I built a Telegram bot that controls ComfyUI video generation from my phone – approve or regenerate each shot with one tap

A developer has created a novel workflow solution that liberates AI video creators from being tethered to their workstations. By building a custom Python pipeline, LooPene44 connected the popular node-based interface ComfyUI to the Telegram messaging platform via its Bot API. This setup allows users to define a video scene in a JSON file—specifying parameters like StartFrame, prompts, CFG scale, and steps—and then initiate the generation process remotely. Once a shot is complete (taking approximately 130 seconds on an RTX 5070 Ti GPU), the system automatically sends both a preview frame and the full RIFE-interpolated 32fps MP4 video directly to a Telegram chat. The user then receives two simple buttons: one to approve the shot and proceed, and another to regenerate it with a new random seed.

The technical stack is deliberately lightweight, using Python with the `requests` library for API calls and `ffmpeg` for frame extraction, avoiding complex webhooks in favor of simple `getUpdates` polling. The AI generation itself leverages the Wan 2.2 I2V 14B model in a GGUF quantized format (Q6_K), run through a dual KSampler setup for managing high and low noise levels. The core innovation is the shift from passive waiting to active, asynchronous review and decision-making. This transforms video creation into a mobile-friendly, iterative approval process where creators can be productive from anywhere, significantly reducing downtime and manual intervention. The project demonstrates a growing trend of developers building bespoke automation and interface layers on top of powerful but complex AI toolkits to streamline creative workflows.