month
Momentum
0.0
growth rate 0.00%/day
<div align="center">
# 🎬 Forge
**One story, multiple AI models, zero manual stitching.**
[](https://github.com/F-R-L/forge-film/actions)
[](https://www.python.org)
[](LICENSE)
[](https://github.com/F-R-L/forge-film)
[中文文档](README.zh.md) | [Quick Start](#-quickstart)
</div>
---
Making a multi-scene AI film means logging into Kling, CogVideoX, Seedance separately — downloading frames, color-correcting between models, stitching manually. An 8-scene short can eat half a day.
**Forge automates the entire pipeline.** You write a story. Forge compiles it into a scene graph, routes each scene to the right model, runs them in parallel, keeps visual continuity across model boundaries, and outputs a single `final.mp4`.
---
## What Forge does
🧭 **Story → DAG** — GPT-4o (or Claude / DeepSeek) compiles your story into a dependency graph. Scenes with no dependencies run in parallel.
⚡ **CPM parallel scheduling** — Critical Path Method finds the longest dependency chain and prioritizes it. N workers run simultaneously, not one by one.
🎯 **Scene-type routing** — dialogue → Kling, landscapes → CogVideoX (free, local), action → Kling. Fully configurable in `forge.yaml`.
🎨 **Cross-model continuity** — when scene B (CogVideoX) follows scene A (Kling), Forge extracts A's last frame, applies histogram color matching, and passes it as the i2v seed. No jarring cuts.
🎬 **Final assembly** — once all scenes are generated, clips are concatenated in a single ffmpeg pass. Normalized resolution and frame rate. Outputs `final.mp4`.
---
## End-to-end walkthrough
Given this story (`examples/detective.txt`):
```
A weary private detective takes on a missing-person case. His client is an anxious middle-aged woman whose husband vanished without a trace three days ago.
The detective searches the husband's office and finds an unsent letter and a basement key.
Meanwhile, a mysterious man begins tailing the detective through the streets.
The detective locates the basement and discovers a secret ledger the missing husband had hidden away.
The mysterious man suddenly appears — a confrontation erupts.
The detective delivers the secret ledger to the police, and the truth finally comes to light.
```
Forge compiles it into a DAG and schedules:
```
forge plan examples/detective.txt --scenes 6
Plan: 6 scenes
DAG: {'S1': ['S2', 'S3'], 'S2': ['S4'], 'S3': ['S5'], 'S4': ['S6'], 'S5': ['S6'], 'S6': []}
Routing:
S1 dialogue → kling_light
S2 action → kling_heavy
S3 landscape → cogvideo
S4 action → kling_heavy
S5 dialogue → kling_light
S6 dialogue → kling_light
Critical path: S1 → S2 → S4 → S6 (longest chain)
Estimated time: 20 min parallel vs 30 min serial
```
Then runs:
```
forge run examples/detective.txt --workers 4
[00:00] S1 started (kling_light)
[00:00] S3 started (cogvideo) ← parallel
[05:00] S1 done → S2 unlocked
[05:00] S2 started (kling_heavy)
[07:00] S3 done → S3→S2 color calibration applied
[10:00] S2 done → S4 unlocked
[10:00] S4 started (kling_heavy)
...
[20:00] S6 done → assembling final.mp4
Done. Output: ./output/final.mp4
```
---
## Parallel scheduling
Forge uses the **Critical Path Method (CPM)** to find the longest dependency chain in your scene DAG and prioritizes those scenes first. Scenes with no blocking dependencies start immediately.
In the example above: S1→S2→S4→S6 is the critical path. With 4 workers, S1 and S3 launch simultaneously at t=0. The total wall time drops from 30 min (serial) to 20 min.
Speedup scales with scene independence — a story where half the scenes are parallel will run roughly 2× faster.
```mermaid
gantt
title Without Forge — 6 scenes × 5 min Excerpt of 11,437 characters
Read on GitHub
38
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:f18fc9ce68d6c801, topic:video-generation, topic:text-to-video
matched fp:f18fc9ce68d6c801, topic:gradio
matched fp:f18fc9ce68d6c801, topic:workflow-automation