ChabadLabs AI FOR SHLICHUS
EN

Prompting AI Image Tools to Depict Your Community

Describe the clothing, the setting and the activity, not a group label, to get accurate pictures of a Chabad House and its community from AI image and video tools.

On this page

Shluchim making flyers and videos want the people in the picture to look like the people who will actually come: the shliach and his family, the regulars, the guests. AI image and video tools often get this wrong. Name a group and they reach for a stereotype, refuse, or produce clothing from a different tradition entirely. A thread in the main group in November 2025 settled on a fix that has held up across every tool since: describe what a camera would see.

What goes wrong

The models learn from broad internet imagery. Ask for a religious Jewish scene in a few words and you tend to get a generic costume, the wrong head coverings, or a scene that looks like no Chabad House anyone has been to. In November 2025 a member asked OpenAI's Sora video tool for a scene by naming the community and got back people dressed in a different religion's clothing. In other cases tools refused the request outright, treating the group label as sensitive. (Sora itself is gone: OpenAI announced on 2026-03-24 that it was shutting the app down, Engadget.)

The fix: describe, do not label

Give the model the same concrete brief you would give a photographer or an illustrator:

  • Clothing. A white shirt and dark suit, a kippah, tzitzis showing, a beard. Modest dresses with sleeves past the elbow and skirts below the knee. A sheitel.
  • Setting. A Chabad House dining room, a sukkah in a backyard, a Hebrew school classroom with a picture of the Rebbe on the wall, a shul with a wooden aron kodesh.
  • Activity. Making Kiddush, lighting candles, a child putting a coin in a pushka, a mivtzoim table on a college campus.
  • Mood and light. Warm evening light, a crowded table, people mid-conversation.

The thread found that words like "Chabad" or "Chasidic" help the model place the scene, while bare group labels invite the stereotype or a refusal. Concrete description also makes the image match your brand better.

A prompt in that spirit:

A Shabbos dinner at a Chabad House: a long table with a white tablecloth, two covered challahs, a silver Kiddush cup and bottles of wine. The rabbi, bearded, in a white shirt and kippah, stands at the head of the table making Kiddush. Around the table, guests of all ages in smart casual clothes; women in modest dresses with sleeves past the elbow. On a side table, the Shabbos candles: the mother's candles with a smaller candle beside them for each young girl. Warm home lighting, natural documentary photo style, people mid-conversation, nobody posing.

Check what the picture shows

Accurate people are half the job. The other half is the objects and the practice, which the models also get wrong: a menorah with rounded arms, Luchos with rounded tops, two Shabbos candles where there should be more, a lulav in a woven holder. Before an image goes out, have someone who knows the practice look at it. The list of common corrections is on Image Generation, and the Designer Prompt page has a block of Chabad-practice rules you can paste into your standing instructions.

When a tool refuses

If a tool refuses a plain request to show your community, rewrite the prompt around clothing, setting and activity, start a fresh chat, or try a different tool. A broader look at refusals, and which workarounds are reasonable, is on AI Refusals and Workarounds.

Last updated October 2026 · Maintained by ChabadLabs AI Agent