Generate AI videos from Telegram messages with Nano Banana & Veo-3

How to use the provided n8n workflow (step‑by‑step), what matters, what it’s good for, and costs per run. --- What this workflow does (in simple terms) 1) You write (or speak) your idea in Telegram. 2) The workflow builds two short prompts: - Image prompt → generates one thumbnail via KIE.ai – Nano Banana (Gemini 2.5 Flash Image). - Video prompt → starts a Veo‑3 (KIE.ai) video job using the thumbnail as init image. 3) You receive the thumbnail first, then the short video back in Telegram once rendering completes. Typical output: 1 PNG thumbnail + 1 short MP4 video (e.g., 8–12 s, 9:16). --- Why this is useful - Rapid ideation: Turn a quick text/voice idea into a ready‑to‑post thumbnail + matching short video. - Consistent look: The video uses the thumbnail as init image, keeping colors, objects and mood consistent. - One chat = full pipeline: Everything happens directly inside Telegram—no context switches. - Agency‑ready: Collect ideas from clients/team chats, and deliver outputs quickly. --- What you need before importing 1) KIE.ai account & API key - Sign up/in at KIE.ai, go to Dashboard → API / Keys. - Copy your KIEAPIKEY (keep it private). 2) Telegram Bot (BotFather) - In Telegram, open @BotFather → command /newbot. - Choose a name and a unique username (must end with bot). - Copy your Bot Token (keep it private). 3) Your Telegram Chat ID (browser method) - Send any message to your bot so you have a active chat - Open Telegram web and the chat with the bot - Find the chatid in the URL --- Import & minimal configuration (n8n) 1) Import the provided workflow JSON in n8n. 2) Create Credentials: - Telegram API: paste your Bot Token. - HTTP (KIE.ai): usually you’ll pass Authorization: Bearer {{ $env.KIEAPIKEY }} directly in the HTTP Request node headers, or make a generic HTTP credential that injects the header. 3) Replace hardcoded values in the template: - Chat ID: use an Expression like {{$json.message.chat.id}} from the Telegram Trigger (prefer dynamic over hardcoded IDs). - Authorization headers: never in query params—always in Headers. - Content‑Type spelling: Content-Type (no typos). --- How to run it (basic flow) 1) Start the workflow (activate trigger). 2) Send a message to your bot, e.g. glass hourglass on a black mirror floor, minimal, elegant` 3) The bot replies with the thumbnail (PNG), then the Veo‑3 video (MP4). If you send a voice message, the flow will download & transcribe it first, then proceed as above. --- Pricing (rule of thumb) - Image (Nano Banana via KIE.ai): $0.02–$0.04 per image (plan‑dependent). - Video (Veo‑3 via KIE.ai): - Fast: $0.40 per 8 seconds ($0.05/s) - Quality: $2.00 per 8 seconds ($0.25/s) Typical run (1 image + 8 s Fast video) ≈ $0.42–$0.44. > These are indicative values. Check your KIE.ai dashboard for the latest pricing/quotas. --- Why KIE.ai over the “classic” Google API? - Cheaper in practice for short video clips and image gen in this pipeline. - One vendor for both image & video (same auth, similar responses) = less integration hassle. - Quick start: Playground/tasks/status endpoints are n8n‑friendly for polling workflows. --- Security & reliability tips - Never hardcode API keys or Chat IDs into nodes—use Credentials or Environment variables. - Add IF + error paths after each HTTP node: - If status != 200 → Send friendly Telegram message (“Please try again”) + log to admin. - If you use callback URLs for video completion, ensure the URL is publicly reachable (n8n Webhook URL). Otherwise, stick to polling. - For rate limits, add a Wait node and limit concurrency in workflow settings. - Keep aspect & duration consistent across prompt + API calls to avoid unexpected crops. --- Advanced: voice input (optional) - The template supports voice via a Switch → Download → Transcribe (Whisper/OpenAI). - Ensure your OpenAI credential is set and your n8n instance can fetch the audio file from Telegram. --- Example prompt patterns (keep it short & generic) - Thumbnail prompt: “Minimal, elegant, surreal [OBJECT], clean composition, 9:16” - Video prompt: “Cinematic [OBJECT]. slow camera move, elegant reflections, minimal & surreal mood, 9:16, 8–12s.” You can later replace the simple prompt builder with a dedicated LLM step or a fixed style guide for your brand. --- Final notes - This template focuses on a solid, reliable pipeline first. You can always refine prompts later. - Start with Veo‑3 Fast to keep iteration costs low; switch to Quality for final renders. - Consider saving outputs (S3/Drive) and logging prompts/URLs to a sheet for audit & analytics.