The 'Looks Too AI' Debate

AI can produce a decent-looking flyer in a minute. The trouble is that by mid-2026 everyone was doing it, and people in the community learned to spot the result at a glance. In July 2026 a member put it bluntly: "Every flyer you look at today is basically one of four AI designs and people are sick of it." This page collects both sides of the argument, the signs that give a flyer away, and the fixes members have tested.

The two positions

Use it, and stop worrying. Most members fall here. Their points, mostly from a February 2026 thread:

• An AI flyer already beats what most shluchim make by hand in Canva.
• Many people, especially older community members, neither notice nor care.
• The flyer matters less than the food and the program.
• In the admin's words: "I'd rather use the time on more important parts of my shlichus and settle for a design worth 10 minutes of AI prompting."

It costs you something. A vocal minority, and their case got stronger as 2026 went on:

• "Too much going on... dizzy" (May 2026).
• "People want to hear you. They see AI, they see it's not you" (May 2026).
• A rabbi in July 2026: flyers, videos and posts made with AI, "even with the amazing prompt shared here, are usually very recognisable."
• In July 2026 one member said a plain graphic that looked hand-made in a basic paint program got a better response than his polished AI flyers.

The rough consensus: use AI for the parts it does well, keep the flyer short, and finish the last part by hand. A February 2026 version of the community prompt put the split at about 70 percent AI and 30 percent human. A member in June 2026 added the simplest fix of all: "your AI flyer doesn't have to have every single bit of information about whatever event you're doing."

The 7 AI Tells

The taxonomy at the heart of the community's Designer Prompt, first posted in February 2026:

1. No intent. Everything is "nice," nothing is purposeful. Fix: one visual idea you can say in one sentence.
2. Default symmetry. AI centers everything. Fix: asymmetry by default; symmetry only for something formal or solemn.
3. Everything shouts equally. Fix: one dominant read, one secondary read, the rest quiet.
4. Fake type space. Generic poster space with no real text in mind. Fix: design around your actual words, with exact placement and size.
5. Global polish. Everything smoothed the same way. Fix: directional light, real material textures, one deliberate imperfection.
6. Stylistic mush. Several styles blended. Fix: name one design tradition and commit to it.
7. Dead space versus tense space. Every pixel filled, or empty space that does nothing. Fix: empty space with a reason to be there.

Newer tells members spotted in 2026

The models changed during the year, and so did the giveaways:

• Blue and gold. "Stop using the blue and gold in flyers!!!" (May 2026, repeated a few days later, with several members agreeing). Ask for a different palette by name.
• The default ChatGPT image look. In May 2026 the admin called the new GPT image standard style "nauseating and instantly recognisable."
• Hand-drawn leaves and doodles scattered around the edges (April 2026). Tell the model to remove them.
• Textured backgrounds. A February 2026 tip: ask for "solid color background instead of the ubiquitous textured," and "drop half the elements."
• A small dash under the headline, centered text, and a sunset color theme (September 2026). The quick fix offered: left-align the text, remove the dash, change the colors.
• AI-built websites have their own version: the same rounded gradients, the same fonts, the same "learn more" buttons. Members argued in August 2026 about whether visitors notice. Some said savvy visitors and kids can tell; others said most people only want clear information on a pleasant page. No consensus.

Fixes that members report work

Ordered roughly by how often they came up:

• Redesign an existing flyer instead of starting from nothing. Upload last year's flyer (or one you like) and ask for a redesign. In April 2026 members called this "far better than asking it to create from scratch." The prompt is on Flyer Design.
• Ask what looks AI, then fix it. Upload the result and ask "Give me a list of all the things that are way too AI about this," then "make the new one" (July 2026).
• Make your own brand rules. Give the model your logo and your website, tell it what you like and dislike, and have it write design rules for your organization. Then tell it to follow those rules every time (July 2026).
• Bring a mood board. Give it a few real designs you like, "just like every graphic or branding team asks you for a moodboard" (September 2026).
• Name a style, a medium and a palette together. For example "WPA poster style, flat layered mountains, three-color print, warm sunset palette," or "two-color screen print, cream paper, no gradients." A Claude-generated list of 20 named traditions shared in September 2026 also advised dropping words like "stunning," "hyper-detailed," "4K" and "cinematic," which pull the result back toward the glossy AI look.
• Generate the art without lettering, then add the text yourself in Canva. This also lets you reuse one piece of art across a series.
• Go plain on purpose. Half joking, a member in August 2026 suggested: "no graphics, no colors, only B&W text. No additional text other than what I gave you."

Hebrew and right-to-left text

Through 2025 every image model garbled Hebrew letters and reversed word order, so the rule was to let AI make the picture and type the Hebrew yourself. The newer models do better. OpenAI says its April 2026 image model improved non-Latin scripts, and members found short Hebrew (a title, a name) often comes out right. But in May 2026 members were still fixing garbled Hebrew on AI flyers, so proofread every letter. Two workarounds from that thread:

1. Remove, then retype. Finish the flyer, tell the model to remove the Hebrew words and change nothing else, then add the Hebrew in Canva.
2. Feed it exact text as an image. Type the Hebrew yourself, export it as an image or PDF, and tell the model to keep it exactly as given.

For typeset documents (a shul magazine with Hebrew phrases), a member found in May 2026 that Claude had to be reminded to write right-to-left and keep the sentence order. See How do I keep AI image tools from garbling the Hebrew on my flyers? and Hebrew & Yiddish.

Chabad objects the models still get wrong

Every image on a shliach's flyer should show what a Chabad House actually does. The models keep slipping on these:

• The menorah. In May 2026 members reported ChatGPT getting the Chabad menorah wrong even with a reference image, and AI drawing six branches instead of eight, or rounded arms. Gemini did better for one member with the phrase "straight armed Rambam Chabad menorah." Another suggested simply calling it a Rambam menorah. In April 2026 a member noted Google "does not yet know to make the Menorah with straight branches or the Luchos square."
• The Luchos should have square tops, not rounded ones.
• Shabbos candles. In February 2026 a shliach pointed out that the Rebbe was particular that Neshek publicity show a third, smaller candle between the two larger ones. Chabad practice is a smaller candle for each young girl, who lights her own. AI images usually draw two.
• Tefillin. In August 2026 a member found ChatGPT could not draw the batim and straps correctly. His tip: give it a real photo and ask for a prompt that reproduces it.
• Matzos. In February 2026 Nano Banana Pro could not be talked out of putting four matzos on a seder plate.
• Kippah and tzitzis drift from figure to figure in group scenes. Add it to the edit prompt ("boys with kippah and tzitzis, modest dress").

The Chabad-practice add-on on the Designer Prompt page puts these rules in one block you can paste. The admin mentioned in May 2026 that a custom Gem for Chabad imagery had been made, but its prompt was never found or shared.

Open problems

• Consistent series. Getting five matching flyers is still hard. Members upload the same style images every time, or ask an AI to describe a reference image as a structured JSON prompt and reuse it (May 2026). See Image Generation.
• The model cannot see its own output well, so each round of edits drifts back toward the defaults. Start a new chat with the best version as the reference when that happens.
• Real photos. Image models regenerate faces rather than pasting them, so a real photo of a person, or of the Rebbe, can come back subtly changed (2025 and again in September 2026). Place real photos yourself in Canva.

Related

• Designer Prompt, the full system prompts.
• Flyer Design, the workflow and the ready-made prompts.
• How do I make an event flyer with AI that doesn't look AI-generated?
• Canva, Nano Banana, ChatGPT.