Blog

  • AI Receptionist: 把 AI Agent 放進 WhatsApp 群組:Skill-Wise 設定教學(Manus、ChatGPT、Zo Computer)

    Channel: 阿石OMP

    Introduction

    The video from 阿石OMP demonstrates how to place an AI agent inside a WhatsApp group using the Skill‑Wise platform. The presenter shows a live demo where an “OMP Demo” agent answers a question about yesterday’s event attendance (97 %) and explains the approval flow that lets a colleague tag the agent and require an owner emoji before the agent can access company data.

    What the Video Covers

    • Why a 1:1 WhatsApp chat with an AI is less useful than putting the agent in a group where teammates can @‑mention it.
    • A live demo of the OMP Demo agent responding to a query about event attendance and the optional human‑approval step.
    • The end‑to‑end setup process for connecting Manus, ChatGPT (paid), or Zo Computer to a dedicated WhatsApp number via Skill‑Wise.
    • Technical detail: a WhatsApp message creates a Manus task; when the task finishes, the reply is returned to WhatsApp.
    • Beta pricing information, the free‑for‑~100‑users offer, and the promotional pitch for CEF‑funded courses and related tools.

    Step‑by‑Step Walkthrough

    1. Create a Skill‑Wise account – sign up for free at skill‑wise.ai.
    2. Connect an AI model
      • Choose ChatGPT (requires a paid subscription), Manus, or Zo Computer.
      • If using Manus, paste the Manus API key obtained from the Developers portal.
      • Zo Computer is noted as the preferred option for fine‑grained permission control.
    3. Add a dedicated WhatsApp number
      • Do not use your personal WhatsApp account.
      • Register a new WhatsApp number (or a secondary SIM) and link it via the QR code shown in Skill‑Wise (WhatsApp Web flow).
    4. Configure approval settings
      • Set the default behaviour: the AI ignores requests until it receives explicit permission.
      • Define per‑colleague rules – either auto‑allow or require an owner emoji (e.g., 👍) before the agent may access CRM or other internal data.
    5. Invite the agent to a WhatsApp group
      • Add the dedicated WhatsApp number as a participant in the desired group.
      • Team members can now @‑mention the agent to ask questions.
    6. Observe the interaction flow
      • A user’s @‑mention triggers a WhatsApp message.
      • Skill‑Wise forwards the message to the selected AI (Manus/ChatGPT/Zo) as a task.
      • When the AI finishes processing, the result is sent back to WhatsApp as a reply.
      • If approval is required, the agent waits for the owner’s emoji response before proceeding with any data‑access actions.

    Tools, Numbers, and Key Details

    • Demo answer: the agent reported yesterday’s event attendance as 97 %.
    • Skill‑Wise beta: free for the first ~100 friends/users; potential future charges if server costs increase.
    • Supported AI connections: ChatGPT (paid), Manus (API key required), Zo Computer (preferred for granular permissions).
    • WhatsApp integration: uses the standard WhatsApp Web QR process to link a dedicated number (not a personal account).
    • Underlying tech: WhatsApp message → Manus task → task completion → reply returned to WhatsApp.
    • Additional mentions: Whisper ASR was used for transcription (no YouTube captions); the video ends with a pitch for CEF‑funded AI marketing/office courses and related tools such as Manus, VPN, and Alibaba office solutions.

    Caveats / Hype Check

    “將 AI 放進 WhatsApp Group 上是比較實用的”
    “我們建議你是使用一個新的 WhatsApp 號碼… 不要連接你自己的個人號碼”
    “預設 AI 是不會應付他們的,所以你一定要設定權限給它”

    The presenter notes that Skill‑Wise is still in beta, with a free‑for‑~100‑users offer that may transition to a paid model later. The transcript was generated via Whisper ASR, which can introduce noise (e.g., mis‑recognised terms). No independent security audit of the WhatsApp‑to‑AI bridging process is mentioned. The video concludes with promotional CTAs for courses and related tools, which should be evaluated separately from the core functionality.

    Concrete Next Actions

    • Evaluate Skill‑Wise (or a comparable WhatsApp‑group AI bridge) for an AI receptionist or HK‑based operations workflow, focusing on mandatory human approval before any CRM write or data‑access action.
    • If exploring alternatives, compare Skill‑Wise’s multi‑user group + approval model to tools like Speedy WhatsApp computer‑use or Grok Bot, noting differences in permission granularity and pricing.
    • Set up a test environment using a fresh WhatsApp number, connect a preferred AI (e.g., Zo Computer for fine‑grained control), and run a small internal pilot with a handful of colleagues to verify the approval flow and response latency (approximately 10 seconds in the demo).
    • Document any observed ASR‑related transcription issues and assess whether they impact the reliability of the agent’s answers in your specific use case.
  • The Drop Drive: 1 Secret Prompt to Create Unlimited Motion Graphics Videos with AI (FREE)

    Channel: Biet Thinks

    Introduction

    The video from Biet Thinks presents a workflow that combines a Claude Project, a master prompt, Google Flow, and CapCut to produce motion‑graphics videos without paying for premium AI credits or needing After Effects expertise. The approach is positioned as a free method for creators who want to turn any script into a series of short clips and then stitch them together for faceless explainers or Shorts.

