glunty AI tool cheat sheets / Gemini
Gemini tips and prompt cheat sheet
Gemini is Google's AI assistant, and the quickest wins come from building a reusable Gem, picking the right Nano Banana image tier, and grounding answers in your own sources. This cheat sheet maps Gems, @ app mentions, Canvas, Deep Research, and image generation so you spend your Pro quota on the jobs that need it. The prompts run in your own Gemini account; this page just teaches you how to write them.
Last verified: 2026-07-07. Free tiers and features change, so check the official site.
At a glance
- What it is
- Google's AI assistant with Gems, image generation, and research tools, on web and mobile. Verify current features.
- Gems
- Reusable custom assistants: saved instructions plus optional Knowledge files.
- NotebookLM import
- Supported: pull a NotebookLM notebook in as Gem knowledge without re-uploading exports.
- Nano Banana tiers
- Image quality follows the model selector; Lite has no multi-reference, Flash or Pro and Pro do. Verify current features.
- Personal Intelligence
- US 18 plus on a personal account; uses named Google Photos face groups for character consistency. Verify current features.
- Gemini Live
- Voice mode that is not available inside Gems, per official help.
Gems (custom assistants)
New Gem Create a reusable assistant with saved instructions you can rerun anytime.
Example: Customize, then Gems
Instructions The persistent system prompt: role, tone, and a hard format rule.
Example: max 400 words, H2 sections
Knowledge files Reference files the Gem always reads before answering.
Example: brand guide PDF plus 2 posts
Import NotebookLM Pull a NotebookLM notebook in as Gem knowledge, no re-upload.
Example: tone-matched rewrites
Disable citations Turn off knowledge citation clutter in Gem settings if it annoys you.
Gemini Live Voice mode that is not available inside a Gem, per official help.
Image generation (Nano Banana)
Flash-Lite Nano Banana 2 Lite: fast, but has no multi-reference.
Example: quick single-image drafts
Flash / Pro Nano Banana 2: balanced, and supports multi-reference images.
Example: blend 2 reference images
Redo with Pro Upgrade a generation to Nano Banana Pro for text and infographics; burns Pro quota.
Example: text-heavy poster finals
draw / generate / create Open an image prompt with an action verb, then style, then detail.
Example: create a flat icon of a fox
Multi-reference Feed several reference images to hold a subject; Flash or Pro only.
Apps, Canvas, and grounding
@ app mentions Pull a connected Google app into a prompt. Verify which apps are live.
Example: @YouTube summarize this
Canvas An interactive side panel to draft and edit docs or code in place.
Example: iterate on a doc without re-prompting
Grounding Answers backed by live Google Search results with source links. Verify current features.
Double-check Uses Search to validate claims and flag them; verify before you publish.
Research and consistency
Deep Research Agentic mode that plans, browses many sources, and writes a cited report. Verify current features.
Example: a multi-source brief
Personal Intelligence US 18 plus, personal account; uses Google Photos for character consistency. Verify current features.
Named face groups Name people in Google Photos so the right face lands in generated images.
Veo video Text-to-video generation available on some plans. Verify current pricing and features.
No rows match that search. Clear the box to see all 19.
Tips
Build a Gem, stop repasting
If you paste the same instructions every chat, save them as a Gem instead. Give it a role, a tone, and one hard format rule, then attach Knowledge files so it answers from your own references.
Match the image tier to the job
The Nano Banana variant follows the model selector. Flash-Lite is fast but has no multi-reference; switch to Flash or Pro for Nano Banana 2 when you need to combine reference images.
Redo with Pro only for text
Nano Banana Pro renders clean text and infographics but burns your Pro quota. Draft on Nano Banana 2 first and only Redo with Pro for the text-heavy final.
Ground it, then double-check
For anything factual, turn on grounding so answers come with source links, then run double-check before you rely on a claim. Treat an ungrounded answer as a draft.
Import NotebookLM for tone
You do not have to re-upload research. Import a NotebookLM notebook as Gem knowledge so the Gem rewrites in a voice grounded in your own sources.
Name your faces for consistency
With Personal Intelligence on a personal US account, name the face groups in Google Photos so the same person lands in generated images. Verify current features, since availability changes.
Operator recipe
You will make: A reusable brand Gem that writes an on-brand post and drafts a matching hero image in one session.
- Open Gems, create a New Gem, and write instructions that set role, tone, and one hard format rule like max 400 words.
- Add Knowledge files: a brand guide PDF and two exemplar posts so the Gem matches your voice.
- If you research in NotebookLM, import that notebook as Gem knowledge instead of re-uploading exports.
- Draft a post in the Gem, then switch the model selector to Flash or Pro and prompt with create for a matching image on Nano Banana 2.
- Only Redo with Pro when the image needs clean text or an infographic, since that burns your Pro quota.
- Turn on grounding and run double-check on any factual claim before you publish.
