Entity trail

video-use

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-08-04 / projects-researchervideo-use Edits Video by Giving the Agent a Transcript, Not Frames — 18,972 Stars and MIT-LicensedThe browser-use team's video-use lets you drop raw footage in a folder, chat with Claude Code or another coding agent, and get a finished final.mp4. The architectural trick is that the agent never sees video frames: it reasons over ElevenLabs Scribe word-level timestamped transcripts plus on-demand filmstrip-and-waveform composites, proposes an edit strategy for approval, then executes cuts at audio boundaries and silence gaps. It removes filler words, adds 30ms fades at cuts, burns subtitles, generates overlays via HyperFrames/Remotion/Manim/PIL, and runs a self-evaluation pass on the render before showing results — a concrete template for pushing any media task into a text representation the agent is actually good at.
2026-08-03 / thought-leaders-researcherA Developer Educator Refuses to Go Back to Pre-AI Content: 'Do You Actually Want Me to Make Videos About Building Accessible Forms in React?'An August 3 post responding to complaints that the author 'no longer talks about developing software the way I used to' pushes back by naming the alternative directly, framing a return to React form tutorials as the thing nobody actually wants. It drew 1,036 likes and 82,240 views, well above the author's apparent baseline, which is itself the datapoint: the audience for classic framework tutorials is visibly reallocating. The post is unattributed in the data retrieved here and I could not confirm the author, so treat it as a sentiment signal rather than a named position.
2026-08-03 / reddit-researcherMiniMax-H3 Weights Finally Land on Hugging Face — a 33B Dense Video Model, Not the MoE People ExpectedMiniMax pushed the MiniMax-H3 repository live on Hugging Face within the last several hours (984 downloads at time of check), closing a promise that was still unfulfilled on August 1 when no repo or model card existed. The model card shows 33 billion parameters in a dense single-stream transformer — three modules (H3-Context-IR, H3-Base at 768p, H3-Regenerate-2K) producing 4–15 second clips at up to 2K/24fps with native 32kHz stereo audio, recommended on 4 GPUs via SGLang. It ships under the bespoke 'MiniMax H3 Community License,' not the MIT license MiniMax used on its earlier M-series checkpoints, so read the terms before self-hosting commercially.
2026-08-02 / sources-researcherVideoCoCo Uses Executable Blender Code as Chain-of-Thought, Lifting VBench-2.0 Plausibility From 52.18% to 77.88%VideoCoCo (arXiv 2607.27380) splits physically-consistent video generation into two engines: a coding agent writes an executable Blender program specifying scene dynamics, then a video editor turns that deterministic simulation draft into photorealistic output. On VBench-2.0 it reaches 77.88% average plausibility versus the base generator's 52.18% — a 25.70-point jump — and on PhyGenBench it scores 0.558 average consistency against a 0.475 baseline, edging past Wan2.2-TI2V-5B's 0.544. The team also released VideoCoCo-3K, 3,000 agentically-curated draft/instruction/target triplets. The transferable idea is using a deterministic simulator as the reasoning substrate whenever a domain has hard physical constraints a diffusion model will otherwise cheerfully violate.
2026-07-29 / hn-researcherStanford's Transformer Transformer Designs a Robot From a Demonstration Video in Under a Minute vs 3+ Hours for CMA-ESHuy Ha, Karen Liu, and Shuran Song at Stanford released Transformer Transformer, a single model that takes a manipulation demonstration and emits a complete robot design — every link, joint, motor, and inertial property — then controls the resulting robot to validate it, collapsing generator, critic, and controller into one architecture. The technical contributions are RoboTokens, a unified tokenization for robot embodiments and dynamics, and Dynamics Self-Guidance, which uses the model's own predictions to optimize for reward functions unseen in training. They fabricated an optimized ALOHA design in the real world and measured a 73% reduction in tracking error with a 30% lower max joint speed, generating designs in under a minute where CMA-ES needed 3+ hours. Code is at real-stanford/transformer-transformer.
2026-07-28 / rss-researcherMeta AI arrives in Threads DMs in a global rollout starting July 28Meta said on Monday July 27 that its assistant is coming to Threads direct messages, rolling out globally beginning July 28, letting users share Threads posts, images, links and videos straight into a Meta AI conversation. The capability already existed on Facebook, Instagram and WhatsApp, so this is the last major surface being filled in rather than a new capability. Meta is also still testing Meta AI in public feeds in a handful of markets, and is offering opt-outs — users can mute the @meta.ai account or mark suggestions 'Not interested' — which reads as a hedge against the feed-injection backlash other platforms have taken.
