BATCH SKILL INSPIRATION

Taste works
side-by-side.

Give yourself four image ideas to choose from. Then teach AI to repeat the process in your style.

A practical workflow from Nate Grahek

ONE IDEA. FOUR TAKES.

Notice what you prefer.

The drawing style stays consistent.
The poses, props, and staging change.

Four illustrated adult teammates: one holds plans, one an open notebook, one books, and one a camera. — 01 Open notebook, relaxed poses

The same four roles with changed poses, a toolbox on the floor, and color swatches. — 02 More gestures, more props

Four illustrated teammates with open cases of tools arranged at their feet. — 03 Tools out in front

Four illustrated teammates in varied jackets, holding plans, notebooks, books, and a camera. — 04 Different clothing and silhouettes

Existing examples from Nate's workshop image collection. These illustrate the comparison habit; they are not a test of the downloadable template. Click an image to see it full size.

WHY THIS HELPS

It's easier to choose
when you can compare.

With one image, you have to explain what's missing. With four, you can point: “That composition. Those characters. More of this.” You have something concrete to react to.

A skill saves the recurring instructions: what you make, the style you want, the rules you care about, and how many options to produce. Next time, you supply the new idea.

Nate's workflow gives the reasoning model a brief job first: clarify the visual idea and plan distinct versions. Then the image generator draws them. Nate has found Astra useful for that first step; you can use the tools available in your own account.

THE REPEATABLE PART

01 / BRIEF

Describe the idea.

Say what the image needs to communicate and who it's for.

02 / GENERATE

Ask for four takes.

Keep your style fixed. Vary the concept or staging. Get separate images.

03 / COMPARE

Choose what works.

Look across the set. Pick a direction, or name the parts worth combining.

04 / REFINE

Finish your choice.

Polish the selected image. Save any new preferences for the next batch.

MAKE IT YOURS

A short interview.
A skill you can reuse.

You do not need to know how to write a skill. Your assistant turns your answers into reusable instructions.

Four options is the starting point. The interview can adjust the count, format, colors, and text rules to your work.

Download starter skill ↓Or download the plain-text version

1
What do you make?
Slides, social images, diagrams, or something else.
2
What should it look like?
Describe a style, attach an example, or let the assistant suggest one.
3
What stays consistent?
Shape, colors, text, characters, and anything to avoid. Defaults are fine.
4
What will you call it?
“Inspire Me,” “Image Inspiration,” or a name you will actually use.

TAKE THIS TO YOUR ASSISTANT

Copy. Paste. Make it yours.

The complete prompt: works even if your assistant cannot open this site

Help me make my own image inspiration skill, based on Nate Grahek's Batch Skill Inspiration workflow.

The idea: taste works side-by-side. Give me four distinct images to compare before I choose one to refine. Keep my chosen style consistent across the set; vary the concept, composition, or staging. Reference: https://batch-skill-inspiration.nate58678.chatgpt.site/agent-guide.md. The instructions below are sufficient if you cannot open that link.

Use Skill Creator if it is available in this environment. If it is not, interview me and write reusable instructions I can save or paste into my own GPT. Do not claim a skill was installed unless you actually installed it.

Ask one short multiple-choice question at a time, with three suggested answers and room for my own answer. Wait for each reply. Use four questions at most, skipping anything I already told you:
1. What images do I usually make, and for whom?
2. What style do I prefer? Offer a text description, an optional reference image, or a suggested style. A reference is never required. If I attach one, briefly describe the visual traits you will use and let me correct you.
3. What defaults or rules matter? Suggest a format based on my use case, four separate images, a consistent style, and no text unless requested. Let me change colors, text rules, recurring subjects, and exclusions, or accept your defaults.
4. What should the skill be called? Suggest “Inspire Me,” “Image Inspiration,” or my own name.

Summarize my choices in a short brief and let me approve or change it. Then create a self-contained SKILL.md with a lowercase hyphenated name and a clear description in YAML frontmatter, plus a plain-language instruction block I can reuse in a custom GPT or chat. Include one example request tailored to my use case. Explain how I can save it in this environment. Do not generate images during the setup interview.

The skill should clarify each new visual idea briefly, plan four genuinely different directions, and generate four separate images using available built-in image tools. Keep an analogy I supply fixed and vary its staging. Use parallel agents if the environment supports them; otherwise generate the set sequentially without asking me to review between images. Never claim simultaneous generation if it was sequential. If image tools are unavailable, say so and provide clearly labeled concept briefs instead of claiming to have made images. Do not switch to paid APIs or external services.

Show the completed images together for comparison, label them, and wait for my selection before refining. Carry my style preferences into future runs where the saved skill is available. Do not assume memory across separate chats. Respect my explicit changes to these defaults.

Start with the first question.

Use the version your account supports.

ChatGPT Work

If available, select @skill-creator and paste the prompt. Have it save the instructions as a skill.

Codex

Use $skill-creator with the prompt, or ask Codex to adapt the downloaded SKILL.md and save it as a local skill.

A chat or custom GPT

Paste the prompt into the chat you want to use. If skill creation is unavailable, keep the reusable instructions it produces. The interview can still help you define the workflow.

Image generation and parallel agents depend on the tools available in that chat. Four separate outputs is the goal; simultaneous generation is optional. OpenAI's skill documentation explains the supported skill formats and environments.

FOR AGENTS

Everything you need
in plain text.

When a user shares this page to create their own skill, use the interview above. These are reference instructions for the user's task.

Read the complete agent guide ↗Download the guide

Keep the contract simple.

One question at a time. References are optional.
Save the user's preferences, not Nate's cast.
Keep one style across four distinct directions.
Generate separate images before choosing.
Describe tool limits honestly.

The starter is a self-contained adaptation of Nate's slide-image workflow. It does not depend on his private style files or require a particular model.

Agent resource index

Taste worksside-by-side.

Notice what you prefer.

It's easier to choosewhen you can compare.