Vidu S2 rewrites the video while it is still playing

Share
Vidu S2 rewrites the video while it is still playing

The two stories nobody outside China is covering this morning are both about the same thing: what happens to AI video and AI serving once the model has to answer in real time.

Shengshu Tech made the live video stream editable while it plays. Vidu S2, released September 16, is two models — Avatar for live interactive digital humans and Editing for changing a running video — and the avatar's live output moved from 540p to 720p while the frame rate holds, so a viewer's request to change someone's outfit, swap the person, replace the background or restyle the whole scene lands inside the stream instead of in a re-render queue. Reference images can be handed over during a conversation and take effect from a chosen timestamp, which is what makes the thing feel directed rather than prompted. The engineering, published alongside the release, is the interesting part: a Backbone-Refiner split that generates structure at low resolution and repairs detail in an asynchronous second pass, a Self-Replay Forcing training step that teaches the model to absorb its own drift instead of compounding it, and a vision-language agent that watches what has already been generated before deciding how to execute the next instruction. Shengshu also demonstrated real-time spatial video for headset viewing — the same territory ByteDance is reportedly building toward — and the version gap is 69 days from S1, which tells you what the video-model race looks like from inside.


Intel is making the argument that the KV cache belongs on the CPU, and its numbers are hard to dismiss. A KV cache is the note-taking layer of transformer inference: the Key and Value tensors for tokens already processed, kept so the model doesn't recompute its own context. It is also pure HBM appetite: for one mid-size open model's layer configuration, caching a hypothetical one-million-token context works out to roughly 147 GB of Key-Value data before the model weights take their share. Intel's answer is tiered — hot cache stays in GPU memory, cooling cache drops to CPU-side DDR, cold cache sinks to SSD, with the processor acting as traffic cop. Its KV Shrink layer pairs that scheduling with QuickAssist compression, which Intel reports lifted lossless space savings from about 10 percent to the 20–30 percent range by reordering the data before compressing. In Intel's own test — dual Xeon Gold 6554S, two Nvidia L20 GPUs, Qwen3-32B, an 80 percent cache hit rate — time to first token came in up to about five times faster than a stock vLLM baseline without offloading, with hardware compression running roughly twice as fast as software compression and adding under 10 percent of extra first-token latency. Treat those as vendor numbers on a curated setup, not a benchmark result: the honest read is that once an agent is waiting on a tool call, its session history is squatting on expensive accelerator memory, and that is a bill someone will keep trying to move.


ByteDance refreshed Doubao-Seed-2.1 Pro to a v0915 snapshot, and the changelog is all agent reliability. The update, live on the Volcano Engine Ark platform, targets long-horizon task delivery and multimodal coding rather than headline benchmark scores. It lands on top of the launch version introduced at the June Force conference, where Volcano Engine president Tan Dai framed the model as crossing a "production-grade" threshold on code delivery, long-range agent tasks, multimodal understanding and enterprise stability. A snapshot refresh about twelve weeks after a flagship's debut, published without a launch event, is how frontier labs operate; Chinese model vendors are now matching it in public.

Real-time means the model can no longer take an hour to think — does that push video generation ahead of text in the race for inference investment? Write to us via our contact page.

Source: arXiv — Vidu S2: Real-Time Interactive, Editable, and Spatial Video Generation

Get SiliconNoon in your inbox

One email a day with the AI stories that matter — what happened, the numbers, and why it matters. No spam, unsubscribe in one click.

Read more

Doubao 2.1 Pro clears 83% of a 387,000-line repo's open issues

Doubao 2.1 Pro clears 83% of a 387,000-line repo's open issues

ByteDance spent the last 24 hours upgrading three layers of the same stack at once: the model, the workplace suite it lives in, and the agent that ties them together. Volcano Engine pushed Doubao 2.1 Pro to a 0915 snapshot, and the number it is selling is a repair run: multiple sub-agents worked the open-source game Luanti — roughly 387,000 lines of code and 1,000 real historical issues — in parallel for 36 hours and brought 83% of those issues to merge-ready standard.

