I put together another scenario that builds on the comment and DM auto reply flow I shared earlier, this one adds image and audio understanding to the mix. A few people asked if the AI could actually understand a photo or a voice note someone sends over DM instead of just plain text, so here’s how I set that up.
What it does
The scenario listens for incoming Instagram messages, and instead of only handling text, it can read images and audio that people send in and generate a relevant reply from them. So if someone sends a screenshot with a question in it, or a voice note asking about a product, the flow can actually understand the content and respond accordingly using Claude, ChatGPT, or Gemini.
How it’s structured
- A webhook module catches the incoming message from Instagram, whether it’s text, an image, or audio
- A router checks the message type and sends it down the right path
- Image and audio content gets processed and converted into something the AI model can understand
- The message, along with any extracted content, gets passed to Claude, ChatGPT, or Gemini with a prompt that shapes the tone of the reply
- A datastore module keeps track of conversation history so replies stay relevant across multiple messages
- The generated reply gets sent back through Instagram
Why this setup matters
Understanding images and audio is only half the problem though. The bigger issue is still getting that reply delivered properly within Instagram’s official API rules, especially the 24 hour messaging window. A lot of people default to something like ManyChat for this kind of thing, but ManyChat and most similar tools weren’t built as official Meta business partners, so they tend to route around the platform instead of through it. For the sending step here, I went with InstantDM again. It’s a Meta business partner and works directly through the official Messaging API, so replies actually land properly instead of getting blocked or flagged, even when the conversation involves richer content like images and audio.
Where InstantDM sits in the flow
Once the AI model processes the message, whatever came in as text, image, or audio, and generates the reply text, that text gets handed off to InstantDM to actually send. InstantDM manages the token refresh cycle and the messaging window logic behind the scenes, so the scenario itself stays focused on understanding the message and deciding what to say, while InstantDM handles getting it delivered.
Handling image and audio content specifically
The trickier part of this scenario is making sure image and audio content actually reaches the AI model in a usable format. Images get passed through as visual input so the model can describe or answer questions about what’s in them, and audio gets processed so the model can respond to what was actually said rather than just acknowledging that a voice note arrived. This matters a lot for use cases like product questions sent as photos or quick voice messages instead of typed text.
A few practical notes for anyone setting this up
- Make sure your Instagram account is connected as a business or creator account, since the Messaging API only works with those account types
- Test with a range of image types and audio lengths early on, since quality varies a lot depending on what’s actually being sent
- Keep your prompts specific about how to handle unclear images or audio, so the model asks a clarifying question instead of guessing
- Watch your messaging window logic closely if you’re running high message volume, since replies sent outside the window will fail silently in some setups
Where this is useful beyond basic replies
This kind of setup works well for product questions sent as photos, voice note inquiries from people who would rather talk than type, or support requests where a screenshot explains the issue faster than text would. The AI model is doing the understanding, and InstantDM is what makes sure the reply actually reaches the person reliably within Meta’s rules.
If anyone wants to try this out, here’s the shared scenario link: Instagram AI Auto-Reply Image & Audio - Make.com Automation Scenario
Happy to go deeper on any specific part of this, like how the image and audio conversion is set up, or how it compares to the text only version I shared before.