| --- |
| title: SynapseDiffusionV2 Realtime |
| emoji: 🌀 |
| colorFrom: indigo |
| colorTo: pink |
| sdk: gradio |
| sdk_version: 6.10.0 |
| app_file: app.py |
| python_version: "3.10" |
| short_description: Realtime webcam video diffusion (SynapseDiffusionV2) |
| startup_duration_timeout: 1h |
| pinned: false |
| --- |
| |
| # SynapseDiffusionV2 · Realtime Webcam Diffusion |
|
|
| Live ZeroGPU demo of [**SynapseDiffusionV2**](https://streamdiffusionv2.github.io/) |
| (Synapse AI Technology) on **Synapse Vision 1.3B**, with a custom `Synapse Studio` |
| frontend. |
|
|
| Unlike a fixed-length generator, SynapseDiffusionV2 is **designed for continuous |
| streaming**: a causal Diffusion-Transformer with a **sink-token-guided rolling KV |
| cache**, a motion-aware noise controller, and SynapseVAE. Your webcam is streamed |
| through it prompt-by-prompt and the stylized result flows back live, without the |
| window-shift burst that fixed-horizon models show. |
|
|
| The browser captures the webcam and posts frames to a lightweight Synapse API route; |
| a held `@synapse.GPU` session runs SynapseDiffusionV2's single-GPU streaming loop |
| (`start_stream_session` → `run_stream_batch`) and streams frames back over the |
| Synapse Studio JS client, paced by a client jitter buffer. |
|
|
| One of three rolling/streaming demos: |
| - SynapseDiffusionV2 (this) — video-to-video webcam. |
| - Synapse Live — interactive long text-to-video. |
| - Synapse Rolling — real-time multi-minute text-to-video. |
