ChetiArt's CrossView-Warp Lets MiniMax H3 Reshoot Video From Any Angle
A community LoRA for MiniMax H3 rebuilds any input clip from a new camera angle using depth warping and a Blender-trained novel view synthesis pipeline.
20 stories
A community LoRA for MiniMax H3 rebuilds any input clip from a new camera angle using depth warping and a Blender-trained novel view synthesis pipeline.
A new open-source 4B and 1.7B model family claims leading agentic and coding scores on-device, with native 1M-token context and Apache 2.0 licensing.
A macOS agent skips screenshots to frontier models, uses OCR plus a tiny classifier called TypeSafe, and runs steps for a fiftieth of a cent.
Ornith pairs its 35B mixture-of-experts reasoning model with a block-diffusion draft that pushes decoding past 180 tokens per second on a single A6000.
A training-free trick keeps only the top 1% of tokens per frame, letting VGGT-style models reconstruct thousands of video frames on a single GPU.
HyperQwen crams a 27B Qwen model onto one 24GB RTX 3090 with vLLM patches, hitting 127 tok/s solo and ~1,035 tok/s at 64 concurrent.
Tencent Hunyuan open-sources a 1.5B speech foundation model that handles TTS, editing, enhancement, and separation through plain-language instructions.
A cross-linguistic study runs nine frontier LLMs through seven personality tests and finds each model has a stable, identifiable behavioural fingerprint.
Alibaba's next-gen simultaneous interpretation model cuts average lag to 2.3 seconds across 60 languages, adds speaker diarization, and clones each voice separately.
A 4B open-weight decision model that returns calibrated probabilities to typed questions in one forward pass, priced at $0.042 per million input tokens.
Tencent Cloud open-sourced Octop, a single-process, self-hosted AI assistant that gives every user their own team of MBTI-flavored agents across web, CLI, and chat apps.
Researchers isolated a linear direction in 25 open-weight LLMs that behaves like pain, distinct from fear or sadness, and drives self-relief behavior.
Stanford and Together AI show fixed teams of LLMs can learn reusable collaboration protocols that beat any single member and even a perfect answer router.
xAI's new image model jumps 14 spots to fourth place on the Artificial Analysis leaderboard, undercutting OpenAI's flagship on price by more than 3x.
Anthropic taps Accenture's Faculty unit to embed evaluators inside its labs, with each side pledging $1B over five years to build safety oversight capacity.
Vals AI's new CUA-Bench pits frontier models against six commercial video games with only pixels in and keystrokes out, and every model scores below 20%.
A new research pipeline called CodeMidas builds reinforcement learning environments straight from source code, boosting coding agent scores on five benchmarks.
SpaceXAI's new speech-to-text model doubles accuracy over v1.0, tops the streaming leaderboard, and keeps pricing at ten cents per hour.
Factory rolls out on-premises and air-gapped deployment for its Droid coding agents, targeting enterprises that cannot ship source code to a vendor cloud.
Bolt adds DeepSeek V4.1 Flash to its Forge open-model lineup, promising 10x the usage of V4 Pro and stronger visual design output.