GML-MMGroup/AdCraft

AdCraft — 首个面向营销广告的 Agentic 视频制作平台。无限画布 · 拉片复刻 · 全流程自动化，从创意到成片一站直达。AdCraft — The first agentic video production platform for marketing and advertising. Infinite canvas · shot-by-shot replication · fully automated, from idea to final cut.

🎬 AdCraft – An Agentic Video‑Ad Production Platform

What it is – AdCraft is an open‑source AI‑powered system that turns a short textual brief (product, selling points, style, length, platform) into a full‑featured advertising video. It does this by chaining specialised AI agents that each handle a stage of the creative pipeline – from concept planning all the way to final compositing – while keeping every intermediate asset in a unified, version‑controlled library.

Why it matters – Traditional ad creation requires a team of writers, designers, editors, and a lot of back‑and‑forth. AdCraft automates that workflow, yet remains editable: you can intervene at any step, ask the agents to revise a single character, a shot, or the background music, and the system will regenerate only the affected pieces.

Core capabilities (as described in the README)

Feature	What you get
End‑to‑end ad generation	Input a brief → script → product visuals → characters → scenes → storyboard → assets → editing → final cut.
Editable workflow	Add, remove, reorder, or re‑run any stage; assets you like are preserved.
Conversational control	Chat with the AI (e.g., “revise the third shot”) and the system understands what and how to change.
Specialised AI agents	Creative Director, Scriptwriter, Product Visual, Character Designer, Scene Designer, Storyboard Designer, Video Director, Sound Director, Final Compositor – each runs its own model and passes results downstream.
Unified asset library	All generated/uploaded images, video clips, audio, and metadata live in a single library with version history, source tags, and reuse flags.
Reusable character & scene libraries	Pre‑built, production‑ready character turn‑arounds and scene world‑building storyboards that can be dropped into any project.
Fine‑grained editing & rollback	Edit any element in isolation; candidate versions appear for preview before acceptance.
Real‑time progress UI	Canvas shows the exact status of each node (queued, generating, awaiting confirmation, etc.).
Final compositing timeline	Drag‑and‑drop clips, trim, edit subtitles, tweak audio, then render the finished ad.
Video cloning	Upload a reference commercial, have the system deconstruct its shot‑by‑shot structure, and rebuild a new ad with your own product/characters while preserving the original rhythm and pacing.
Preset creative flows	Ready‑made templates for common ad types (product showcase, game UA, e‑commerce seeding, …).
Multi‑model support	Plug‑in dozens of image, video, and audio generation services (e.g., Jimeng, Doubao, Tencent Hunyuan, MiniMax, CosyVoice, etc.) and configure them centrally.

Typical use‑case workflow

Provide a brief – product name, key benefits, desired style, length, target platform.
AI Creative Director creates a high‑level plan and orchestrates downstream agents.
Scriptwriter drafts copy and subtitles.
Product Visual generates product shots; Character Designer builds brand‑aligned avatars; Scene Designer assembles environments.
Storyboard Designer maps shots; Video Director turns storyboard frames into short clips; Sound Director adds music/SFX.
Final Compositor assembles everything on a timeline.
User reviews, tweaks via chat or canvas edits, accepts candidates, and renders the final video.

Getting started (as per the repo)

Follow the Getting Started doc for installation steps.
Choose a deployment method: native (no Docker) or Docker‑based.
Download the recommended assets release (character & scene catalogs) and place the extracted folder under runtime-data/api/assets/catalogs/recommended/v1.0.0/ as instructed.
Verify the download with the provided SHA‑256 checksum.
Run the backend and frontend services; the UI presents an infinite canvas where you can type your brief and watch the pipeline execute.

Who might benefit?

Marketing teams that need to produce many short ads quickly while retaining brand consistency.
Creative agencies looking for a rapid prototyping tool to explore concepts before committing to full production.
Start‑ups / SMEs without a large in‑house video crew, but with access to generative AI models.
Developers & researchers interested in building or extending multi‑agent creative workflows.

Contribution & contact

The project welcomes pull requests for new features, bug fixes, or additional asset packs. Contact the maintainers (Ma Fei, Xu Hongbo) via the listed email addresses for questions or collaborations.

TL;DR – AdCraft is a full‑stack, agent‑driven platform that lets you describe an ad in plain language and automatically builds a complete, editable video production pipeline, with a focus on modular assets, conversational refinement, and support for many generative AI models.