UI image
architecture diagram

Inspiration

Acid attacks remain a devastating, under-discussed issue affecting survivors in Pakistan and beyond. Most awareness campaigns require budgets and production skills that small NGOs and activists simply don't have. We wanted to build a tool that lets anyone turn a cause they care about into a real, watchable short film — no animation or filmmaking skills required.

What it does

Awareness Short Generator is an AI agent that automates the entire short-drama production pipeline: scriptwriting, scene breakdown, video generation, and final editing. A user enters a topic (and optionally, detailed scene descriptions), and the agent handles the rest — producing a finished animated short film with a consistent visual style throughout.

We used it to produce a short film honoring the resilience of acid attack survivors, following a woman's journey from a determined young student to a doctor whose life is changed by violence, and her strength afterward.

How we built it

Script generation — Qwen's language model (qwen3.7-plus) writes the narrative and breaks it into structured scenes
Video generation — Alibaba's Wan2.7 text-to-video model renders each scene, using a locked style descriptor and consistent seed value across all clips for visual coherence
Editing — MoviePy stitches all scenes together with smooth transitions into one final video
Interface — A Streamlit web app lets anyone use the tool without touching code, entering just a topic and optional scene details

Challenges we ran into

Balancing AI-generated automation with the need for a human-in-the-loop option, since fully automated scene-writing didn't give us the creative control this sensitive topic needed — we added a manual scene override so users can write their own detailed prompts when precision matters
Hit a free-tier quota lock on our Qwen Cloud account partway through building that blocked paid API calls despite having active hackathon credits — worked around it collaboratively as a team while pursuing a fix with support
Getting visual consistency across independently-generated video clips took real iteration — locking in identical style descriptors and seed values across scenes was key
Keeping AI-generated visuals respectful and non-graphic for a sensitive real-world topic, using symbolic rather than literal imagery

What we learned

How to architect a multi-step AI agent pipeline (not just single API calls), the tradeoffs between full automation and human control, and how much small technical decisions (like a shared seed value) affect the coherence of AI-generated video.