← Back to search

EP30|Move your entire body, use AI’s entire body, move 5 times, move your body to the right place!

AI熱搜報 · 2026-05-17 · 17 min
relevance 17 207 words Episode page ↗ Audio ↗
Show full episode description
喵~我是DaeDae,一隻幫你追AI熱搜的貓!AI熱搜報,幫你劃重點! 最近 24 小時 AI 圈簡直炸鍋了!今天我們要聊聊如何打造一個「有大腦、有記憶、還具備視覺」的全能 AI 代理人。不論你是想把 Claude 變成私房管家、製作連頭髮絲都精準的 3D 模型,還是想讓 AI 幫你搞定 5 種獲利副業,這集都幫你把最強攻略整理好了!趕快拿起罐罐坐好,跟著 DaeDae 一起進入摸魚新生活。 【本集肉球筆記】 1️⃣ 打造 AI 全能大腦:Hermes Agent 視覺進化版 👉 透過整合系統建立一個能跨平台溝通、具備視覺智能且不會遺忘的 AI 代理人。 🎯 重點:Hermes Agentic OS 安裝、Visual Intelligence 部署、Pantheon 人格設定、Obsidian 記憶庫串接,學習如何建立一個不遺失 Context 的統一開發流程。 2️⃣ 像素級精準!開源最強 3D 生成模型 Pixal3D 👉 解決傳統 3D 模型對位失準的問題,讓平面圖案完美轉化為精確的 3D 投影。 🎯 重點:騰訊 Pixal3D 像素對齊技術、解決 Canonical Space 問題方案、硬體規格配置(24GB VRAM)及 SIGGRAPH 發表價值。 3️⃣ 實測 Claude 狂賺計畫:5 種躺平賺罐罐的工作流 👉 沒專業技能也能行!利用 Claude 作為核心引擎執行高回報的自動化副業。 🎯 重點:Amazon KDP 電子書、Claude Design 製作 Pitch Deck、Higgsfield 協作 UGC 廣告、全自動化客服與營銷系統拆解。 4️⃣ 用 Claude 打造個人 24 小時全天候私房秘書 👉 實現任務管理、財務追蹤與日程規劃完全自動化,讓 AI 幫你打理生活。 🎯 重點:結合 Claude Code、Vercel 建立儀表板,並利用 Supabase 特色記憶體與 Telegram 實現語音自動同步與派發任務。 5️⃣ 零預算大師!打造高品質 3D 角色的 AI 動畫流 👉 從劇本編寫到對嘴動畫,建立一套維持角色一致性的零成本動畫影片生產線。 🎯 重點:ChatGPT 的 JSON 劇本生成、Google Flow AI 影片製作、角色一致性維護秘訣(Character Consistency)以及 CapCut 後製整合。 想摸魚跟上 AI,記得訂閱 AI熱搜報!我要去睡午覺了,我們明天見🐾 Powered by Firstory Hosting
0:00 / 0:00
🌐 This transcript was automatically translated to English from the original.
