🎨 GPT Images 2.0 is now live in Pickaxe

[abhi]

It’s a BIG update.
ChatGPT just launched its latest image model and we added it to Pickaxe right away.

What’s new

Actually understands text
You can now generate images with readable, clean text. Posters, slides, ads… this used to break. Now it works.

Better prompt accuracy
What you ask is much closer to what you get. Less trial and error.

Multilingual support
Text inside images works across languages. Huge if you’re building global tools.

More realistic + structured outputs
Cleaner layouts, sharper visuals, better composition.

What this means for you

You’re not just generating images anymore.

You can now build:

• Marketing creatives directly inside your Pickaxe

• Social media posts with text baked in

• Slide previews and visual reports

• Branded assets without leaving your tool

How to use it (takes 10 seconds)

1. Go to Actions tab

2. Search “Generate GPT Images”

3. Toggle it ON

4. Add a simple instruction in your prompt like:
   “Generate an image when the user asks for visuals”

That’s it.

If you want to go deeper on Actions, this guide is worth a quick skim:

Pro tip

Be specific with prompts.

Instead of
“Create a poster”

Try
“Create a modern startup poster with bold headline text ‘AI for Everyone’, minimal design, soft gradient background”

You’ll see the difference instantly.

If you try something cool with this, share it in the thread. Curious to see what people build

[thomasumstattd]

This is totally broken for me. I can’t get any images to generate in any pickaxe.

[carsondev]

hey thomas ! @thomasumstattd thanks for flagging, mind trying again ? should be good to go

[thomasumstattd]

It works! This is such a relief!

My inbox has been blowing up with users not able to use the image generators.

[motheraigaia]

Nice! Great job guys, how does this compare cost wise to nano banana? I’m wondering ballpark what it costs to create an image using GPT images 2?

[danny_support]

Hi @motheraigaia ,
Thank you for your question!

Image generation costs can vary depending on image size, quality, and the request itself.

OpenAI prices GPT Images 2 by tokens, so the cost varies by output size and quality. Their examples show a 1024x1024 image at around $0.006 for low quality, $0.053 for medium, and $0.211 for high.

Nano Banana Pro / Gemini 3 Pro Image Preview is priced more like a per-image cost, with Google listing it around $0.134 for 1K/2K images and $0.24 for 4K images.

[enochleffingwell]

Thanks so much for this breakdown, @danny_support — super helpful and very thorough.

I really appreciate you walking through the different pricing models and giving concrete ballpark numbers for each option. That’s exactly what I was wondering about too, so I’m also grateful to @motheraigaia for asking the question in the first place.

[Kent]

This creates awesome images. take a look, this is by just asking for an image for a bakery, in A4 format, thats all

[abhi]

@Kent : wow! pretty cool. Thanks for sharing.

[enochleffingwell]

I’m not sure if this is the best place to put my questions about Image 2.0 or better to put a new post about this.

But I have some questions about how the Image 2.0 thinking works.

Maybe @danny_support you could take a look? I’d love to hear your feedback or someone else from the team.

Video #1

Video #2

[enochleffingwell]

If I wanted to also include a branded style guide reference image like this with my prompt. How would I go about doing that?

Currently I added it to the Knowledge files and am referencing it in my Role Prompt.

BTW: Image 2.0 generated this style guide. It was very helpful.

[briang]

Hi Enoch,

Thanks for documenting your findings, your questions dig deep into not only how Pickaxe interacts with the GPT Image 2.0 action but also how OpenAI have designed their model to work.

I’ll try to answer them to the best of my understanding.

1. If GPT Image 2 generates the image, does the selected agent model still matter?
   Yes. GPT Image 2 is the model actually rendering the image. But in a workflow tool like Pickaxe, the LLM you select like Gemini, GPT, or Claude is usually the model reading the transcript, deciding what to do, preparing the prompt, and calling the image action.

2. How do I get ChatGPT-quality infographics in Pickaxe
   I noticed you selected the model used in Pickaxe as GPT-5. The beautiful infographics you were able to create in ChatGPT are actually using the most recent model released, GPT-5.5. This makes a pretty big difference and through my testing should net you the largest improvement in infographic quality.

Example made by an PickAxe agent on GPT 5.5:

3. Where does the “thinking” happen? Is it the the agent or the image model?
   The reasoning slider controls how long and thorough the chat LLM’s reasoning is before answering and calling the image tool. Only the final prompt string the LLM emits ever reaches the image model (doesn’t pass the transcript). Internally OpenAI could be providing that to the image model but we’re not entirely sure.

Lastly, if you are worried about image generation timing out - you can try lowering the reasoning level and cutting the prompt down that the Wingman created since over-long prompts can be detrimental.

Let me know if you have any questions or interesting findings but it is very cool to learn about how these things work behind the scenes.

[enochleffingwell]

Wow, this was really helpful. Thank you so much for getting back to me.

It makes perfect sense that GPT‑5.5 is what I’ve been using to get these results I really like.

I did end up creating a duplicate, and that was really helpful to simplify the prompt.

I see your photo was great looking. That’s along the lines of what I was hoping for. Can I ask you what level of reasoning you used for your 5.5 infographic you made to get those results?

Your Response was so helpful, and I appreciate your graphic example too.

Thank you for getting back to me.

[briang]

Glad to hear what I said has helped.
You should definitely use reasoning but it doesn’t need to be too high. I think I used the “Fast” setting.