Hi all,
As announced in December last year, this is the first release of the year for 4AI, and it's focused on making images!
Image generation - and modification - in 4AI
I have been very cautious so far about including image generation into 4AI, as my experiments with available models from OpenAI and others had mixed results, to say the least.
Getting images matching one's description required some kind of magic prompting, numerous experiments, and a fair bit of patience. Sometimes still with average results, and it would probably have been easier to get these images from existing pictures on specialized websites or of our own creation.
In addition, existing models were limited to square images, up to 1024 pixels. While such images are fine for social network use, I feel they are not as suited for display on desktop websites.
With the release of DALL·E 3 last year, it seems to me OpenAI has reached a quality level where generating images as part of 4AI, inside your Joomla website, could become both usable and useful. In addition, DALL·E 3 handles square images as before but also vertical or horizontal images up to 1792 pixels.
And so 4AI 4.0, released today, implements not only image creation but also image modification!
Image generation
In all operations in 4AI, my goal is to:
- simplify so-called "prompting", that is giving instructions to the AI to obtain the desired output
- integrate easily that result into your Joomla website
4AI already helps you Transform or Create content, doing SEO work, create content for Social Networks, and lastly Chat directly with either GPT-3 or GPT-4.
Now another tab lets you generate and use these generated images on your website with just a few clicks:
Image modification?
What is image modification, and why do we need it? Despite major improvements in quality, sometimes the images generated by the AI model may not follow your description exactly. Or they have one or more areas where you would like to change something rather than asking for the creation of an entirely new image.
This is where image modification happens:
- Take the image originally created by the AI
- Select an area to modify with your mouse
- Describe what you want to change
- Click the button!
Or as a moving image is always clearer:
This allows an iterative process that can give you better results sometimes. A word of caution: image modification is only possible at the moment using the DallE-2 model. It's the older, cheaper model from OpenAI, which:
- can only handle square images, up to 1024 x 1024 pixels
- I find less able to understand instructions
I suspect DallE-3 will soon be able to not only generate but also modify images, and that will likely be a nice improvement.
And now use that image!
But that process would not be complete, nor really useful without easy ways to use the generated image. Once you're happy with the result, using the image is also one click away.
First click the Save Image button. This will not only save the original png generated image to your server but also create different;
- sizes (small, medium, large)
- format (jpg, webp)
All of them are now regular images on your server, that you can use through the media manager as usual.
And then another set of buttons let you:
- use as Intro text image
- use as Full text image
- Insert at cursor
In all cases, the image Alt text will also be added and when inserting at the current editor position, a fully responsive image element is used, for the best Core Web Vitals performance.
What else?
- Both 4SEO and 4SEF are also having maintenance updates, with a few fixes and some smaller improvements in various areas
Until next time, be sure to keep all your websites and extensions up to date!
Cheers,
Yannick