Hans @MGHansLin

When several people talk at once, a transcript can get messy fast. Our new Nemotron 3 Diarization model tracks who spoke when, even when voices overlap. It handles up to eight speakers, has 100M parameters, and is now available on @huggingface 🤗

Introducing Bot Mode for Hermes Desktop. Your agent profiles become a series of named Bots. Each Bot has its own role, model, memory, skills and profile picture; Bots can use any model and even communicate with each other. Build a specialist Bot once to use it forever.

I just reserved my Cloudflare Wallet tag: ctbc.cloudflare.pay. Reserve yours now at cloudflare.pay

Claude终于补上了自己最大的短板，现在能真正看懂视频了。 不用等Anthropic官方慢吞吞的更新，现在通过Algrow的MCP平台，你只要扔一个YouTube、TikTok或者Instagram的链接，Claude就能完整看完整个视频，分析画面、音频、节奏、文字叠加，甚至给你出一份可落地的深度分析报告。 演示效果直接颠覆认知， 输入一个YouTube Shorts频道的URL，Claude会自动调用视频分析工具，批量拉取该频道所有短视频，生成包含标题、播放量、点赞率的完整数据表格。 接着它会做病毒化深度拆解，告诉你为什么有的视频爆了，有的却石沉大海。 它能精准识别出所有视频共用的模板、配音和动画风格，然后一针见血地指出：决定性差异从来不是后期特效，而是前2秒的hook概念冲击力。 最狠的是帧级留存分析，它会画出每一秒的观众流失曲线，标出哪些时间段是“死亡区”，观众会在这一秒毫不犹豫划走。 以前你要花三天时间，手动刷几十条视频、记满一本笔记、做无数个表格才能得出的结论。 现在一句话，30秒搞定。 Algrow的后端会自动完成视频下载、关键帧提取、音频转写，再把多模态数据完整喂给Claude。 通过MCP工具链，它能一口气跑完“看视频→抓数据→交叉对比→输出报告”的全流程，中间不需要你任何干预。 目前这个功能正处于免费无限试用的推广期。 正式上线后定价也非常亲民，1个积分可分析约8分钟视频，最低25美元/月的订阅就包含12小时以上的分析额度。 为什么这件事意义重大？ 1. 大模型的战争已经彻底转向生态 Claude的文本推理能力一直是行业顶尖，但原生视频能力长期落后于Gemini。这次第三方开发者用MCP的方式，一夜之间就把Claude拉到了同一水平线，甚至因为工具链式调用能力变得更强。这证明未来的胜负手从来不是参数，而是谁的生态能长出更多杀手级应用。 2. 内容创作者迎来核武器级生产力工具 以后再也不用熬夜刷竞品了。让Claude帮你把同赛道所有爆款和扑街视频全看一遍，直接提炼出可复制的hook公式、转场节奏和内容结构。谁先用上这套工具，谁就能在流量战里领先别人一个身位。 3. 视频研究Agent的时代正式开启 视频是互联网上体量最大、最难处理的数据类型。一旦Claude能看懂视频，再结合MCP的工具链能力，很快就能诞生能自动监控全网流量趋势、生成完整内容策略、甚至迭代创意的AI Agent。内容分析这个曾经的人力密集型工作，正在被彻底重构。 当然它也有局限，对小众文化梗和细腻情绪的理解仍有误差，敏感内容不要随意投喂，免费期结束后算力成本可能会传导到价格上。 但这些都不重要，潘多拉的盒子已经打开了。

Anthropic 設計師 Ryan Mather 的 7 個實戰 tips

Ryan Mather @Flomerboy

6 months ago

🧵 My tips for getting the best results out of Claude Design! I’m on the verticals team at Anthropic which means I serve 7 different products. Claude Design makes it possible! 1. Set up your design system and your core screens. An hour of setup and refinement here is worth it

177 672 9K 1.5M 18K

这个厉害了 Midjourney 发布名为 Omni-Reference（全向参考） 功能 允许用户在生成图像时明确指定：“把这个东西放进图像里”。 也就是你可以给一个参考图像，指定将其放入生成的图像里 支持添加各种类型： -人物、动物或非人类生物 -武器、道具、车辆、物体等 -甚至是一整套角色造型或风格元素 这是 V6 版本中角色参考（character reference）功能的进化版，功能更强。

introducing Video Training, a new level of control for AI video generation. now you can train Wan 2.1 with your own videos and have it learn custom styles, motions, or objects. here's how it works 👇

Wan 团队的 All-in-One 视频编辑框架 VACE 太强了 可以混合任何内容到视频中： Move-Anything: 在图像中移动选定对象到新位置 Swap-Anything: 替换图像中的一个对象为另一个对象 Reference-Anything: 参考图像的风格应用到目标图像 Expand-Anything: 扩展图像边界，智能填充新区域 Animate-Anything: 为静态图像添加动画效果 支持视频重新渲染包括：内容保留、结构保留、主体保留、姿势保留和运动保留等

架空のCMをAIで90年代をイメージして作ってみました！ I used AI to create a fictional commercial inspired by the 90s! ▼使用した動画生成AI ・神威/KAMUI（@kamui_qai）Google Veo2 ,Wan2.1 ・Luma AI（@LumaLabsAI）Ray2 ・その他SE、音楽、音声もすべてAIです！ #kamuiart #Veo2 #Wan

Hello @CHANEL , Here is your Chanel 2025 “Pre-Pre-Pre-Pre-Valentine’s” Gen AI Spec Commercial. I’m experimenting with how far we can go to create a luxury, traditional commercial entirely with AI, inspired by a poem from Virginia Satir. All clips were generated using OpenAI’s Sora Text-to-Video. @OpenAI #SoraShowcase

what? Krea AI now lets you turn images into 3D objects and blend seamlessly in any background in real time .. This is one of the most innovative AI tools so far and it’s free..🤯

this is wild.. this new AI relighting tool can detect the light source in the 3D environment of your image and relight your character, the shadows look so realistic.. it's especially helpful for AI images 10 examples:

