开启时会模糊预览图，关闭后正常显示
If current submissions don’t pan, probably going to opt out indefinitely: conference acceptance is increasingly becoming lower signal for me as a reader anyway.
Artisanal baker of reasoning models @pleiasfr
Когда ИИ пришёл за картинками из одноклассников, мы промолчали, но потом... Nano Banana Pro дорос.
Пример тут https://t.co/Q6VfNJsFVb
评论区一人一个 Claude Code AI 编程技巧， 我先来 🙋♂️ claude --dangerously-skip-permissions
🚧 building https://t.co/AJfZ3LMlgq https://t.co/606cFUoda3 https://t.co/s0m0tpQMDH https://t.co/UQ5vrrYdAG 🐣learning/earning while helping others ❤️making software, storytelling videos 🔙alibaba @thoughtworks
通义 Z-image 实测! 这个模型只有 6B! 效果逆天! 给大家带来通义 Z-image-Turbo 实测! 这是个文生图模型, 模型支持定义画面输出比例, 最高分辨率能接近 650 万像素(超过后不是不能生成, 只是效果变差), 甜区在400-500万像素(即生成得比较好). 这个模型最大的亮点是性能强劲的同时只有6B大小! 显存大一点的游戏显卡(16G)都可以运行! 如果用专业卡, 甚至可以实现秒级图片生成! 本次测试聚焦在电影海报生成(复杂场景还原+文本生成) 和表情包生成(人物表情控制上). 直接说结论, 我测试的电影海报生成场景直接秒杀了, 甚至你可以看我这个封面图, 马龙白兰度两张脸甚至都差不多能合到一起了, 牛皮! 当然这里还是要提一下我测试过程中发现的问题, 其实也是跟模型大小有关. 由于模型大小非常小, 所以它的 instruction 部分会非常小, 对指令的推理能力会非常有限. 如果使用复杂的抽象prompt, 比如"画一个黑板，然后请计算4的三次方加上1300，得到结果后写在黑板上", 这是肯定无法实现的, 直接要求模型"画一个黑板，上面写着1364", 会更靠谱. 当然成败都在大小, 6B的大小完全可以作为本地部署来进行动态生成图片了. 而且文本生成, 指令遵循效果都非常好. 可以预见到接下来 Z-image 的各种应用和 LoRA 会层出不穷. 总之我先赶紧本地存一份用起来. #zimage #通义实验室 #Nanobanana #文生图大模型 #KCORES大模型竞技场
大白鲨&教父海报
I’ve also seen this happen in my life over and over again. It’s fascinating! The latest example was with writing my newsletter. I started without having a clue on what to write about. All I knew was that I want to do it to get better at. And then slowly but surely, each issue got easier and easier to write. Imaging how this is going to be 12 months down the road! 🤩
I build stuff. On my way to making $1M 💰 My projects 👇
Stop saying Chat Control “passed.” It didn’t. It’s not law. We can still stop it. Wording it like that kills momentum and makes people think the fight is over. If you actually want to help: go to https://t.co/w2GO9SjWma and contact your MEPs now.
CPO @Cyfrin | Ex @Alchemy | Created @cyfrinupdraft and @AlchemyLearn | Robotics | Making web3 mainstream