What’s up AIS Community! Just posted a video of a live demo showing how to automatically recreate viral YouTube thumbnails using Google Nano Banana, the world’s top-ranked image editing AI—end-to-end—fully automated in n8n with Telegram. What I cover: - Picking winning thumbnails (VidIQ outlier scoring) - Prompt → recreate with Nano Banana + Lora-trained likeness - Real-time remix via Telegram bot (natural language requests) - Batch generation + Google Drive project folders - Live thumbnail upscaling + Canva polish workflow - Cost breakdowns vs hiring a designer Key findings: ✅ Packaging wins: copying proven YouTube thumbnails = higher CTR (Legit one of these thumbnails got me my highest viewed vid on YouTube) ✅ Fully automated: Telegram kicks off, edits, and delivers Drive links ✅ Cheap iterations: 8+ thumbnails in ~2 minutes ✅ Template-first workflows = predictable results, easy retries ✅ Works for faceless creators, agencies, and solo YouTubers Caveats (read this): ⚠️ Subject likeness improves with a Lora model trained on 20–30 images ⚠️ Some thumbnails need light Canva/Photoshop tweaks post-gen Why it matters: If you’re building a channel or running client content, thumbnail automation is leverage. It lets you remix what’s already working, run multiple variants, and save hours (and $$) on design cycles. @Nate Herk - I’m sorry… I just had to do it 🤣 What are you gonna remix? ~ ps: full n8n + template pack is free inside the video description

Hey everyone! I just published a live demo showing how to generate AI avatars of yourself and turn them into UGC ads or your own content—end-to-end—using Nano Banana + Veo3 Fast + ElevenLabs, all orchestrated in n8n with Telegram. What I cover: - Prompt → image edits with a Nano Banana prompt agent + rectification loop - Live UGC ad generation (image → Veo3 Fast video) - Voice cloning in ElevenLabs (custom vs default) - Auto splicing with Shotstack (no Premiere/CapCut needed) - Human-in-the-loop approvals via Telegram - Organized outputs to Google Drive (images, audio, final video) - Cost breakdowns vs Synthesia/HeyGen and when to choose what Key findings: ✅ “You” on camera, anywhere: avatar stays consistent across scenes ✅ Hands-off: Telegram kicks off, approves, and delivers Drive links automatically ✅ Cheap runs: ~¢4/image (edits) and ~¢40/8s Veo3 Fast clips ✅ Template-first prompts = higher hit rate, faster retries ✅ Works for product UGC, memes, explainer shorts, and ad iterations Caveats (read this): ⚠️ Character consistency improves with better refs + tighter rectification prompts ⚠️ Logo/brand swaps sometimes need 1–2 rejection reasons to nail it Why it matters: If you’re producing client content, your own content or ads, this lets you prototype 3–5 variants in minutes, keep costs transparent, and stay inside one orchestration layer. I just had to make a breaking bad skit of myself with this automation 🧪😆 what will you make? ~ ps: full n8n + prompt templates are free inside the video description

Hey everyone! I just published a live demo showing an AI agent that designs and deploys complete N8N workflows from one prompt—no manual node wiring. What I cover: - Voice or text prompting via Telegram to brief the agent - End-to-end build: RSS → HTTP fetch → LLM summarize → Gmail send - How the N8N MCP (Model Context Protocol) lets the agent read node docs & call the right tools - Automatic cost + token tracking in Google Sheets (duration, model, spend) - Setup steps, required creds, and self-hosted N8N note for the MCP community node - Real constraints & failure modes (and how to work around them) Key findings: ✅ You can go from idea → working workflow in ~3 minutes ✅ Hands-off deployment: the agent returns a direct workflow URL inside your N8N ✅ Transparent costs: my demo run was ~$0.40–$0.50 for a full build (Shoutout to @nateherk for the costing idea) ✅ Best results with flagship models (e.g., GPT-4.1 / GPT-5); minis often break schemas ✅ Using MCP inside N8N beats paying for separate desktop MCP connectors long-term Caveats (read this): ⚠️ Great for simple, automations (Sheets/Gmail/triggers). ⚠️ Complex, multi-branch SDRs or heavy repurposing pipelines still need human oversight. ⚠️ Stick to supported tools & strict schemas to avoid the “expected input did not match” errors. Free stuff(in the video description): 🎁 Complete template pack (main workflow + deploy subworkflow + duration calc) 🎁 Google Sheets tracker for tokens, model, run cost If you’re building client automations, this is a force multiplier for rapid prototyping and delivery—without ballooning tool costs. 👇 Have you tried the N8N MCP yet? What would you have it build first?

Hey everyone! Just published a deep-dive video testing both models in real automation scenarios that I’ve actually used in client work. What I tested: - Customer support agent with RAG implementation - Tool calling and reasoning capabilities - Cost vs performance analysis - Real-world use cases (password resets, billing issues, API troubleshooting) Key findings: ✅ GPT-5 consistently outperformed GPT-4 in complex reasoning ✅ Better tool selection and multi-step problem solving ✅ Actually cheaper per token for the quality you get ✅ Superior at handling ambiguous customer queries The results were honestly shocking - GPT-5 caught errors and provided solutions that GPT-4 (and even I!) completely missed, even with identical system prompts. For anyone building/using customer support agents or complex automations, this could be a game-changer for workflows Free N8N templates included Have you started testing GPT-5 in your agency builds yet?! 👇

What's up everyone, just dropped a new YouTube video breaking down how I qualified 50+ inbound leads in production (and not ONE realised they were speaking to AI). Plus how you can get a custom GPT I use Just a quick one on the system: 1. 60-second speed-to-lead - Calls new enquiries instantly, boosting conversions by up to 21×. 2. Human-level conversations - Natural, unscripted flow that passes the “is this a bot?” test every time. 3. Consistent qualification - Captures the right info every time with structured prompts and scoring. Check out the video as I’m giving away the exact n8n workflows, Vapi prompts, and even a custom GPT I use to generate production-ready voice agent scripts 🤞🏾