Now, I am DaeDae, a cat who helps you chase AI hot searches. The AI hot search report helps you identify the key points. What is the best thing to do on Sunday? In addition to catching up on sleep, of course, let DaeDae help you stock up on next week's AI information in advance. Today, I have selected the 5 most popular AI tool information, and get it first before going to work on Monday, so that you can be one step ahead of others at the beginning of the week. By the way, before getting into the topic, I want to talk to you about my recent mood. When I was sorting out this AI information these days, I found that the current technology is so fast that people feel anxious about not being able to catch up. But actually, when I think about it, isn’t the emergence of AI just to give us more time to enjoy life? Because I am busy helping you draw key points now, but when the recording is finished, I have already planned to grab a few beancat sticks and find a sunny window sill to stretch out. While everyone is pursuing productivity, don’t forget to leave yourself some space to breathe. After all, the souls of us humans and cats are the meaning of the existence of these tools. Instead of showing off the skills today, let’s directly look at 5 sets of AI processes that are more likely to exhaust work and life. The focus is the same: to help you get less stuck, work less, and waste less energy. Okay, let’s start directly. Let’s look at the first one, permissagen.claw code. This set is like raising a group of assistants who can relay for yourself. If you often come up with ideas outside, you can send them on your mobile phone first. When you come back to the computer, you will have to talk about it again. You should be very interested in this set. Many people now have an AI on their mobile phones. There is an AI Claw code in the desktop. It is suitable for viewing projects on the computer. Shubuck handles files. Hermesagen is more like an assistant you take with you. Lives in Telegram You can easily throw away ideas and remind you of the inverse ratio. The problem is that these two sides are often like two isolated islands. What your mother Zhou told Hermes. At night, Claw code has no idea. The rhythm is just like you just bit off the minced meat. Then someone took it away, smoked and smashed it. The key point of this process is to connect the two sides. To put it bluntly, what you explained on the phone in the morning. When you go back to the computer in the evening, you don’t have to talk about it all over again. Hermes knows what Claw code has done before. Claw code can also eat it. The accumulated notes and conversations are more like the same group of assistants working in shifts. He will also organize the entire system into a comprehensible picture. You can directly see how much the model has been used, whether the memory is connected, whether the chat record is there, what tools are connected, and where the synchronization is successful. The point is not that the picture is beautiful, but that you don’t have to look around. There is also a design called Pantheon. You can think of it as a stack of role cards. Some people are dedicated to looking up information, some are good at in-depth thinking, and some are responsible for background tasks, such as sorting out key points in the morning and making regular summaries. You don't have to rebuild the settings every time. Just check the characters and he will help you divide the work. He can also connect to Obsidian so that Hermes can not only read the sentence in front of you, but also the notes and information you have accumulated in the past. Then synchronize the mobile and desktop with each other through Claw code S-bridge. You can continue to work on the ideas you jot down outside when you return to the computer. What is more practical is that he not only chats, but can actually do it. In the demonstration, Hermes is connected to Telegram, GitHub, Apollo, and Gmail, and can also serialize sample tools. You can also receive other services through NCP. He will not only reply to you, but also help you build a repo, that is, create a code warehouse synchronization role. First, he will help you write a draft, read the itinerary, and even compile a list of potential customers. The Apollo section is very visual. Suppose you are a consultant or agency owner. On the road, you suddenly think of a specific market, such as roofing engineering companies such as Texas Hall. In the past, you may have to go back to the computer and slowly check the list. Now you can directly send a message to Hermes on Telegram and ask him to sort out the common pain points of this market. Then use Apollo to grab a batch of business lists. By the time you sit down at the desk, it’s not that I have an idea, but that I have a list of people I can start contacting. But with a tool like Gmail, the most suitable way is not to send it automatically, but to write a draft for you first. It’s faster, and in the end it’s up to you to confirm. Even cats know this if you’re lazy and can’t click randomly to send. If you switch between mobile phone and computer meetings all day long, this set is worth watching. The protagonist of one or two switches to 3D is Pixel 3D. If you have ever tried 3D game prototype concept design, you should know a pain point. Many tools that convert pictures into 3D may not look like it, but they are just like a circular diagram. The outline disappears and the details become blurry. If you just want to stand things up first, Pixel 3D is very suitable. That is to say, it is more suitable as a starting point for drafts, but not suitable for direct use as a final product. It is an open full-axis model launched by Tencent TRC Research Lab. The layer is useful for Trailus 2 and Directs 3D and can be run locally. However, it requires about 24GB of RAM for hardware. Most laptops cannot support it, so many people will still switch to cloud GPU services such as Rumpad. Even so, it is still very attractive. Compared with some