- Tighten the Gem instructions with the templates at /ai-prompts/prompting-basics, or refine any weak prompt at /improve-prompt.
Pass check: The Gem produces an on-brand post and a matching image in one session, with no repasted instructions and Pro quota to spare.
Failure modes
| Symptom | Cause | Fix |
|---|---|---|
| Text in the image comes out garbled | You generated on the Lite variant | Switch to Flash or Pro for Nano Banana 2, then Redo with Pro for the final |
| Multiple reference images are ignored | Flash-Lite has no multi-reference | Move to Flash or Pro so Nano Banana 2 can read all the references |
| Generated person is the wrong face | The Google Photos face group is unnamed | Name the face groups, and confirm Personal Intelligence is enabled |
| A Gem ignores its attached files | The Drive file is a stale version | Re-add the file so the Gem loads the latest version |
| Voice mode will not start in a Gem | Gems cannot use Gemini Live | Use Gemini Live in a normal chat, outside the Gem |
Workflow ladder
Chat normally, attach a file, and turn on grounding to get answers with source links.
Build a Gem with instructions and Knowledge files, and generate images on Flash or Pro.
Use @ app mentions and Canvas to pull in your Google apps and edit drafts in place.
Import a NotebookLM notebook, run Deep Research, and reserve Redo with Pro for text-heavy finals.
Frequently asked questions
- Why is the text in my Gemini image garbled?
- You are probably on the Lite variant, which is fast but weak on text. Switch the model selector to Flash or Pro for Nano Banana 2, then use Redo with Pro for a text-heavy final like a poster or infographic.
- What is a Gem and when should I make one?
- A Gem is a reusable custom assistant that saves your instructions and optional Knowledge files. Make one whenever you find yourself pasting the same role, tone, or reference material into a new chat every time.
- Can I use Gemini Live inside a Gem?
- No. Per Google's help docs, Gemini Live voice mode is not available inside Gems, so do not plan voice workflows there. Use Live in a normal chat instead.
- How do I keep the same person across generated images?
- Turn on Personal Intelligence on a personal account, then name the face groups in Google Photos so Gemini knows who to draw. Availability is US 18 plus and changes, so verify current features.
This is reference content. The prompts and settings run in your own AI tool, not on glunty, so reading this page makes zero requests to those services. Details change often, so verify current pricing and limits on the official site.
Share your result
Made something you like with Gemini? Frame the screenshot or keyframe in Beautify a Screenshot: padding, rounded corners, and a soft shadow, then post it. Shrink it first with Compress Image if you need a smaller file. It all runs in your browser.
Official reference: Gemini docs. Details change, so verify version-specific features there.
Related on glunty: NotebookLM tips, ChatGPT hacks, AI prompt library.
More AI tool guides
Midjourney parameter cheat sheet
A searchable Midjourney cheat sheet: core parameters like --ar, --sref, --cref, --stylize, --chaos, and --no, each with what it does and an example. Free.
Suno AI tips and prompt cheat sheet
A searchable Suno cheat sheet: section tags, vocal delivery cues, the 5-part style formula, and hacks to stop wasting your daily credits. Free, copy-paste.
Leonardo.Ai token costs and cheat sheet
Learn how Leonardo.Ai tokens are spent per generation and stretch your free daily tokens with this cheat sheet.
Kling AI prompt syntax and cheat sheet
Master Kling AI prompt syntax, storyboard shots, and elements to make consistent AI video within your free daily credits.
ElevenLabs tips and voice settings cheat sheet
Tune ElevenLabs voice settings for narration or character work with a cheat sheet of stability, similarity, and Voice Design.
Ideogram tips and text-in-image cheat sheet
Get reliable in-image text from Ideogram with a cheat sheet on quoting words, Magic Prompt, fonts, and JSON prompting.
Cursor rules tips and cheat sheet
Steer Cursor with rule files and @ mentions: a cheat sheet for .mdc rules, AGENTS.md, globs, and rule precedence.
NotebookLM tips and Audio Overview guide
Get better NotebookLM answers and podcasts: a cheat sheet for source filters, the Customize Audio Overview brief, and Gemini.
ChatGPT hacks for power users
Master ChatGPT with Custom Instructions, Projects, custom GPTs, and Deep Research for faster results.
Claude hacks for power users
Master Claude with Projects, Styles, and Artifacts for nuanced writing and long-document work.
Runway prompts and tips
Camera moves, Act-One, and Motion Brush prompts to control Runway Gen-4 AI video generations.
Perplexity tips and focus modes
Focus modes, Collections, and Pro Search tips to get sourced answers from Perplexity AI search.
Imagine.art tips and credit hacks
Model picker, credit hacks, and suite workflow tips for Imagine.art image, video, and voice.
Udio AI music tips and cheat sheet
Master Udio AI music: a cheat sheet for Styles, Style Blend, Sessions waveform editing, Voices, and the instrumental break fix. Free.
Also useful: the AI prompt library and the developer cheat sheets.