AI text and image processing in Telegram

In FLOWIN, you can automatically process messages with AI before sending them to another Telegram chat or channel: rewrite text, edit photos, create illustrations and extract data from images. You write an instruction once, and it is applied to new messages in the selected forward.

How to apply an AI assistant to a forward

Before you begin, you need a configured forward in “Copy” mode, which allows you to change message content. The connection procedure is the same for all types of processing. The form screenshot shows settings for a photo, and ready-to-use tasks for different messages are provided below.

0. Open @flow_in_bot, send the /start command, and click “Control panel”.

   

1. Open “Menu” → “AI Assistants” and click “Create”.

   

   

2. In the “Mode” field, select the type of processing you need. In the “Task” field, describe what to do with the message. Click “Create”.

   

3. Click “Apply” next to the AI assistant you created. In the “Selecting forwards” window, select the forward you need and click “Save”. A “Forwards [1]” button will appear: the AI assistant is now applied to this forward and will process its new messages.

   

   

   

Practical examples

Each example specifies the mode, task and result. To use another example with an AI assistant already applied to a forward, change the assistant’s settings.

Recalculate prices in a product description

A supplier sends product descriptions with wholesale prices, and you want them to appear in your channel with a 20% markup. Select “Text processing” mode.

Task: “In the following message, increase all prices by 20%. Do not change product quantities or other numbers. Return only the modified message, without comments.”

In the original message, the product costs 1,000 rubles and there are 12 units in stock. After processing, the price becomes 1,200 rubles, while the quantity stays the same.

For other tasks, you can use instructions like these:

• “Translate the message into Chinese.”
• “Shorten the message, keeping the main point.”
• “Rewrite the message in a businesslike style.”

Style a cake photo

A bakery sends an ordinary photo of a cake. Your channel needs a shot with soft lighting and a tidy background, like a food magazine photo. Select “Image modification” mode.

Task: “Style the cake photo like a food magazine shoot: soft side lighting, a tidy background and a striking composition. Preserve the cake itself, its shape, frosting and decorations. No text.”

A photo arrives in the source with the caption “Strawberry cake from our bakery”. The destination chat receives the processed photo with the original caption.

You can replace the background or change the color scheme in the same way. For example: “Make the image black and white, preserving the composition.”

Create a news poster

A channel publishes local news. To automatically add a themed poster to each news roundup, select “Image generation” mode.

Task: “Create a poster based on the news roundup in the message. Combine the main topics in an editorial illustration. No text or logos.”

In our example, the source receives a sample roundup titled “City news”:

• A new science museum has opened.
• 300 trees have been planted in the city park.
• The outdoor concert season has begun.

The destination chat receives a poster based on these three events, with the original roundup preserved in the caption. For the next poster, simply send a new roundup: the task sets the style, while the message supplies the content.

Read a trading plan from a chart

You can add a text plan to a technical analysis chart: the ticker, trade direction, entry point, target and stop-loss. Select “Image analysis” mode.

Let’s use a practice chart for a fictional stock called DEMO. It shows a buy entry at 100 rubles, a target of 110 rubles and a stop-loss of 95 rubles. We need to transfer these labels into text and calculate the ratio of potential profit to risk.

Task: “Read the ticker, direction, entry, target and stop-loss from the chart. Calculate reward/risk. Use 6 short lines without markup. For missing data, write ‘not specified’.”

The result includes the values from the chart and a reward-to-risk ratio of 2:1: 10 rubles to the target and 5 rubles to the stop. The image and original caption are preserved, with the answer placed after the caption.

The levels in this example are marked on the chart itself. The task asks the model to read them and avoid inventing missing values. For another image, you can ask for a description of objects, text recognition or extraction of the data you need.

How to change the task and settings

1. Return to the “AI Assistants” list and click “Change” next to the AI assistant you need.

   

2. Make your changes and click “Save”. The updated settings will apply to new messages.

   

   

Mode

Determines what is processed and how the message changes. Four options are available:

• “Text processing” changes the message text or media caption: for example, it can translate, shorten or recalculate prices. Each message is processed separately. This mode does not combine neighboring messages into a digest or edit lettering inside a photo.
• “Image modification” edits an existing image according to the task and uses the result to replace the original. The message text is preserved and is not used as an instruction for modifying the photo.
• “Image generation” creates a new image based on the task and the message text. The original text is preserved in the caption. To generate an image while forwarding, you need a text message without photos, videos, other attachments or a link preview.
• “Image analysis” preserves the image and adds a text response after the original caption. If there is no caption, only the response will appear. The instruction comes from the “Task” field; the caption is not used as an additional task.

Modification and analysis support Telegram photos and static JPEG, PNG and WebP images sent as files. Images in an album are processed individually. Videos, stickers and PDFs are not supported in these modes.

Changing the mode changes the available models and selects the default model for the new mode. Your custom task text is preserved, so check that it suits the new type of processing.

Task

The instruction applied to each incoming message. It can specify the change you need, the style and the response format. For example, when generating a poster, the task describes the illustration’s style, while the specific news arrives in the message text.

Model

The “Model” field offers options for the selected mode. You can search for a model by name. Different models may follow the same instruction differently, so compare them using the same messages.

Description

You can leave a comment in this field to help you find the AI assistant in the list later. The description is optional and does not affect message processing.

Text processing parameters

In “Text processing” mode, there is a “Parameters” link next to the “Model” field. It opens the formatting and Internet access settings. This link is not available in image modes.

The “Formatting” field determines how text is sent to the model and how its response is processed:

• “Text” is selected by default. Styling is removed from the text being processed.
• “HTML” sends the message with markup and processes the response as HTML. This option allows supported formatting to be preserved if the model returns valid markup.

“Access to the Internet” is needed for tasks that require information from the web. It is not needed to recalculate a price or change the style of text already received.

Additional actions

The “More” menu lets you apply a saved AI assistant to other forwards, copy its settings or delete it:

• “Apply” opens the selection of forwards to process.
• “Duplicate” creates a copy of the AI assistant with its settings.
• “Select project” adds the AI assistant to a project.
• “Delete” deletes the AI assistant and its settings.

You can also open the forward selection using the “Forwards [N]” button, where N is the number of forwards. If there are no connections yet, “Apply” appears instead of the counter.