Message a description, get a picture back in Telegram

Send your Telegram bot a description - 'a fox reading a map by lamplight' - and it draws that picture and sends it back to the same chat, as a photo captioned with what was asked for.

Every request goes through a safety check before anything is drawn. A model is asked whether the description is one this bot should draw, and the picture is only made when the answer is a plain ALLOW. It fails closed: a refusal, a garbled answer, or a model that cannot be reached all mean no picture, and the chat is told why. The check refuses real, named or identifiable people, sexual content, violence, hateful content, anything meant to pass as a real document or a real company's branding, and instructions for making a weapon or a drug. It is a filter rather than a guarantee - a determined sender can word their way past a model - so keep the bot to chats you trust.

Pictures go back only to the chat the request came from. Nothing is posted anywhere else and nothing is stored in your workspace beyond the generated file itself.

To set it up, paste the ID of the chat to answer onto the trigger step. Left blank, the trigger fires for every message your bot receives in every chat it has been added to, so fill it in to keep the bot to one chat. Then fill in 'pictureStyle' with the look you want - 'digital-art', 'photographic', 'comic-book' and 'line-art' all work - and leave 'pictureShape' at 1:1 or change it to another of the shapes the generator offers. A style or shape it does not know stops the run with a message naming the step, rather than failing on every picture.

This needs a connection to an image generator, an image host and a model. The generated file is saved to your workspace and a copy is hosted publicly for a week so Telegram can fetch it - Telegram will not accept an address it cannot reach itself, so the workflow checks the address before sending and sends a plain message instead if it is not a real public link.