VivaChat – FaceTime with AI
TLDR: Video chat with realistic AI persons: https://app.vivalabs.ai/chat Hey AI enthusiasts, After GPT’s Advanced Voice Mode launch, we’re more excited than ever about authentic conversations with AI. We think the next frontier will be incorporating video feeds. On VivaChat, you can video chat with an AI therapist, digital marketing consultant, English language tutor, or a tech recruiter. We’re also beginning to support business use cases around customer support, HR, and education. Building this was a blast! We strung together multiple systems to coordinate an entire…
In plain words
VivaChat is a video chat application that connects users with AI characters including a therapist, marketing consultant, language tutor, and tech recruiter. The platform uses speech recognition, language models, text-to-speech, and video generation to create realistic two-way conversations with visual feed. It targets individuals seeking AI-powered conversations and businesses exploring applications in customer support, HR, and education.
written from the facts on this page · September 2026
From the sources
In the maker’s words, at launch
TLDR: Video chat with realistic AI persons: https://app.vivalabs.ai/chat Hey AI enthusiasts, After GPT’s Advanced Voice Mode launch, we’re more excited than ever about authentic conversations with AI. We think the next frontier will be incorporating video feeds. On VivaChat, you can video chat with an AI therapist, digital marketing consultant, English language tutor, or a tech recruiter. We’re also beginning to support business use cases around customer support, HR, and education. Building this was a blast! We strung together multiple systems to coordinate an entire conversation: ASR + VAD models to process user speech, LLMs to generate text responses, TTS providers to generate speech, and a custom video generation model to generate lipsynced video frames. The key technical challenges were achieving respectable visual clarity (HD quality) and reasonable latency (~1s end to end). We trained custom NERF (neural radiance fields) models to generate video frames consistent with the generated audio in real-time. The vast majority of the latency is in the ASR + LLM + TTS flow, so we’re looking forward to replacing it with end to end speech models! Many lipsync models don’t generate frames in real time (at least 25 fps). Our model generates frames at 30-70 FPS, depending on the GPU. Currently, it requires 330 ms of audio generated ahead to generate a lip synced frame. Conversations usually involve periods of silence and sudden interruptions. We built on top of pipecat to handle this flow. Because a generated video frame is dependent on earlier frames, we have to throttle how much audio is processed by the NERF model. We wanted to make sure that users could talk to avatars ASAP while not breaking the bank. To this end, we use serverless GPU providers like Modal, and we spent time optimizing our cold start setup. The end to end cold start time for a conversation is about 10-20 seconds. It involves starting the containers, loading libraries, and loading model artifacts. Shoutout to Modal’s Memory Snapshot feature! And shoutout to the Daily team for pipecat and video chat SDKs! We’re actively experimenting with approaches to increase visual clarity, emotional expressiveness, rendering speed, and training robustness. While the implicit 3D representation of NERFs is powerful, we’ve found them to be quite sensitive to training data when representing dynamic, time-dependent scenes. We’ll be exploring other 3D model approaches like Gaussian Splatting, which also has faster rendering speeds. We’re considering 2D approaches, such as GANs or latent consistency models, as well. Our model’s handling of emotional expressions is still limited, as we focused primarily on generating accurate lipsync. To this end, we’d like to more explicitly model how facial expressions and head poses vary with audio. -- Try VivaChat for free: https://app.vivalabs.ai/chat Any inquiries (e.g. custom avatars, API access, custom conversation topics, etc.): [email protected]
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 26d ago · cactuscompute.com
Launched alongside, September 2024
the whole month →
BeforeSunset AI 2.0▲1,267Personalized AI daily planning that suits your life
AI · 2024 · beforesunset.ai