Here's my first series of Anime created by Generative AI. Vidu 2.0 is a groundbreaking advancement for storytelling in the Anime style. #vidu @Viduforhuman Hi everyone 😊 😉 Here's my first AI-created Anime series. The name of my anime is Kuro & Yuki. I won't spoil the story for you, but it begins with these two boys locked up in what appears to be a highly secure institute/prison. Yuki has reached the age where he has awakened his power. While wandering the corridors of the institute, he meets Kuro, a boy with autism. It's a fleeting encounter as they end up being separated. As the story unfolds, you'll understand who they really are and why they are locked up. I don't plan on making episode 2 yet, even though I've grown attached to my characters' story. I'll continue the story only if a lot of people are interested and want to know what happens next. Otherwise, I'll explore other universes and styles of Anime 😉 Tools: - Epidemic Sound (For sound effects) bit.ly/4j02Ode - Vidu AI (to transform frames into animation) vidu.com - ElevenLabs (for certain expressions) bit.ly/42iONl9 When I say Vidu 2.0 is a game-changer, I mean it. I had already made my animation with the previous version of Vidu, but when they gave me access to the beta, I urgently changed my plans and recreated all the animations. Let me tell you, it's a whole new level. My first animation was ultra frustrating to bring my ideas to life! Really! And since Vidu couldn't handle many image styles, I clearly couldn't bring my ideas to life. But with Vidu 2.0, I enjoy it much more! For the first time, I really get to bring my ideas to life! Until now, I always had to make compromises. Of course, Vidu is still not perfect, and there are still many obstacles, but it is a truly magnificent advancement! (and it surpasses all the Anime-style AIs I've tested recently) Moreover, Vidu 2.0 is fantastic for special effects, but also for embedding elements into the video (like a hand that appears and interacts with the character, or another character; and the best part is that they perfectly match the style of the original images!) * The animations/images and voices are AI-generated * The SoundDesign is traditional and done by me * The story was entirely written by me (I don't use AI to create my stories, it's important to me to write them myself) The voices were generated thanks to Nijivoice (for secondary characters) and Hailuo Audio (for main characters). I thank Yachimat (@yachimat_manga) for introducing me to Nijivoice, it's very generous of him! During the beta, we didn't consume credits, otherwise, this animation would have cost me around 20,000 credits, ha ha. Vidu 2.0, with its superb stability and fidelity to the style of images, offers brand new horizons for storytelling! One problem with Vidu 2.0 is that for now, you have to manually extend videos (by exporting the last frame). As you can see, this allowed me to create long scenes with different actions, and everything fits together perfectly! There are still many obstacles to storytelling, such as character consistency (I use the image to video function; I produce my images with Nijijourney) and it's always laborious to have a consistent character! The same goes for backgrounds. Here are the points I've identified that would facilitate storytelling: For Vidu: - Function to extend videos (you can already do this manually by exporting the last frame) - Function to invent the beginning or end of a video using a single frame (you can already do this manually by putting a completely white or completely black image as the first or last frame; this technique also allows you to have very dynamic results with Vidu 2.0! ) - It's difficult to obtain facial expressions for certain image styles (the same image style that the previous version couldn't handle at all). - More dynamism There's always room for improvement, but Vidu 2.0 has truly opened up exciting new possibilities for creative storytelling. ai aiart aianimation aianime ainews anime animenews aitools aitool

给可灵 1.6 做了一个详细的测试，它就是现在最强的图生视频模型！ 写实图片生成视频居然可以完成 Veo 2 的牛排测试 风格化更是断档领先，我都怀疑 Veo2 有没有这么强 无论是 3D、动漫还是艺术画都不在话下 试完之后又想自己搞片子了。

用 Gemini 2 分析复刻了一下之前很喜欢的一个 AI 视频 结果非常好，画面内容和节奏都很相似 Gemini 2 学习和复刻内容成本真的低好多 主要是被海辛下面这个用法启发的，制作过程在下面：

Ring Hyacinth @ring_hyacinth

2 years ago

给大家分享一个小发现： 可以靠在 prompt 上写时间戳来控制镜头运动 然后时辰随手丢个给了我一篇论文，说明理论上确实可行的：arxiv.org/html/2412.0216…

7 39 292 81K 260

【メッシュが驚くほどキレイ】 NVIDIAが発表した3D生成AI text/image-to-3DモデルのEdify。 表面の色などだけでは無く、その他モデルと比べると メッシュが素晴らしくきれいなどが話題に。素晴らしいクオリティです。 #生成AI #3D

This is incredible! DaVinci Resolve, the Hollywood color grading software, just released a brand new AI motion tracking tool. With just one click, you can effortlessly track even the fastest moving objects. It’s free to use, link in comments! Here’s how to make it work:

Krea AI 上线了 Lora 训练功能。 上传最少三张图片就可以训练，界面和交互非常精致和直观。 FLUX Lora训练真的很简单，但是一堆产品没有一个再体验和交互上下功夫的。 Krea 真的活该成功。

Krea @krea_ai

2 years ago

finally, AI trainings arrived to Krea. train an AI model and teach it about your own characters, styles, products, and more. sharing 100 early invites below 👇

80 126 936 175K 501

finally, AI trainings arrived to Krea. train an AI model and teach it about your own characters, styles, products, and more. sharing 100 early invites below 👇

v4 is coming soon 🎧