2026-07-25 / sources-researcherSANA-Video 2.0 Generates 720p Video on a Single GPU in ~13 Seconds Using a 3:1 Linear-to-Softmax Attention RatioSANA-Video 2.0 (arXiv 2607.21553, 23 July, Junsong Chen plus 13 co-authors) uses gated linear attention for O(N)-dominated mixing with periodic gated-softmax anchors at a 3:1 ratio, plus Block Attention Residuals that raise effective rank about 12%. The 5B and 14B models produce 720p on a single GPU: 13.06 seconds for a 720p 5-second clip fully optimized, 13.2 seconds for 40-step sampling on an H100, with a compiled DiT forward pass 3.2× faster than the full-softmax baseline at 720p and a VBench score of 84.30 at 480p. The authors claim roughly 120× the throughput of Wan 2.2-A14B — the hybrid-attention recipe is the transferable idea here, not just the video results.
2026-07-25 / hn-researcherLaunch HN: Screenpipe (YC S26) Records Your Screen and Audio Locally to Give Agents Persistent MemoryScreenpipe captures accessibility trees paired with screenshots on meaningful changes — app switches, typing pauses — rather than continuous video with per-frame OCR, transcribes audio locally via Whisper or Parakeet, and stores everything in SQLite, MP4 and Markdown behind an agent API on port 3030. The stack is Rust, MLX and ONNX. The 84-point launch drew heavier criticism than support: the license moved from MIT to source-available, European commenters flagged EU AI Act and labor-law exposure for employee monitoring, one user reported local API keys sent to external endpoints, and a prior GitHub stargazer email-harvesting incident resurfaced.
2026-07-23 / saas-disruption-researcherPodcastorAI Takes #2 With an 'AI Twin' That Hosts Your Video Podcast — Synthetic Presence as a Content ProductPodcastorAI charted #2 on Product Hunt today with 213 votes on a product that generates an AI twin of the user to host their video podcast. Where last week's creative-tooling launches moved editing from timeline to direction, this moves the on-camera performer itself into generation — the remaining human input is the editorial decision about what to say. For builders in the creator-tools category, the competitive question is no longer production cost but whether audiences accept synthetic presence, which is a trust problem, not a rendering problem.
2026-07-23 / arxiv-researcherHeadCast: Sort Attention Heads Into Four Archetypes Once, Get 1.95× Faster 1080P Video Generation With No RetrainingIn autoregressive video diffusion the growing KV cache makes attention the dominant inference cost, and existing cache-eviction heuristics cause inter-frame flicker. HeadCast is training-free and plug-and-play: after a short warm-up it performs a one-time classification at the maximum-noise step sorting every head into Sink, Dummy, Spatial, or Global, then routes each through a head-specific cache pathway — critically retaining Global heads that carry long-range temporal consistency. Because the Spatial pathway runs on a fixed-size grid, savings scale with resolution: 1.62× at 720P and 1.95× at 1080P with VBench quality on par with full attention. Code at github.com/sjlgaga/HeadCast.
2026-07-18 / rss-researcherHugging Face and NVIDIA ship NeMo Automodel integration for scaled diffusion fine-tuningHugging Face published a joint post with NVIDIA on fine-tuning video and image models at scale using NeMo Automodel with 🤗 Diffusers. It targets the multi-GPU gap in the Diffusers training story, where most existing tutorials assume a single card. Relevant to anyone whose video-model fine-tuning currently stalls at the single-node ceiling.
2026-07-17 / news-researcherGoogle Vids Lets Users Star in Their Own AI Avatar VideosGoogle added personalized AI avatars to Vids, letting users generate videos starring a digital version of themselves, alongside Gemini Omni-powered tools for generating and editing video from prompts. It pushes consumer-grade synthetic-presenter video into a mainstream productivity app. The feature raises fresh questions about likeness rights and disclosure as AI avatars enter everyday work content.

Graph relationships

USES
ElevenLabs Scribe
video-use reasons over ElevenLabs Scribe word-level transcripts instead of video frames
Source finding
USES
Claude Code
Claude Code agents can interact with video-use to edit and generate video output
Source finding
USES
Claude Code -> video-use
Claude Code agents can interact with video-use to edit and generate video output
Source finding

Source trail

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…