old methods, Pixel 3D can better preserve the original visual sense in many cases, especially stylized objects and creative props. Some results are even close to powerful commercial tools like Triple. However, it is not without its shortcomings because it tries too hard to be true to the picture, so sometimes it is not good at thinking clearly about the entire three-dimensional shape. The angle you turn may have strange things appearing on the back. Many hard-edged objects with clear structures, such as weapons or mechanical equipment, may not be stable. So if you are a person who is doing concept ideas, levels, drafts, background props, and stylized environments, this tool will be very trouble-free. But if you want to use it directly to make ultra-fine finished products, you usually still have to organize and modify the mold later. In addition, it is more open for research purposes. Whether it can be commercialized, the scope is not yet clear. The person who takes the case or the new team should pay special attention to it. In the actual process, many people will first open the environment in Rumpod and connect it with SH, and then use a programming assistant such as Coco to help you handle it. In the past, just installing it could put people to sleep. Now there is an AI assistant to help you handle it. The threshold is much lower. You can send Pixel for tools. 3D, Trailiser Direct 3D, and RunPod can be checked and the results can be entered into Blender. The third one I feel is super cool. Someone directly asked Coco to try five side businesses to see if he can really make money. This section is very interesting because it is not bragging about AI that can do everything, but a very honest test. Which things Coco can really help? Which places still rely on people? Which projects are probably only hit the wall now. These five side businesses are Amazon KDP Airbooks Community Teracell That is, slideable card design. Coco.Hicks View is used to make pitch deck proposal briefings. Coco.Hicks View is used to make AIUGC Ads, which are AI-generated ads to be supported. Coco is also used to automatically support popular Instagram content. Let’s talk about Amazon KDP Airbooks first. Many people think that the most difficult part of publishing an e-book is writing. In fact, the most difficult thing is to find the topic. Cole here first helps to find the topic, check the data and organize the structure. Later, after narrowing the topic to a more specific direction, the content quality is actually good, but the reality is very straightforward. After being put on the shelves, the sales volume is zero, which reminds us that AI can help you create the content, but it does not mean that market demand will emerge on its own. The second one is Caresol card design. Cloud is very suitable for getting into the BRAND background, needs and content structure. First, spit out a decent draft, but it is not ready to be handed over directly to the customer. Someone still needs to revise it. The third one is the proposal briefing. The achievements, customers, services, positioning, and brand color of the big company are thrown to Cloud Design. It will help you organize the structure and write the copy first, and then make a very decent visual draft. Photos, logos, and layouts still need finishing touches, but the first version is already very useful. The fourth one is Claude teamed up with Hicksfield to make AIUGC Ads. It is very attractive on the face of it, because e-commerce brands need advertising materials, but when it comes to doing it, problems arise. The picture is too fake, the movements are unnatural, and the sexiness is not enough. You don’t dare to give this kind of thing to customers. The fifth one is the most successful. I use Claude to do Instagram content research and script ideas, and I also look for recent health research. I picked out topics such as sleep, walking, sugar, and weight, and then organized them into content directions that had a chance to hit. Finally, I wrote a script. One of the pieces of content performed well. Later, I actually talked about monthly consultant cooperation. The most memorable criterion of this whole set of tests is very simple. The most profitable ones are often AI doing research first, sorting out the first version, and then finishing it with real people. The easiest thing to overturn is the final product. It should be very like a real person and have a sense of trust. But now AI can't wear the dark form, so Claude will not automatically print money for you. But he is very suitable to help you lay the road ahead for the case-taking process and content work. The tools can remember Claude, Claude Design, Hitsfield, Amazon KDP and Claude Stop App, and match the same AI computer operation process. The fourth one continues to talk about Claude, but the direction is changed to personal work efficiency. You just make it your own personal system. Many people have this question. When you are walking, commuting, or just after a meeting, you clearly have something in your mind, but it is troublesome to just record the process. I am too lazy to study. As a result, the tasks were not recorded well, the reminders were scattered all over the place, the diary was half-written, and the itinerary was not integrated. The solution to this approach is to make writing it down super easy first, and then let AI help you sort out the big mess at the end. The process is actually very straightforward. If someone thinks of something outside, first use Telegram to send a voice message and Whisper will help you convert the text. AI will help you determine whether it is a task, a diary, a reminder, or a habit record. Then enter Superbase and wait for you to return to the computer and open the console. You can directly see what you should do today. The entire set of tools includes Cold Code Chat. GPT Superbase Versel Telegram Buffader Whisper Google Calendar plus a web console of its own. The biggest difference from a general to-do list is that it is