I’m building a fully automated internal employee magazine published monthly. The pipeline is split into 3 Make.com scenarios:
Scenario 1 – News Collecting: Fetch article URLs via RSS feeds → append them to Google Sheets automatically.
Scenario 2 – Article Curation: Merge collected articles using a Text Aggregator → send to Gemini with a selection prompt to pick employee-relevant topics → receive the curated list with reasoning via Gmail for manual review. (Gmail checkpoint is intentional — it ensures quality and editorial control before content generation begins.)
Scenario 3 – Content Generation: Read selected topics from Scenario 2 → Gemini fetches original article links and generates 3 subheadings + 300–400 character body per topic → results sent via Gmail for review AND routed via Router to a Parse JSON module → Google Sheets (Sheet 2) to store each topic, subheading, and body in separate columns, prepping data for image generation and Canva template population.
End goal: Auto-populate a branded Canva template with generated body text and AI-generated images that match each article’s content.
What is the problem & what have you tried?
The core problem is with the AI image generation step (planned as Scenario 4). When I generate images based on the article topics/body text, the results are often off-target — images don’t visually match the intended content or feel inconsistent with the magazine’s brand tone.
I haven’t finalized the image generation module yet precisely because of this concern. I’m unsure whether to:
Use Gemini’s image generation directly
Use a third-party image generation API (e.g., Ideogram, DALL·E, Stable Diffusion via API)
Add a human-in-the-loop checkpoint for image approval before Canva population (similar to the Gmail checkpoints in Scenarios 2 & 3)
Additionally, I’d like feedback on whether my overall 3-scenario architecture has any gaps, inefficiencies, or improvements worth considering before I finalize the pipeline.
Error messages or input/output bundles
No hard errors at this stage — the issue is architectural/quality-related rather than a runtime error. The image generation module has not been implemented yet due to output quality concerns. Open to suggestions on prompt engineering strategies, module choices, or intermediate review steps for image generation within Make.com.
The overall setup looks fine. Splitting it into separate scenarios makes sense, and the Gmail review points are probably a good idea if quality matters.
For the image step, I would not make it fully automatic yet. The usual problem is that article text is too broad, so the image tool fills in the gaps in ways that do not match the tone you want.
A better approach is to add one step before image generation where Gemini creates a short image brief for each article. That brief should describe the subject, setting, mood, style, and anything to avoid. Then use that brief as the prompt for the image tool.
I would also add a review step before Canva. Even a simple approval check will save you from pushing off-brand images into the final magazine.
So the safer version is:
topic approved
Gemini writes image brief
image tool generates one or two options
review
approved image goes to Canva
If you are choosing between image tools, test the same 10 prompts across each one and compare consistency rather than picking one upfront. The best choice is usually the one that matches your brand style most reliably, not the one that looks best once or twice.
Dr. Tanvi Sachar
Monday Certified Partner, Tuesday Wizard
Thank you for your helpful feedback. I’m still new to using make.com, so I’m still learning a lot and have a lot of things to learn. I’ll give the method Dr. Tanvi Sachar suggested a try! Thank you. Have a nice day.
What is your goal?
What is the problem & what have you tried?
Error messages or input/output bundles
Screenshots (scenario flow, module settings, errors)