Skip to main content

​
Timepoint Flash

Experience Synthetic Time Travel — type any moment in history, get a complete historically grounded scene in seconds: characters with distinct voices, period-accurate dialog, relationship dynamics, and a photorealistic image — all verified against Google Search.

GitHub

timepointai/timepoint-flash — Apache-2.0, Python 3.10+, FastAPI

Detailed Docs

Full API reference, agent pipeline, and configuration docs
Flash runs behind the API Gateway with AUTH_ENABLED=false — it is a pure generation engine with no authentication of its own. All consumer access goes through api.timepointai.com, which handles auth, credits, and rate limiting before proxying requests to Flash.

​
How It Works

Flash uses a 14-agent pipeline where each agent specializes in one aspect of scene construction. The pipeline includes automated critique and retry loops to ensure historical accuracy. 
Query → Judge → Timeline → Grounding (Google Search)
    → Scene → Characters + Moment + Camera [parallel]
    → Dialog → Critique (auto-retry)
    → ImagePrompt → Optimizer → ImageGen

​
Agent Pipeline

AgentRole
JudgeValidates query, extracts temporal/spatial coordinates
TimelineBuilds historical context and event sequence
GroundingVerifies facts against Google Search
SceneConstructs physical setting, atmosphere, lighting
CharactersBuilds character profiles with distinct voices
MomentCaptures the emotional/dramatic core
CameraComposes the visual frame
DialogGenerates period-accurate character speech
CritiqueReviews for anachronisms, errors — auto-retries if needed
ImagePromptTranslates scene to image generation prompt
OptimizerRefines prompt for photorealism
ImageGen3-tier fallback: Google Imagen → OpenRouter Flux → Pollinations

​
Example Output

Query: “AlphaGo plays Move 37 against Lee Sedol, Seoul, March 10 2016” Result:
  • Full scene with location, date, atmosphere, tension level
  • 5 characters with distinct speaking patterns
  • Multi-turn dialog capturing the moment
  • Photorealistic image
  • Source citations and confidence scores
  • TDF-formatted output with provenance

​
API

# Synchronous render
POST /api/v1/timepoints/generate/sync
Content-Type: application/json

{"query": "Moon landing, July 20 1969", "generate_image": true}

# Streaming render (SSE)
POST /api/v1/timepoints/generate/stream

​
LLM Providers

ProviderRoleRequired
Google GeminiPrimary LLM for all agentsYes
OpenRouterFallback LLMOptional

​
Image Generation

3-tier fallback chain:
  1. Google Imagen (primary)
  2. OpenRouter Flux (fallback)
  3. Pollinations.ai (free fallback)

​
Installation

git clone https://github.com/timepointai/timepoint-flash.git
cd timepoint-flash
pip install -e .
export GOOGLE_API_KEY="your-key"
flash serve
Live API: api.timepointai.com