Dima T. @Dimentorio

The authors’ ratings: method 1 got a B, method 2 from Mark an A, method 3 from Iman a B, method 5 an A with a big caveat, method 6 a B, and method 4 they placed somewhere between B and C. They didn’t reveal the seventh method, but both would pick it if they were starting from zero. They called it the main opportunity right now. It will be covered on November 7 at the free 5-day AI Income Challenge, link in the video description. The overall takeaway: most of these ideas are services, so they’re capped by your time. Scale starts where an agent does the work. 作者的评级:方法 1 是 B,Mark 的方法 2 是 A,Iman 的方法 3 是 B,方法 5 是 A 但有重大保留,方法 6 是 B,方法 4 被放在 B 和 C 之间。 第七个方法他们没有公开,但两人都说,如果从零开始,都会选它,并称之为当下最大的机会。11 月 7 日的免费 5 天 AI Income Challenge 会详细讲解,链接在视频简介里。 总结:这些点子大多是服务型的,收入受限于你的时间。规模化从让智能体干活开始。

Method 5: AI ads for TikTok and Instagram. You upload a model avatar and product photos, describe the actions and the script, and generate a UGC-style clip. Pay runs $100-600 and up per video, and a monthly package brings several thousand dollars per client. But it’s the most overheated niche: making a pretty clip is easy, while making an ad that actually sells has to be learned. Method 6: support chatbots on Tidio or Chatbase. You feed the bot the website and the return policy, write the rules, and above all test it. A bot without training gives wrong answers, which is worse than having no bot at all. 方法 5:TikTok 和 Instagram 的 AI 广告。 上传模特形象和产品图,描述动作和台词,生成 UGC 风格的视频。每条视频报酬 100 到 600 美元甚至更高,按月打包能从单个客户拿到几千美元。但这是最拥挤的赛道:做出好看的视频很容易,做出真正能卖货的广告却需要学习。 方法 6:用 Tidio 或 Chatbase 做客服聊天机器人。把网站和退货政策喂给机器人,写好规则,最重要的是测试。没经过训练的机器人会给出错误答案,这比没有机器人还糟。

Almost everyone trying to make money with AI makes the same mistake: they build their own tool or startup. 99% of those projects will get crushed by companies with millions in funding. Iman Gadzhi and Mark broke down 7 ways to earn from AI differently. Every business knows it needs AI, but almost nobody knows how. The money goes to people who apply ready-made tools inside someone else’s business. Breakdown 👇 几乎所有想靠 AI 赚钱的人都犯同一个错误:自己做工具或创业。这类项目 99% 会被融资数百万的大公司碾压。Iman Gadzhi 和 Mark 拆解了 7 种不同的 AI 赚钱方式。每家企业都知道自己需要 AI,但几乎没人知道怎么用。钱属于那些把现成工具用在别人业务里的人。详细拆解 👇

@abxxai @FishCreativeHQ ai video generation is getting scary good, the lighting and camera movement look like an actual commercial 👀

@DCyberCherries not every project needs millions of players to be a success, sometimes making your childhood dream real is enough

@vib3coded game dev is changing fast — what used to take a team days to prototype can now start with one prompt 🚀

HE LEAKED HIS CLAUDE CODE WORKFLOW FOR AI DROPSHIPPING: FROM PICKING A PRODUCT TO A FINISHED VIDEO Jake Dabrowski shows how he builds organic dropshipping while handing every routine step to Claude Code inside Visual Studio Code. It's not a collection of separate tricks but a standardized process that runs the same way from start to finish. It has three blocks. First, product selection. Then generating custom assets for that product. And finally, editing high-quality promo videos through to the final export. The core idea is that the advantage comes not from the toolset but from the order. When every step repeats the same way, videos come out faster and in a consistent style instead of starting from zero each time. For organic dropshipping that decides everything, because you need content in volume and on a regular schedule. The video runs 31 minutes, so there are surely details on every stage inside. I worked only from the short description, so there are no specific tools, prices or numbers here. Send the transcript with timestamps and I'll rework this into a step-by-step guide with those details.

