elder-plinius/i-llm

1. What this repository is

- I-LLM is an interactive AI “movie maker” workflow built around Streamlit.

- The core use case is generating a movie concept, style guide, outline, screenplay, and voice/dialogue assets from user prompts.

2. Primary runtime entrypoint

- app.py is the documented main application entrypoint (streamlit run app.py).

- The extra variants (appv2.py, appv3.py, appv4.py, appdalle.py, movie_maker_app.py) suggest iterative UI/prototype versions rather than separate services.

3. Presentation layer

- The Streamlit app is the user-facing boundary.

- It likely gathers creative inputs from the user, displays generated text/assets, and coordinates the multi-step creation flow.

- screen.md likely documents or captures the UI/output flow.

4. Orchestration / workflow layer

- agents.py is the main sign that the app uses a higher-level orchestration layer rather than calling models directly from the UI.

- This layer likely sequences tasks such as concept generation, style development, outline writing, script drafting, and dialogue/voice preparation.

- prompts.py appears to hold the prompt templates that define those generation steps.

5. Text-generation model adapters

- openai_chat.py is the OpenAI text/chat integration.

- gemini_chat.py is the Google Gemini text-generation integration.

- These appear to be thin service boundaries that isolate vendor-specific API calls from the rest of the application.

6. Multimodal / image-related functionality

- openai_vision.py suggests image understanding or vision-based input handling.

- dalle.py and appdalle.py indicate image generation capability, probably for concept art or visual accompaniment.

- This creates a separate model boundary for image generation/analysis alongside text generation.

7. Voice and audio synthesis

- speech.py is the audio/voice layer.

- elevenlabs_client.py, elevenlabs_library.py, and 11labslibrary.py indicate ElevenLabs integration for voice synthesis or voice matching.

- The README explicitly calls out voice matching and dialogue generation, so audio is a distinct subsystem, not just an afterthought.

8. Shared utilities and support code

- utils.py likely contains common helpers used across the app, model wrappers, and generation flows.

- requirements.txt defines the Python dependency boundary for the whole system.

- .env.example shows the app depends on runtime secrets, especially OPENAI_API_KEY and ELEVENLABS_API_KEY.

9. External services and dependencies

- OpenAI is used for chat/text generation and possibly vision/image-related tasks.

- Google Gemini is an alternate or complementary text-generation backend.

- ElevenLabs is the external speech synthesis provider.

- DALL-E is implied for image generation.

- These services are the main external architecture boundaries the app integrates with.

10. Deployment/runtime shape

- Procfile suggests the project is prepared for a process-based deployment platform such as Heroku or similar.

- setup.sh likely automates local or cloud environment setup.

- The repository is therefore a Python application with a Streamlit frontend and vendor API integrations, not a traditional multi-service backend.

11. How data flows through the system

- User input enters via Streamlit.

- The app routes that input through prompt templates and orchestration logic.

- Model adapter modules call external AI APIs.

- Generated text, images, or audio are returned to the UI for display or playback.

- The architecture is centered on “UI → orchestrator → provider adapters → external AI services → UI.”

12. Architecture diagram boundary summary

- Draw one clear application boundary around the Python/Streamlit app.

- Inside it, show separate subsystems for UI, orchestration/prompts, text models, image/vision, and speech.

- Outside it, show OpenAI, Gemini, ElevenLabs, and image-generation services as external providers.