My faceless pages have done nearly 70 million views on TikTok and YouTube in about a month. I have not recorded a single video for them. In this video I show the back end in Claude Code. The skill is free, links below. Here is the flow in the video: Step 1, Setup: Open the Claude Code desktop app (VS Code works too), open the skill folder and tell Claude to set it up. It asks for three keys: the Higgsfield API, Apify and Blotato. It tests each key with a free call, then lists the social accounts you have connected to Blotato. Step 2, Connect your pages: In Blotato it is one button per platform. Log in with TikTok, log in with YouTube, done. That is what lets Claude post for you. Step 3, Research a niche: One command. I used oddly satisfying. It runs six YouTube searches plus an Apify TikTok scraper at the same time and comes back with the clips doing the numbers, each with its link. Not happy with what came back? Run it again until you find a style worth recreating. Step 4, Make the page (if you need one): No page yet? The skill gives you name ideas, checks the handle on both platforms, then makes a logo and a banner. Step 5, Recreate a clip: Pick a row and ask Claude to recreate it with Seedance 2.5 through the Higgsfield API. My first version was not great, so I told it what to change and the second one was much closer. Step 6, Post it: Tell Claude to post it through Blotato. It can drop clips into your posting times, or post one straight away. I posted mine unlisted first to check it landed. The real power is batching. Once your style is locked in, ask for a month of posts in one go and schedule them all. Setup guide, you only do it once: https://github.com/automatorsplus/faceless-page-factory-skill/blob/main/.claude/skills/faceless-page-factory/SETUP.md Resources: The skill: https://github.com/automatorsplus/faceless-page-factory-skill

I recorded one raw intro clip, pointing at where I wanted things to go. GPT-6 Astra and HyperFrames turned it into a finished edit: a title, cards, captions, the websites I've built sliding past and last week's video playing in a window. In this video I break down how it works, then run it live. Here's the five-step process: Step 1, Transcribe: It transcribes your raw clip word by word. You can use OpenAI's Whisper model or an external tool like ElevenLabs, which is a bit faster. Step 2, Cut: From the transcript it makes the cuts: silences, stumbles, the dead space at the start and the end. Step 3, Plan the beats: Every cue you say becomes a moment on screen. When I say the cards coming in from the left, cards come in from the left. If you record yourself pointing and saying it, most of the plan is already done. Step 4, Build with HyperFrames: HyperFrames is HeyGen's locally hosted video editor. Astra pushes the cut video and the beat plan into it and builds the motion layers on top. Step 5, Verify: It watches the result back. If something's wrong it replans the beats and rebuilds. In the live run I show the exact prompt: where the raw clip is, which cues I say, where my websites and last video live. I ask for a plan first. It comes back as a timed map of every cue and what goes there. I say go, it builds the edit. When I wanted one change, keeping my background when I shrink into the corner, I just asked for it in the chat. To get started you need the HyperFrames skill. Copy the GitHub link, paste it into your GPT-6 Astra chat and tell it to install the skill. At the end I show three shorts built with HyperFrames from my long-form videos. Fair warning: Astra burns through tokens. For the quality you get, it's worth it. Resources: HyperFrames → https://github.com/heygen-com/hyperframes

Anthropic and OpenAI shipped new models on the same day. This is the quick rundown: what each one costs now, how much smarter they are, and what it means if you use Claude Code or Codex. Both got cheaper. Opus 5.5 is $4 in and $20 out per million tokens, 20% less than Opus 5. GPT-6 Sol is $2 and $10, half the price of 5.6 Sol. On the Artificial Analysis Intelligence Index, an independent test, Opus 5.5 went up 7 points to 58, which puts it top of their leaderboard. Sol went up 1 point to 48. OpenAI went for cost, Anthropic went for cost and brains. The bit most people miss: Opus 5.5 on medium effort scores the same as Opus 5 on max, for 77% less per task. So stop running max effort on everything. I also gave both models the same prompt to build a landing page for Falco AI, so you can see what each one does with it. Sources: https://artificialanalysis.ai https://www.anthropic.com/claude-opus-5-5 https://openai.com/index/introducing-gpt-6-sol-and-luna/ Four minutes, all the numbers you need.

One product photo and one sentence of angle. That's the whole prompt. Claude Code writes the script, builds the cast, generates every frame, records the voice and stitches the finished ad. Same character in every shot, the real label intact, about $4 an ad on fal.ai. The hard part with AI video is that the character changes every few seconds. This skill fixes it by locking two images first, the world and the character, then attaching both to every frame it makes after. It face-checks every still against the master before it makes the next one, and it checks the product label before it spends a penny on video, because a clip costs ten times a still. It also catches its own mistakes. In the video it fails a clip because the drawer's mouth turned cartoonish, regenerates it, and carries on. You watch it happen. The skill is free and it's yours: https://github.com/automatorsplus/claymation-ads-skill --- send this prompt to Claude Code **copy below this line** I'm giving you a skill called Claymation Ads. Get the files with: git clone https://github.com/automatorsplus/claymation-ads-skill Follow the README to set it up. Recommend how it would best apply to our setup in plain language, and ask me a few questions to clarify my intent before you install it.

Every site Claude Code builds looks the same. Centred hero, three identical cards, same font, same buttons. If you build these daily you can spot one from a mile off. So can your clients. In this one I take a single prompt for a coffee roaster's landing page, build it with no skills installed so you can see the baseline, then run the exact same prompt through four different design skills. Same brief every time, only the skill changes. taste-skill: The anti-slop framework. Install it and run the same prompt. The fonts, hero, buttons and nav all come back different. It is the biggest single jump in the video. awesome-design-skills: Not one skill, a registry of 67 of them. Bento, brutalism, editorial and more. Each one is two files. I pulled brutalism onto the same coffee site and it came back as a completely different brand. Impeccable: The one to watch if you already have a site. Run the detector and it lists what reads as AI slop on the page, ten things on mine. Then run live and a bar appears in your browser where you click any element, get three variants and accept one. No going back to the chat. img2threejs: Give it a flat product photo and it builds a 3D model you can put on the page. Works best on a simple object on a white background. My first render was rough, the second was decent. Install any of them the same way: paste the GitHub link into Claude Code and ask it to install the skill. The four repos: https://github.com/Leonxlnx/taste-skill https://github.com/bergside/awesome-design-skills https://github.com/pbakaus/impeccable https://github.com/img2threejs/img2threejs If you have already built a site with Claude, run the Impeccable detector on it first. That alone is worth the ten minutes.