davidmosiah/creative-forge — explained in plain English
Analysis updated 2026-05-18
Generate and localize paid ad creatives for multiple markets using AI agents.
Run a fictional demo campaign end to end to see the full pipeline without real accounts.
Require verified proof from the ad platform before any ad is considered published.
Render short ad videos automatically using the bundled Remotion pipeline.
| davidmosiah/creative-forge | aclark4life/home-depot-crawl | aetheria-labs1/storefront-skills | |
|---|---|---|---|
| Stars | 6 | 6 | 6 |
| Language | Python | Python | Python |
| Last pushed | — | 2014-08-10 | — |
| Maintenance | — | Dormant | — |
| Setup difficulty | moderate | moderate | easy |
| Complexity | 4/5 | 2/5 | 3/5 |
| Audience | pm founder | developer | pm founder |
Figures from each repo's GitHub metadata at analysis time.
Full video rendering needs a Node.js and Remotion setup in addition to the Python package.
creative-forge is a system where AI agents handle the creative work behind paid advertising campaigns, such as research, writing ad copy, adapting creatives for different markets, and judging quality, while separate automated checks make sure nothing the agent claims is actually false. The core idea is that agent driven ad tools tend to say things like campaign published when nothing was actually published, or report numbers that do not exist. This project is built specifically to avoid that by never letting the agent's own word count as proof. The workflow moves through a fixed sequence of stages: checking readiness, doing research, building a brief, generating creative concepts, producing and localizing them for different languages and regions, rendering final images and video, running quality checks, and finally publishing ads, but only in a paused state that is never allowed to go live automatically. Every stage produces a file on disk that can be reviewed later, and every publish action requires a fresh confirmation from the actual advertising platform showing the ad is genuinely paused, not just a local record claiming so. Three rules run through the whole system. Every ad idea must be traceable back to real research. A local record can never substitute for checking the real state of things on the ad platform. And performance numbers are always reported honestly, using a clear missing data label instead of showing zero when a number is simply unavailable. The project ships with a fictional demo advertising app called sunrise-demo so the entire pipeline can be tried end to end using made up data, with no real accounts or credentials needed. It can be installed as a Python package, and a full checkout also supports rendering short videos using a tool called Remotion. The project is released under the AGPL-3.0 license.
An AI agent driven pipeline for building paid ad creatives that separates what agents claim from what is actually verified true before anything publishes.
Mainly Python. The stack also includes Python, Remotion, Node.js.
You can use and modify this freely, but if you run a modified version as a network service, you must also share your source code under the same license.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly pm founder.
This repo across BitVibe Labs
Verify against the repo before relying on details.