    What the Video Covers

    The presenter explains how to set up a Claude Project with a master instruction set stored in the project description. Once the project is ready, typing “start” triggers the Claude model to lock in visual style parameters (such as aspect ratio, clip length, colour palette, and audio cues). After the style is fixed, the user pastes a script—generated elsewhere, for example with ChatGPT—and Claude outputs a numbered list of clip‑by‑clip prompts that describe each scene in detail.

    Each prompt is then fed into Google Flow, where the user selects the Video mode, sets the aspect ratio to 16:9, chooses the Omni 1.1 flash model, and selects 720p resolution. The system consumes Flow credits to render each 10‑second clip. To maintain visual continuity between clips, the creator captures the last frame of the previous clip (using Shift+Win+S) and re‑uploads it as an image ingredient for the next prompt, ensuring that lighting, colour grading, and background music remain consistent.

    Rendered clips are downloaded at 720p and assembled in CapCut’s free tier. Because Flow already embeds background music, the only post‑production step recommended is adding a voice‑over track. The presenter demonstrates the pipeline with a seven‑clip example, showing how the final video looks after editing in CapCut.

    Step‑by‑Step Walkthrough

    • Create a Claude Project (e.g., name it “motion graphics”).
    • Open the project settings and paste the master instructions from the video description into the Project Instructions field.
    • Type the word “start” in the chat to initialise the style‑locking phase.
    • Confirm the style outputs (cream/minimal/luxury look, 16:9 aspect ratio, 10‑second clip length, no branding, soft whoosh on first words).
    • Paste your full script (the video shows a script generated by ChatGPT) into the chat.
    • Claude returns a numbered series of visual prompts, one per clip (the demo shows clip 1 of 7).
    • For each numbered prompt:
      • Open Google Flow, choose Video → ingredients.
      • Set aspect ratio to 16:9, model to Omni 1.1 flash, resolution to 720p.
      • Paste the clip prompt into the ingredient box.
      • If it is not the first clip, add a screenshot of the last frame from the previous rendered clip as an image ingredient to preserve continuity.
      • Generate the clip; each 10‑second segment costs approximately 15 Flow credits.
      • Download the resulting 720p video file.
    • Repeat until all clips are rendered.
    • Import the downloaded clips into CapCut (free version).
    • Arrange them on the timeline in the correct order.
    • Add a voice‑over track if desired; background music is already embedded in the Flow clips.
    • Export the final video from CapCut.

    Tools and Numbers

    • Google Flow provides 50 free credits per day, refreshed every 24 hours.
    • Each 10‑second 720p video generated with the Omni 1.1 flash model consumes about 15 credits.
    • The example workflow uses seven clips, which would require roughly 105 credits (7 × 15).
    • To stay within the daily free limit, the presenter suggests using multiple Google accounts (four accounts were mentioned) to accumulate the needed credits over several days.
    • Tools named in the video: Claude Projects (for prompt generation), ChatGPT (for script drafting), Google Flow AI (video synthesis), Omni 1.1 flash (the specific model within Flow), and CapCut (free video editor for assembling clips).

    Caveats and Hype Check

    The master prompt that drives the Claude Project is not publicly available in the video; it is gated behind the video description or a linked resource, meaning users must obtain it from the creator to replicate the exact workflow. Relying on multiple Google accounts to harvest free Flow credits may conflict with Google’s Terms of Service, which prohibit creating accounts solely to bypass usage limits. While the presenter describes the method as offering “unlimited” motion‑graphics videos, the reality is that each clip still consumes Flow credits and requires manual editing time in CapCut. No data is provided on audience retention, view counts, or monetisation results for videos produced with this pipeline, so the claim of unlimited output should be viewed as a theoretical capability rather than a proven, scalable business model.

    “All you have to do is paste your script. The master prompt automatically analyzes your script and creates the scene‑by‑scene visual prompts… Everyday, you can use 50 Google Flow credits for free.”

    Concrete Next Actions

    • Locate the master prompt in the video description (or the linked resource) and copy it into a new Claude Project’s Instructions field.
    • Test the pipeline with a short, low‑stakes script (for example, a 150‑word explainer about pickleball rules) to verify that Claude generates usable clip prompts.
    • Run the first prompt through Google Flow using a single account to confirm the credit cost and output quality; adjust the style lock if the results deviate from the desired cream/minimal/luxury look.
    • If more clips are needed than the daily 50‑credit allowance permits, plan to spread rendering across multiple days or across additional Google accounts, while reviewing Google’s Terms of Service to ensure compliance.
    • Download the rendered 720p clips, import them into CapCut, add a voice‑over, and export the final video to evaluate the total production time and visual consistency.
    • Document the credit consumption and editing effort for your test run to decide whether the workflow meets your content‑creation volume and budget requirements.
  • Hello world!

    Welcome to WordPress. This is your first post. Edit or delete it, then start writing!