Editorial

Editorial Standards

Last updated: 2026-05-24

Every prompt and tutorial published on this site goes through the standards below. We publish this page openly so readers, creators, and search engines can audit our decisions. If you see a prompt that violates these standards, email us at [email protected].

1. Source-tier citation

Every prompt carries one of three source tiers, visible on the detail page:

• 🟢 Official — adapted from DeepMind's official Gemini Omni prompt guide or directly from Google's launch demos (mirror-arm transformation, butterfly→bee, bubble sculpture, etc.). The source URL is linked on the detail page.
• 🟡 Community-verified — adapted from a named third-party hands-on test with a working output linked or screenshotted. We cite the source publication (PixVerse, Atlas Cloud, Chrome Unboxed, Medium, digit.in, Seaart, etc.) by name.
• 🟢 Recommendation — patterns we infer by extrapolating from documented techniques but cannot point to a specific verified output. These are clearly labelled as unverified patterns with the caveat called out at the top of the prompt page.

We do not currently publish prompts marked "AI-generated" or "made up by ChatGPT". If we can't tie a prompt back to a documented source or a working test, we mark it unverified pattern rather than publish it as fact.

2. What we publish

• Camera-language and conversational-editing prompts — the categories where Omni is genuinely differentiated
• Source-cited tutorials from creators on YouTube and editorial publications, attributed by name
• Failure modes documented openly — what Omni breaks on, where artifacts appear, where the model is weak
• Field-guide content drawn from DeepMind's official documentation, summarised in our own language

3. What we filter

The following are filtered from the public library entirely:

• Prompts targeting copyrighted IP by name — "Studio Ghibli style", "Marvel character", named celebrities. Stylistic alternatives published instead (e.g. "hand-painted watercolor animation").
• Prompts targeting real public figures' likenesses — politicians, celebrities, journalists. The Avatar feature in particular requires identity verification of one's own face; we do not document workarounds.
• Explicit, NSFW, or violent content — Omni's own filters block most of this; we do not document attempts to bypass.
• Brand camera names ("shot on DJI Mavic Pro", "RED camera footage") — Omni doesn't parse them anyway, and they create misleading impressions.

4. How we update

When Google ships a model update, a new feature, or community testing produces new findings, we update the affected prompt detail page and the field guide. We add an "Updated: YYYY-MM-DD" line to the affected page rather than silently rewriting history.

Prompts that no longer produce a stable output after a model update are either archived with a note or removed and replaced with a re-tested alternative.

5. Takedown / source-rights process

If you are a creator or rights-holder and want a prompt, video embed, or attribution changed or removed, contact [email protected] with the URL and a brief reason. We aim to respond within 48 hours.

We embed YouTube videos by ID — playback is served directly by YouTube, not by us. Embed removal therefore requires either us removing the embed code or the creator restricting embedding via YouTube Studio.

6. Non-affiliation disclosure

This site is not affiliated with Google LLC. Gemini, Gemini Omni, Google Flow, and SynthID are trademarks of Google LLC. We refer to them by name strictly for descriptive, editorial purposes.

The YouTube channels and editorial publications cited on this site are independent creators. We do not have a commercial relationship with any of them. We embed their public content under YouTube's standard embedding terms and link out to their original posts where applicable.

7. Multi-language disclosure

The site is available in English, 中文, 日本語, 한국어, and Español. Prompt text itself is kept in English because that is the input language Gemini Omni Flash currently supports best. Page chrome (navigation, descriptions, calls to action) is translated; prompt bodies are not, by design.

8. AI-generated content disclosure

Some of the prose on this site — particularly category introductions and outro sections — was drafted with the help of an LLM and then human-edited. Prompt bodies themselves, the field guide, the glossary, and source citations are written and verified by humans.

Outputs generated by Gemini Omni Flash using the prompts on this site carry Google's SynthID watermark by default and cannot be falsely attributed as human-made.

Related

• About this site — who runs it and why
• Glossary — the working vocabulary we use across the site
• Contact — submit a correction, source, or takedown request