AstraGen AI: A Multimodal Generation Framework for Converting Text to Cinematic Videos in 60 Seconds

AstraGen AI is an end-to-end multimodal AI video generation framework built on FastAPI. It integrates the narrative capabilities of large language models (LLMs) and visual synthesis technology of diffusion models, enabling the conversion of text prompts into complete cinematic videos in 60 seconds. It achieves a fully automated process from script creation to final rendering, with no manual intervention required throughout.

Text-to-video generation is a challenging task in the field of generative AI, requiring maintenance of temporal coherence, narrative logic, and visual consistency. A single model is difficult to meet the demands; the industry consensus is to combine specialized models: using LLMs for narrative planning and diffusion models for visual generation. AstraGen AI is a practitioner of this approach.

AstraGen AI adopts a four-layer architecture:

1. Narrative Intelligence Layer: LLMs expand user prompts into structured storyboards, planning scenes and shot logic;
2. Visual Synthesis Layer: Calls diffusion model APIs to generate high-fidelity images for corresponding scenes;
3. Automatic Synthesis Layer: The MoviePy engine stitches images, adds transitions and subtitles, and generates MP4 files;
4. Service Layer: FastAPI provides high-performance web services, supporting zero local GPU dependency and fast response.

Tech Stack:

Layer	Technology/Tool	Purpose
Programming Language	Python 3.10+	Core development
Web Framework	FastAPI/Uvicorn	Backend service
Text Generation	OpenAI API/LLM API	Narrative creation
Image Generation	Pollinations AI	Scene visual synthesis
Video Rendering	MoviePy	Video export

Workflow:

1. Input creative prompt;
2. Automatically generate a script with 3 scenes;
3. Generate images for corresponding scenes;
4. Render and output MP4, taking 60 seconds in total.

AstraGen AI is suitable for:

• Rapid Prototyping: Video creators verify ideas and reduce upfront investment;
• Educational Demos: Convert abstract concepts into visual videos;
• Social Media: Quickly generate short video materials;
• Personal Entertainment: AI enthusiasts explore the possibilities of text-to-video.

Current limitations include:

• Static image stitching (not truly dynamic video);
• Lack of audio generation capability;
• Dependence on external APIs (requires network access and may incur costs);
• Limited narrative depth (fixed 3-scene structure). Improvement directions can focus on dynamic video generation, audio integration, reducing API dependence, etc.

Open Source Value: Provides a modular architecture reference, complete end-to-end implementation, and low-cost experimental platform to help developers learn multimodal system integration.

Conclusion: AstraGen AI represents a microcosm of the democratization of AI video generation. Although its quality is not as good as professional models, it demonstrates the potential of combining existing tools to build usable workflows, providing practical value to creators, developers, and researchers.