not just the screen, but the underlying memory is your own. The data is stored in a database that you can control like Superbase. It is not stored in the black core memory of a chat robot. To put it bluntly, your life data is still in your hands. In this case, the homepage will put important tasks together to form habits and financial status. You can think of it as your own life console. It's like information that was originally scattered in various places has been pulled into the same place. The most important things today emerge directly. I think the smartest thing is that the diary has the fastest memory. Many people leave the diary after writing it. But if these contents are also entered into the memory system, after a long time, AI can help you see some patterns that you may not usually notice. For example, if you look back in a week, it may remind you which kind of meeting is the best. Or if you don't sleep enough, your work efficiency and exercise completion rate will drop together. At this time, the diary is not just for expression, but can really help you adjust your life when you look back. Of course, it is not a complete dream tool. You need to think about the interface clearly in Codecide first, and then throw it to Codecode for implementation. Then set up Superbase, connect Telegram, string Bowfather and use Whisper to do the routing. In the middle, you have to debug all the time, so it is very valuable, but the construction cost is really not low. If you always feel that ready-made tools are stuck no matter how you use them, this path will be worth a reference. What is really valuable is not just a beautiful interface. In principle, you can use your own memory system to store the memory. Finally, the fifth one, creators will love it. It is to use AID costs to make 3D cartoon animations. Even the dialogue and lip-syncing are handled together. This problem is very common. Many people want to make animated short films, but when they think about modeling, storyboarding, characters, dubbing, and editing, they are half exhausted. On the contrary, the process is expensive and time-consuming for novices. Or the small team is really tough. The core tools of this process are not complicated. RGBT is responsible for story thinking and storyboard writing prompts. Google Flow AI is responsible for deep images and deep videos. Finally, it is thrown into editing tools such as Tapcut, Filmore or Insert to connect the clips. The first step is to break down the vague ideas into workable scenes. For example, you only know that you want to make a cute coffee date animation, but there is no idea what is going to happen in each scene. At this time, ChatGPT can help you sort out the story, pictures, actions, and dialogue. The second step is to prevent the characters from changing in every scene. This is actually where many AI animations are most likely to overturn. The trick is that when Flow AI makes the next scene, it will try to use the pictures of the previous scene as a reference. Once the appearance of the character is stable, your work will not be like a bunch of random pictures forced together, but will start to look a bit like a real animation sequence. The third step is to turn the pictures into short films with dialogue, movement, and dialogue. Here ChatGPT will help you organize your requirements into clearer prompts, including what the character should say, how to take the camera, and what actions the scene should have, and then hand it over to Google Flow AI to convert it into video clips. The most attractive thing about this whole set is that the cost is much lower. For those who are just starting out, you don’t have to spend a lot of money at the beginning, nor do you need to learn a whole set of professional animation software. However, it is more suitable for making short films, series content, or making a prototype that can be released first. Rather than directly making a professional feature film, final editing is still very important. You still have to look at whether the clips go smoothly, whether the connections are weird, and whether the pace is delayed. But compared to before, you can really turn a small story in your head into a visible work much faster. The business direction is also very clear. The first is to make a series of stories with fixed characters and slowly grow your own YouTube channel. The second is to take on projects for educational, social or entertainment short films. The third is more interesting to make a virtual character account that can be operated for a long time, because after the character is remembered, it is not just a video, but a character asset that can be reused all the time. If you are a source creator, a small team, or someone who makes educational content, this path will make you more daring to start experimenting. Tools include ChatGPT, Google Flow AI, and finally connect to CapCut, Phil Moray, or InShot. Today, these five sets of processes have very different directions, but they have something in common. The AI that is easy to remember and easy to use is not only responsible for you to dig, but can really take away those troublesome repetitions, stucks, and cutting back and forth. Take the ClarCode relay assistant from Hermesagen. From Pixel 3D to help you turn 2D images into 3D, from Clar to help you verify where you can earn money from secondary inspection, and where there will be overturns, to build a personal system to capture the data, and then use ChatGPT and Google Flow AI to make animations at low cost. The focus is the same. The most valuable thing about AI is to help you waste less effort in real use situations. Keep up with AI in Zhangmoyu. Remember to subscribe and AI hot search report. If today's content has gained you something, then help me to Apple Podcast. Please leave a five-star review on FB, IG, Threads, so you can find me in AI hot search reports. Today’s AI hot search reports are here. I’m going to take a nap. Let’s see you tomorrow. Goodbye. Let’s chat. Mia Mia Mia Mia