π§ Inspiration
People experiencing dementia or memory loss often feel disconnected from their past and surroundings. Caregivers rely on reminiscence therapyβusing photos, stories, and familiar soundsβto spark recognition.
We imagined what this could look like with todayβs AI: a system that listens, understands, and recreates meaningful memories through sound and video.
That vision became Memento AI β a conversational companion that turns voice input into cinematic memories.
π¬ What it does
Memento AI listens to the user describe a memory β for example:
βMy family trip to the lake when I was a kid.β
Using Gemini 2.5 Flash, Google Veo 3.1, and ElevenLabs, it generates a short personalized video that visually and emotionally recreates that moment.
It speaks in a gentle, adaptive voice, helping users recall, relive, and reconnect with who they are.
βοΈ How we built it
- Gemini 2.5 Flash β extracts emotional and contextual details from user prompts.
- Google Veo 3.1 β generates cinematic video scenes based on those details.
- ElevenLabs Voice AI β narrates the story with realistic emotional tone.
- LangChain Memory Graph β stores life events for contextual recall.
- Web playback layer β synchronizes video + audio for immersive replay.
π§ Challenges we ran into
- Synchronizing Veo render time with ElevenLabs voice output.
- Avoiding uncanny or distressing visuals for sensitive users.
- Keeping latency and costs low with multimodal pipelines.
- Prompt consistency β faces, lighting, emotional continuity.
π Accomplishments that we're proud of
- Built a fully multimodal memory generator from scratch.
- Designed a tone-adaptive narration pipeline using ElevenLabs.
- Showed how generative media can comfort and connect, not just entertain.
π What we learned
- How to coordinate video + voice + reasoning models cohesively.
- Emotional pacing matters more than raw accuracy for trust-building.
- Ethical prompt design is essential for therapeutic AI.
π What's next for Memento AI
- Integrate real family photos and voices (with consent).
- Expand to a mobile reminiscence app with on-device generation.
- Explore personalized cognitive-care companions that help people hold on to their stories.
Built With
- 2.5
- cloud
- elevenlabs
- ffmpeg
- gemini
- javascript
- langchain
- memory
- python
- veo
- voice



Log in or sign up for Devpost to join the conversation.