I SPENT A WEEK MAKING ALL MY MARKETING ASSETS WITH GPT IMAGE 2

I run a few small SaaS products on the side and the part I hate most is making creatives. Posters for a product launch, ad variants for Facebook, a banner for the newsletter, a quick visual for an X post. None of it is hard work, but it eats hours. So when ChatGPT Images 2.0 rolled out, I cleared a week and tried to replace my whole marketing-asset workflow with it. No Canva, no Figma, no Photoshop except for final crops. Just prompts.

Here is what I made, what broke, and what I am keeping.

The setup

I gave myself five real jobs from my actual backlog:

1. A launch poster for a new feature on one of my products
2. Six Instagram square posts for a content calendar
3. Three Facebook ad variants for the same landing page
4. A Twitter/X header
5. A set of three blog header images in a consistent style

In total, around 25 finished images. Before this week, that would have been maybe four hours in Figma plus stock photo licensing fees, or two days waiting on a freelancer.

Job 1: The launch poster

This is where I expected the most pain and got the least. I described what I wanted in plain English: a vertical poster, dark navy background, a phone mockup tilted slightly to the right, a headline reading "Ship faster. Sleep more." in a chunky sans-serif, a subtle gradient behind the phone, and small product feature callouts in three corners.

First generation came back with most of it right. The phone was there, the angle was right, the headline rendered cleanly (this used to be the killer problem for image models, getting text right), but the callouts were generic English text that looked plausible from a distance and turned into nonsense up close. Common issue. I regenerated with the callouts explicitly written out word for word in the prompt. Second try was usable.

Total time: about twelve minutes including two regenerations. Compare that to opening Figma, finding a mockup, picking fonts, aligning everything. The poster is not going to win design awards, but neither did the one I would have made in Figma in two hours.

Job 2: Instagram content calendar

This is where I learned the most important lesson of the week. Single images are easy. Consistent series are hard.

I wanted six posts that looked like they came from the same brand. Same color palette, same composition style, same "feel." If you just write six separate prompts you get six different brands. What worked: I generated one image I liked, then for every subsequent post I started the prompt with "in the same visual style as the previous image, with the same warm cream background and rounded shapes," and described only the content change. About 70% of outputs were on-brand. The other 30% I regenerated.

For a six-post series, total time including misses was around 45 minutes. Acceptable.

Job 3: Facebook ad variants

This was the surprise. Ad creative testing usually means making the same ad three different ways and shipping all three. Different headline, different image angle, different color treatment. Normally a pain.

Here I just wrote one base prompt and then asked for variations: "same product shot but on a white studio background," "same product shot but in a cozy home setting," "same product shot but flat lay from above." Three variants in about eight minutes. I pushed all three live on Facebook the next morning and the home-setting one is winning by about 1.4x on CTR so far. That alone paid for the month of API usage.

Job 4: Twitter header

Boring job, did it in five minutes, moving on.

Job 5: Blog headers in a consistent style

Similar challenge to the Instagram series, similar solution. The style anchor prompt approach kept the headers visually related. Where I struggled was getting specific concepts rendered. "A blog header about API integrations" produced a lot of generic abstract tech illustrations. The model is much better when you tell it exactly what objects to show, not what topic to evoke. "Two cartoon plug connectors snapping together against a soft yellow background" worked. "An image representing API integration" did not.

What broke

Three things, honestly.

Hands and fine details are still imperfect if you put a person in the frame. I avoided people for this reason and used product shots, objects, and abstract compositions instead. Worked fine.

Brand colors are approximate. If you have a strict brand hex code, the model will get you within visual distance but not pixel-exact. For my products this is fine. For an enterprise brand guide, you would still want a designer touching up the final output.

Aspect ratios for niche platforms (LinkedIn carousel, Pinterest pin) sometimes needed a manual crop. Five seconds in Preview. Not a real problem.

What I am keeping

Almost all of it. The week ended with me cancelling my Canva Pro subscription, which is the most direct vote of confidence I can give. For an indie founder or a small team without a designer on payroll, the workflow now looks like: write a clear prompt, generate, regenerate once if needed, crop in Preview, ship. Total marginal cost per asset is API pennies plus maybe ninety seconds of attention.

The big shift is psychological more than technical. I used to push marketing creative to the end of the week because it felt like a chore. Now I make the visual at the same moment I write the copy, which means I actually ship more campaigns, which means I get more data, which means I can iterate faster. That compounds.

If you are a solo founder, a Shopify store owner, an affiliate marketer, or anyone whose job involves making "good enough" marketing visuals at volume, this is the workflow to copy. Try it on GPT Image 2 before you renew whatever design subscription you are currently paying for. You can decide for yourself whether it covers 80% of what you actually need.

For me, it did.