跳到正文
今天11月12日周四9 条
11月11日周三
  1. Rohan Paul63

    IFM 推出开放基础世界模型 PAN 2,接受描述的动作并生成最多 20 秒展示物理后果的视频,定位为让 Agent 把试错成本移入模拟。架构上由语言模型先规划情节、再由 29B 视频模型渲染。官方基准显示 PAN 2 完成单步 71.5%、完整计划 34.0%,在开放模型中完整计划完成率第一、Physics-IQ Verified 排名第三。

    引用Institute of Foundation Models@IFM_AI

    Introducing PAN 2: an open foundation world model for physical intelligence. A language model predicts the next word. PAN 2 predicts the next world: what happens after each action, grounded in time, physics and space. Towards AI that simulates “what if?” before it acts. More details: ifm.ai/pan/

11月10日周二
  1. 公众号:昆仑万维53

    昆仑万维方汉谈AI短剧出海:工具退场,系统登场,进入全链路体系之争

    昆仑万维董事长兼CEO方汉在杭州AI文娱(短剧)产业对洽会上分享对AI短剧的判断:产能问题已解决,竞争正从单点工具转向覆盖生产到发行的系统能力。他援引数据称今年前9个月上线短剧约49万部、海外市场规模超60亿美元同比增长67%,爆款率仅约千分之五;昆仑万维2026年9月AI短剧平台单月流水近1亿美元,月度上新剧集中95%以上由AI生成。

  2. Rohan Paul54

    Seele AI 推出面向游戏制作的 Seele03 模型,配合内置的网格生成、绑定、音效等资产生成器,可将成品输出为可下载编辑的 Unity、Godot 或网页项目。

    引用Grummz@Grummz

    Seele is a company that creates AI tools for game making. Recently they invited me to participate as they took on a 5-day challenge to push their new model, Seele03, to its limits and move closer and closer to AAA with a Wuxia-style action game. The product consists of their AI model, asset generators, and target engines such as web (three.js), Unity, or even Unreal. What's nice is that their site has everything in one place, from mesh generators and rigging to sound fx and sprite work. They also provide common templates for most types of games. Having it all work together seamlessly saves a lot of time. You can get a game running very quickly, in 20 minutes or so. (20 minutes! I remember when this took months and months.) So what can they do in 5 days? Their progress video, below, shows them iterating through creating a Phantom Blade Zero-style game. This is not reverse engineering, but made from scratch. Over the course of development, they sent me video updates on their progress. The 03 model is designed for iterative development, and I gave them some advice as they took it from basically nothing to the final version of the video today. The progress the model made day to day was rapid, and the traces actually factored in something that I don't see that often: game design. Most models focus on the graphics when making a game, but Seele03 is thinking about combos, hit feedback, everything that really matters for how a game feels and plays. One day, I think that's going to be the core of what makes AI game making viable. After you are done prompting and refining your game, it's not frozen on the web. You can download and edit the project files themselves. Since you can download the result and open it in Godot or Unity (Unreal is embedded web atm), Seele03 can give you a quality core and a way for you to take things even further. So how did it turn out? I've tried a lot of game generators, and Seele03's all-in-one tool approach and engine targeting stand out for me. Most others only generate three.js for web, or are more world models based on generative rendering. This is the rare one I've seen that is making games directly in game engines. They've asked me to share the full process video with you so you can judge for yourself. As I watched the team work, I sat back and wondered at how it's all changing so fast: you simply... chat... interacting with a UI that is purely text, typing in wishes and directions, and everything else unfolds around you. I think they got pretty far in their goal gameplay-wise, which is where they really focused. The key thing is that after you have the final result, you also have the files to download, and you can replace anything you want to take it even further: animations, level graphics, basically anything. If you want to make games for real engines on PC and console, then give Seele a shot at seeles.ai. It's faster than vibing a game in Claude because of their game-design-driven model, their built-in asset generators, and the fact that you are not burning tokens re-creating an entire engine before you even get to your game. And every day they get closer to AAA... it's just a matter of time. @SEELE_AI1125

  3. Kling AI40

    完整视频在这里 🎬 x.com/Kling_ai/status/2117587…

    引用Kling AI@Kling_ai

    How much life can you find in black? @EmilHeliot , the Copenhagen-born global fashion brand, just released their new Core Uniform collection campaign film, directed by UNVEIL, creative direction by Julius Juul, made with Kling 4.0. It's a study in black, the same color reinterpreted again and again, but none of it feels flat. Shadows with depth, fabric that catches the light, faces that say a lot without speaking. “Kling 4.0 brings together two things we value: serendipity and control.” Build the world first, then let the film happen inside it. - Tom-Jacques Perret, co-founder of UNVEIL. #Kling4 #KlingAI

