Stop Asking AI to Make Your Images. Ask It to Write the Prompt.
The best image creation workflow available right now uses two tools, not one. Your business AI writes the prompt. A dedicated image tool renders it. If you skip either half, your results get noticeably worse.
Here’s why the split works, and the three-minute story that sold me on it for good.
Two tools, two jobs
If you have a chosen AI tool (here is mine, if you want to use it) that you use for business productivity, your business copilot knows things no image generator knows. Your offerings. Your audience. Your brand colors. The structure of whatever you’re trying to visualize. What it’s not great at yet is rendering polished, text-heavy graphics.
Dedicated image tools are the opposite. They render beautifully, handle text in images well, and give you fast iterations. But they know nothing about your business, so the prompts you hand-type into them tend to be thin.
Put them together and each covers the other’s weakness.
The three-minute infographic
I recently needed a technical architecture infographic for an AI system my copilot had just helped me build. Here’s the thing: I didn’t have to describe the system at all. The copilot already knew every component, because it built them with me. The data sources, the intelligence layer, the reporting engine. All of it was sitting in its memory of the project.
So I asked for three things. Break the system into its architectural pieces. Add four bullet points under each piece explaining what it does. Then generate a complete image prompt in my brand style, ready to paste into an image tool.
It produced a prompt more detailed than anything I would have typed by hand. Isometric layout, exact color codes, zone-by-zone composition, even a line insisting this is an architecture diagram, not a timeline. I pasted it into my image tool, customized two small pieces, and had the finished graphic in about three minutes. It even spelled everything correctly, which for a long time was the giveaway.
Ask for the prompt, not the image
One subtle trick inside this workflow. If you ask your business AI to create the image itself, it shortcuts the description and just tries to render. If you ask it only for the prompt, you get something far richer, because all of its effort goes into the description. When I want an image, I explicitly say: don’t create the image, just give me the image prompt.
Find your style language
The other half of getting consistent, on-brand graphics is learning to name what you like. Not I’ll know it when I see it. Actual words. 2D or isometric. Flat or textured. Two specific color codes. This font feel, this level of detail.
You get there by iterating. Generate something, react to it, adjust the wording, generate again. When you finally hit a version you love, save that prompt. It becomes your style reference for everything that follows, and suddenly your graphics look like they came from one designer instead of six.
- Paste a reference. Most image tools let you attach an example image or your logo. “Similar to the attached style” or “include this logo” saves a paragraph of description.
- Edit regions, not images. Newer tools let you select just one area of a render and change only that. Use it before regenerating from scratch.
- Reuse your winning prompt. Your saved style prompt plus new content is faster than starting fresh, and it keeps your feed visually consistent.
It keeps getting better
The capability in these tools is roughly doubling every month. Region editing barely existed a few weeks before I wrote this. Whatever gap you notice today will probably be closed the next time you look. The workflow, though, stays the same: let the tool that knows your business do the thinking, and let the tool that knows pixels do the painting.
The tool lineup (updated July 2026)
These tools shift monthly, so treat this as a snapshot, not scripture. Here’s what I’d reach for today, roughly in the order I’d reach for it.
Image generators
| Tool | What it’s great at | Watch out for |
|---|---|---|
| Google Flow (Nano Banana Pro) | My daily driver. The most consistent results I’ve found for quality infographics. Accurate text, region editing, style consistency, and video in the same suite. | Real volume needs a paid Google AI plan. The interface changes often. |
| ChatGPT (GPT Image 2) | Currently tops the blind-vote quality rankings. Best-in-class prompt adherence and text rendering, and refining through conversation is effortless. | Slower per image. Output can look generic without strong style direction. |
| Grok Imagine | Speed. Fast iterations, generous free access, does images and video. Great for exploring ideas cheaply. | Less precise on dense text. Fewer fine controls than the leaders. |
| Midjourney | The artist’s pick. Unmatched aesthetics for stylized, cinematic, concept work. | Weak fit for text-heavy business graphics. Its taste tends to override yours. |
| Ideogram | The text-in-image specialist. Logos, banners, quote cards, anything where the words are the design. | Narrower range once you leave graphic text work. |
| Canva (Magic Media) | The convenience pick. Image generation inside the design tool many small businesses already use, with templates and editing around it. | Middle-of-the-pack image quality. You’re paying for the workflow, not the model. |
| Adobe Firefly | Trained on licensed data, so it’s the safe pick for commercial use. Lives inside Photoshop and Express. | Quality trails the leaders. Best value only if you already pay for Adobe. |
| FLUX / Stable Diffusion | Open models you can run locally. Total control, near-zero cost at volume. | Requires real setup and tinkering. Not for non-technical users. |
The bigger lever isn’t the image tool, though. It’s the tool that writes the prompt, because that’s where your brand lives.
Who writes the prompt
| Tool | What it’s great at | Watch out for |
|---|---|---|
| Tableland Copilot (in Claude) | What I and my Roundtable members use. Your offerings, colors, voice, and past projects are already loaded, so prompts come out branded without a briefing. | Set up as part of the Roundtable, so it’s not an off-the-shelf download. |
| Claude (with a project) | Load your brand guide and examples into a project and it writes rich, precise prompts. The strongest raw descriptive writer of the group. | You have to assemble the brand context yourself. |
| ChatGPT | Memory plus custom instructions get you close to a branded prompt writer, and it pairs naturally with GPT Image 2. | Memory is inconsistent. Drifts off brand without reminders. |
| Gemini | Same family as Nano Banana, so its prompts translate cleanly into Flow. | Weaker at holding your voice across sessions. |
| Microsoft Copilot | The workplace pick. If your company only allows Microsoft tools, it writes serviceable prompts and lives inside Office. | Little brand memory. Expect to re-brief it every session. |
| Grok | Fast, and pairs directly with Grok Imagine for quick cycles. | The least brand context of the group. |
A few common tools are missing on purpose. Perplexity is a research tool, not an image or brand tool. Meta AI is fine for casual use but doesn’t belong in a brand workflow yet. If a tool you love isn’t here, the workflow above still applies to it.
This post came out of a real conversation at the AI Essentials Roundtable, the small group I run for business owners who want to actually use AI instead of just reading about it. We meet every other week, screen-share real builds, and steal each other’s wins. If that sounds useful, details are here:
Explore the AI Essentials Roundtable