StepFun ships five voice models and takes two global No. 1 spots

StepFun ships five voice models and takes two global No. 1 spots

A Chinese lab just shipped the whole voice stack in one afternoon — and the pacing fight in Washington got a very public answer from Nvidia's CEO. StepFun released five StepAudio 3 models on the same day, and two of them now sit at the top of Artificial Analysis's speech leaderboards. StepAudio 3 Realtime, ASR, TTS, Gen and Music went live on StepFun's open platform Tuesday, covering everything from full-duplex conversation to music generation.

Five robot brains in eight days — and the money follows

Five robot brains in eight days — and the money follows

Two things happened to Chinese robotics in the same two weeks: the body makers shipped model after model, and the funding data caught up with why. Three Chinese robot makers released five embodied-AI foundation models in eight days this month — Unitree, AgiBot and Noetix — a pace that says the hardware race is over and the brain is the product. The sequence, per Red Star Capital Bureau: Noetix launched its Scalabot embodied-AI brand and the first HERON world model on September 8, then…

HiDream-O1-Video debuts fourth on Artificial Analysis

HiDream-O1-Video debuts fourth on Artificial Analysis

A Chinese startup put a video model on the global boards at the top on release day. Separately, the US chip buildout is running straight into the thing money cannot buy fast enough: technicians. HiDream.ai released HiDream-O1-Video-1.0 (HD-V1) on September 15, and it landed fourth worldwide on Artificial Analysis's Image-to-Video leaderboard for models with audio — an Elo of 1176, behind MiniMax H3 Max (1195), MiniMax H3 (1181) and Google's Gemini Omni Flash (1177), and ahead of ByteDance's…

Qwen's agent built a research-grade telescope simulator in 3 days

Qwen's agent built a research-grade telescope simulator in 3 days

Two stories about the layer underneath the models: the agent that now plans the observation, and the memory the whole stack is short of. Alibaba says its Qwen Office agent helped a research team at the National Astronomical Observatory of China build a large-aperture, research-grade telescope simulation system in three days, for under 1,000 yuan — work that previously went to an outside software vendor, took about three months and cost tens of thousands of yuan.

Zhipu's GLM-5.3-FlashX hits 200 tokens/s — at 2.5x the price

Zhipu's GLM-5.3-FlashX hits 200 tokens/s — at 2.5x the price

Zhipu shipped the fast tier of its open model today and, unusually for a Chinese lab in 2026, raised the price to do it. GLM-5.3-FlashX is live on the API at up to 200 tokens per second — roughly five times the throughput of GLM-5.3-Flash — and priced at about 2.5x what the base model costs. Zhipu says its 100,000-chip domestic inference cluster was saturated at launch under the weight of demand that never let up, and that the new tier follows fresh infrastructure investment and inference-side…

Qwen's interpreter model cuts live translation lag to 2.3 seconds

Qwen's interpreter model cuts live translation lag to 2.3 seconds

Alibaba's Qwen team shipped a new simultaneous-interpretation model today, and the interesting claim isn't translation quality — it's who is talking. On the same news cycle, Sam Altman's UN appearance got a date. Qwen released Qwen3.8-LiveTranslate, a real-time simultaneous interpretation model that rebuilds the pipeline as one interleaved stream of audio and text — and drops average per-character lag (LAAL) from 2.8 seconds to 2.3 seconds.

Shanghai AI Lab chief: science is the next programming

Shanghai AI Lab chief: science is the next programming

One of China's biggest AI labs says the frontier after code is the wet lab — and it shipped a robot-running world model to make the case. Wenbow Zhou, director and chief scientist of the Shanghai AI Laboratory, used the main forum of the 2026 Pujiang Innovation Forum on September 12 to argue that "scientific research is the next programming." His reasoning: programming is AI's biggest real-world productivity win so far, and research has the same shape — problem-oriented, cognition-driven …