OpenAI launched a crazy good new image engine this week. The main advantage is that it can add text to an image with no problems.
It is an order of magnitude better than Dall-E. I don’t know if it has an API yet, but if it does, is there a timeline when we can use it with Pickaxe? I’m willing to bring my own OpenAI api key to use it.
Any idea on a timeline for it’s support?
@thomasumstattd openAI has not released the APIs yet
Yes, indeed. It would be great to provide our users with this option.
Sam Altman complained that this new engine is melting their servers so it may be awhile before they open up the API. It will be epic when they do though. I already have a bunch of cool Pickaxe ideas.
@thomasumstattd yes, I’ve been playing a bit with it and it will really be a game changer.
To be able to have any text, formula and graph not only rendered correctly but also linked to the underlying subject is incredible.
This is something I created the other day with a prompt as simple as - create an infographic on the law of gravity
That image generation model is not available through API yet. When it is available through API, we will offer it.
Open AI finally added the image generation model to the API. They are calling it GPT Image 1.
Once Pickaxe adds support for this, I might be able to double my pickaxe users.
There is now an gpt image action but when using/testing it I noticed the image will not display. The response states that the image is ready but nothing displays.
I feel we are close to having this functionality added but the images just do not display. I know the action returns a json with base64 and not url now. Guessing that is the issue.
I did notice this issue happens often with Dalle too. Sometimes the image will display but often it doesn’t.
I get this repsonse when using the action: ‘Technical difficulties prevented the direct generation of this image, but this detailed description is intended to evoke the same creative imagery.’
You need to make sure your org is validated in OpenAI. The API is only accessible to validated orgs
To access the OpenAI GPT-Image-1 API, organizations may need to complete a verification process.
Cheers Gil… I’m already verified but it hasn’t made a difference.
Sometime generates a slightly different message ‘Image generation encountered an issue’ or ‘Unfortunately, the image generation failed, but here is a detailed visualisation:’
Yes, while I don’t get that direct issue I have noticed if the image is taking too long or if there are other activities involved that the UI just may be responding with what it has so far. Like old school buffering back in the day.
The UI may not have long term ability to stay open long enough to pull the request?
What prompt are you using? I am willing to enter it and see if it returns anything for me.
Yep - I think some timeout issues may have been at play here when it fails. It is working at the moment albeit not consistently and when it does… it always display the same image (at least) twice but references is if it is a totally different image!
Yes I saw that too. It doesn’t do it if the user explicitly states to do one image in the prompt. But the code is set correctly to n: 1.
I did clone it to my pickax to set the image size to auto. Didn’t want the user to always get a square format.
I didn’t see any code for timing for a timeout. Think the timeout is happening when pickaxe decodes the image for local storage. Which may be happening after the api call. Meaning the timeout is at the aws server to present the url?
Have you encountered this error (or similar): ‘/tmp/tmpw8wellvo/action.py was modified during build process.’?
Getting that now as well, wonder if they are fixing the processing inside pickaxe?
I’m getting this error: “I’m experiencing issues generating the image at the moment. Please try again later. If there’s anything else I can assist you with, feel free to let me know!” Verification did not help.
Waiting 2 days did help. It’s now working!
Were you able to find a solution to the image shape? I’m only getting square images despite the prompt calling for a different ratio.
You have to change the sizing to whatever size that you need or just remove that part of the code.
Then the code will automatically generate a size based on the prompt