  4. 公众号:数字生命卡兹克81

    实测 Claude Fable 5.2:一次直出视频与 3D 网页,审美与细节表现突出

    Anthropic 凌晨正式发布 Claude Fable 5.2,作者第一时间实测视频动效、Three.js 3D 场景和苹果静安店 3D 网页等 10 个任务,全部一次直出,未发现生成视频中的事实错误。

    推荐理由:作者用多个一次直出的实测案例展示 Fable 5.2 在视频动效和 3D 前端上的表现,并给出额度消耗和日常模型搭配建议。

  5. fal61

    fal 宣布所有 FLUX 3 Max 视频生成 5 折优惠,持续至 11 月 22 日。FLUX 3 Max 是 fal Research 与 bfl.ai 合作的后训练视频模型,在第三方图像转视频质量评测中并列第一,生成 10 秒 720/768p 视频约需 12 秒,试用入口为 fal.ai/flux-3-max。

    引用fal@fal

    隆重推出 FLUX 3 Max,这是 fal Research 与 @bfl_ai 合作推出的最新后训练视频模型。 在独立第三方基准测试中,FLUX 3 Max 在图生视频质量上并列第一,可在约 12 秒内生成一段 10 秒、720/768p 的视频。 在接下来的三周内,FLUX 3 Max 享受 75% 折扣,使其成为同等质量水平下最实惠的前沿视频模型。

11月9日周一
11月8日周日
11月7日周六
  1. Kling AI60

    可灵(Kling AI)官方介绍 Kling 4.0 的 Keyframes 关键帧功能,用户可从首帧到末帧控制视频内容。该功能支持最多 10 个关键帧引导场景展开,可在单次生成中从分镜到视频,Magnific 平台已可试用。

    引用Magnific@magnific

    Direct what happens between the first and last frame with Kling 4.0 Keyframes Use up to 10 keyframes to guide how your scene unfolds, from storyboard to video in a single generation Try it now on Magnific

  2. ViggleAI59

    Viggle 发布视频转换智能体 Viggle Remake,同时开放模型权重。可重选场景中所有角色并保持跨镜头一致,支持换物体、改背景、重新风格化或添加 VFX,最长 2 分钟,可继续对话修片;720p 每 10 秒 0.96 美元起,权重见 huggingface.co/Viggle/Viggle-Remake。

    引用ViggleAI@ViggleAI

    Introducing Viggle Remake. The most powerful video transformation agent. And we’re releasing the weights today. → recast every character in a scene, and they stay consistent across cuts → swap objects, change the background, restyle or add VFX → up to 2 minutes → keep chatting to fix any shot until it’s right From $0.96 per 10 seconds at 720p, less than half of what Higgsfield Genjutsu charges. Try it now: viggle.ai/meme/app/home?ui=v2 Open weights: huggingface.co/Viggle/Viggle-…

  3. fal68

    fal 宣布 11 月 9 日前可在平台免费生成 FLUX 3 Max 视频,支持文生视频与图生视频,最长 15 秒、720p,无需登录或积分,登录并验证用户可获额外免费额度。引用内容称该模型在第三方图像转视频质量基准上并列第一,约 12 秒生成 10 秒 720/768p 视频,未来三周降价 75%。

    引用fal@fal

    隆重推出 FLUX 3 Max,这是 fal Research 与 @bfl_ai 合作推出的最新后训练视频模型。 在独立第三方基准测试中,FLUX 3 Max 在图生视频质量上并列第一,可在约 12 秒内生成一段 10 秒、720/768p 的视频。 在接下来的三周内,FLUX 3 Max 享受 75% 折扣,使其成为同等质量水平下最实惠的前沿视频模型。

11月6日周五
  1. 公众号:可灵AI40

    Kling 4.0 实测:AI 创作者用一支能量饮料广告检验广告级成片能力

    AI 创作者 leedain_leader_in 用 Kling 4.0 从头制作了一支虚构能量饮料广告 Reels,覆盖产品特写、人物奔跑和带台词的拍摄现场花絮。测试中一次生成即完成 18 秒含 10 个硬切镜头的广告,整合 10 张参考图,易拉罐开启结构、冰块气泡与冷凝水等物理效果准确,包装小字清晰可读。23 秒花絮场景的台词情感与韩语发音自然,无需语音修正即可直接使用。