GUIDE: HOW TO MAKE CLAUDE EDIT YOUR VIDEO IN CAPCUT: 117 CUTS, GRAPHICS AND MUSIC WITHOUT A SINGLE CLICK The author handed Claude the entire edit of a talking-head video: 117 cuts, motion graphics, music, zooms and a color grade. Everything is built in a real CapCut project, so you can tweak anything before exporting. A video with Claude-made graphics gained almost 600,000 views in a week and a half. On a 51-minute recording full of mistakes plus a screen recording, Claude produced an 18-minute version with precise sync. STEP 1. PREPARE THE CAPCUT PROJECT Create a new project and drop the raw recording and any extra material into the media bin: images, screen recordings, video. Give the project a name you'll remember. Close the project, make sure the name shows in CapCut's main window, then close CapCut completely. STEP 2. THE EDITING PROMPT You need desktop Claude and desktop CapCut. In the prompt the author specifies: the project name; sync screen recordings to the video at the right moments; cut repeats, errors, long pauses and anything out of place; cut between him and the screen recording when needed; add an adjustment layer with saturation, contrast and lowered shadows; add slow, smooth zooms on the camera when an intriguing point lands; take music from your creative folder (with the mood in brackets, for example "intrigue"); any generated assets must be written so CapCut can open them; don't use computer use or open CapCut, write the project files directly; ask clarifying questions before starting. The author works with Opus 5.5 on extra effort. Claude will ask for access to the CapCut folder and the drive with your footage. The edit took 58 minutes. Sometimes one or two files don't link: just relink them through CapCut's pop-up window. STEP 3. CONNECT HIGGSFIELD For graphics you need Claude, CapCut and Higgsfield. In your Higgsfield account, open the MCP section and copy the link. In Claude: Settings, Connectors, Add, Add custom connector. Name it Higgsfield, paste the URL, sign in. Now Claude can make motion graphics, images, video and even voice. STEP 4. REFERENCES FROM PINTEREST Search for documentary-style graphics, smooth UI animation, stop-motion or typography. Copy the link, download the video with any Pinterest downloader and upload it to Claude. STEP 5. THE GRAPHICS PROMPT Attach your references and write: use the Higgsfield Motion Editor MCP, create motion graphics for the edited video, place animations at the important points, first generate one animation for approval and ask questions about the visual direction. Once approved, make the rest in the same style. Files must be readable by CapCut, and at the end they need to be imported into the project at the right times. STEP 6. APPROVE THE FIRST, THEN THE REST Review the first graphic, adjust or approve it. Then Claude makes the rest and inserts them itself along with sound effects. STEP 7. ADD YOUR HUMAN TOUCH Before exporting, open the project and fix the small things. For example, a jump cut right before a graphic appears looks unprofessional: move the graphic slightly earlier and the cut becomes smooth. Claude may have missed a pause or a stutter, and you can tweak the color on the adjustment layer yourself. STEP 8. LET CLAUDE LEARN Close CapCut and ask Claude to re-analyze the project. It compares before and after and picks up your style: graphics, music, sounds, pacing. STEP 9. TURN IT INTO A SKILL Ask Claude to write a skill for future projects based on the chat and the workflow. To install it: Settings, Skills, Add, Upload a skill and drag the file in. The author put his ready-made skill in the description. HOW LONG IT TAKES The first project from scratch: about 3 hours start to finish, but only around 10 minutes of that is your active time. The rest is Claude editing and building graphics while you make tea. After the first project and with the skill installed, it takes noticeably less. WHAT IT COSTS No Higgsfield credits are spent: Claude writes code through the Motion Editor MCP. You need a Higgsfield subscription (to connect the MCP) and Claude Pro: $17 a month on the annual plan, $20 month to month, which is enough. The author estimates he saves 20-30 hours per project. TAKEAWAY Prepare the project, give Claude the prompt, approve the first graphic, fix small things by hand, send the changes back for analysis and save everything as a skill. Every next edit goes faster and lands closer to your taste.

Robotaxi makes the roads safer for everyone

@0xPrajwal_ the fact that one prompt can turn into a playable racing game is actually insane, the barrier to game dev is disappearing 👀

Claude Opus 5.5 vs GPT-6 Astra: building a 3D training dummy from scratch 👀 same $20 subscription same prompt same Blender MCP but the weekly quota usage tells a different story: → GPT-6 Astra: 12% → Claude Opus 5.5: 3% that’s 4x less quota used for Opus on this task even if Opus has been nerfed, the efficiency gap is hard to ignore 🔥

a 19-year-old student allegedly turned $68 into $750K with a trading bot built using Claude Code in just 48 hours 🤯 here’s the setup: → Claude Code handles the trading logic → Binance API streams live BTC prices every second → the bot monitors 50+ markets simultaneously → an iPad serves as a second screen for market monitoring → trades trigger when arbitrage opportunities appear the idea is simple: spot price discrepancies across markets and act before they disappear no staring at charts all day, no emotional decisions — just automated execution based on predefined logic but there’s a catch 👀 the claims of $6,732 in first-night profits and $750K in total earnings need independent verification. automated arbitrage also comes with fees, slippage, execution risks, and potential losses AI can help build the system, but it doesn’t guarantee an edge still, the bigger takeaway is fascinating: you can now prototype trading systems with AI coding tools that once required a much larger engineering effort

@PJaccetturo spending $2k a day on generations until one simple trick changes everything is a painful lesson 💀

@ZenithcoreAi the wild part is that you don’t need to build a startup anymore, you can just deploy bots and let them work 👀

10 Sonnet agents vs 1 Opus 5.5 — the results are wild same prompt, same Blender challenge: build a cozy village at sunset from scratch, with lights turning on one by one zero assets. everything generated through code 10 Sonnet agents → 77 minutes → ~$73 1 Opus agent → 87 minutes → ~$21 the multi-agent team finished just 10 minutes faster, but cost 3.5× more so the real question isn’t which setup is faster it’s whether those 10 minutes are worth an extra $52 👀