I don't know if you have encountered AI generated materials before. The generated video looks great, but it's full of typos I don't know what the common solution is now, Anyway, I have a headache. There may be some external tools, but I think they are still very inconvenient. Why can't AI directly make them right? Then my friend asked me to try Fat Goose AI. Using the same prompt to generate a steam eye mask, Figure 1 is generated by Fat Goose, and Figure 2 is generated by another AI. The video effect is similar, but the typo in Figure 2 is completely unusable. The process of generating videos with ordinary AI is to throw your prompt to a low-level model and wait for it to spit out whatever it says. Errors in spelling, proportions, and style deviation are all blind boxes. And xUbble added a layer in the middle. Fat Goose AI will continue to run experiments in the background, taking different models, prompts, and tool combinations for the same type of task and conducting reverse testing to determine who has the most stable output. The optimal set is packaged into an SOP and stored. SOP is actually the most effective fixed formula for this type of task using this combination. And the longer it is used, the more it understands how to do similar tasks. Just like an experienced traditional Chinese medicine practitioner who has mastered many prescriptions proficiently. XUbble is mainly divided into two parts: The Bubble Engine has been running experiments. For the same type of task, it will use different models, prompts, and tool combinations for reverse testing to determine who has the most stable results, and then package the optimal set into SOP for storage. SOP can be understood as the best fixed formula for this type of task using this combination. The other one is called Bubble Pilot, which is a conversation between the front-end and the user. For example, in the previous conversation with the user, a request was received: scan which smart money has been adding AI tokens in the past 24 hours. It will first determine what type of task this belongs to and go to the backend to see if there is a ready-made SOP. Hit it, I'll give you the best solution directly. Failed to hit, use the general model to barely run one first, and at the same time, write down your request and tell the Engine to practice this more in the future. So the longer you use it, the more it understands how to do your type of task. XUbble has two operating environments. one ️⃣ Bubble Computer is a cloud based project workspace. This is where the zero to one job of creating materials, creating dashboards, and creating web pages comes from. When encountering multi-step tasks, automatically open the sandbox and load tools as needed. You don't have to worry about the intermediate process. Research, generate, verify, and produce finished products in one go, I received an independent URL. Directly forward it to the customer or add it to the group. two ️⃣ Bubble Personal is in local mode. Tend to work in your existing environment. You can read files on your computer, operate browsers, use your applications, and check your schedule. The heavy tasks of installing software and modifying systems are run and destroyed in cloud containers. This machine only moves things that you have explicitly authorized and does not pollute your computer. When I first learned about this project, I was thinking about why I never thought there would be such a sophisticated entrepreneurial direction before It can only be said that the xUbble team's dappos understands the pain points of current users using AI. dappos itself is a mature and well-established web3 project, and its parent company DappOS has received investments from well-known institutions such as Sequoia China and Yzi Labs. I suggest everyone to try it out. When something new comes out, only by trying it out can you know if it's good or not. link：https://(dappos.com)/xbubble/en/invite? referral=z7t4ws