I spent a lovely snowy afternoon using Adobe Firefly to generate images to for a 2.1″ 480x480px round TFT display. The results look amazing.
Using the IOT capable Adafruit Qualia Board and 2.1″ Round TFT with capacitive touch it’s programmed using CircuitPython.
I look forward to getting it animated and taking advantage of the touch and internet capabilities.
Generative fill in Adobe Photoshop (beta) is a pretty amazing tool to extend images. Background artists, web designers, print layout artists, or anybody who deals with incorrect aspect ratios or odd image dimentions will be thrilled by the output.
The image below represents a photo I took near Iceberg Lake in Whistler BC. Portrait aspect on 1920×1080 canvas.
To extend the image CTRL+SELECT image.
Select Inverse – COMMAND+SHIFT+I
SELECT > MODIFY > EXPAND – 3px
EDIT > Generative Fill…. (No Prompts)
En Voila! a landscape that does not exists.
Notice how it recognizes the light source, maintains correct shadows, tones and palette.
Two thirds of the above image was created by AI/ML yet the middle remains completely untouched and my creation. Does this mean the image was “generated by AI”? “Enhanced by AI”? “Extended by AI”?
Or just fake?
Prompts allow for text based editing and generation with mixed results. Note that prompts will only add to the selection. In this case outside of the main image here.
Ultimately however it finally gets classified, anyone looking to extend beyond an image will be happily surprised with the results of Generative Fill.
“Degenerate Phil” on the other hand… t
Text prompt “Degenerate Phil” creates some nightmare fuel.
Nighty-night.
Generative AI is neither good nor evil. It’s up to you, the human in the loop, to decide.
Adobe Firefly Express (beta) whatever “Text to Image” takes the old Colab Notebooks to a new refined level.
However it’s still easy to cherry-pick “good” images for every 10 mediocre or just plain wrong images.
Don’t trust the robots. PEACE. T.AKE IT IT ELASY!
* All images above generated by Adobe “text to image” software.
[EDIT]
I’m crossing the streams. OpenAI conversation where I phrase bomb “T.AKE IT IT ELASY”. A generated text-to-image response for “Take it easy”
Would this be considered adversarial AI?
I’ll admit I’m a HUGE believer of advances in Artificial Intelligence. Deep learning or just general machine learning. From Bayesian beginnings to GANs and CNNs, I’m all about data, ML and all of the promises it’s future holds. The problems it can solve today seemed impossible just 10 years ago. So naturally when I learned of the 2021 release of Atlas of AI I immediately ordered it.
When it arrived I dove into the text, eager for more knowledge and understanding in this exciting field.
But instead of excitement and being a sponge for fresh knowledge, I found the first chapters draining. Like earth-depleting depressing.
At first I pushed back in my mind – “Not that bad” etc.
Honestly I think in my exuberance I didn’t even read the book’s subtitle “Power, Politics, and the Planetary Costs of Artificial Intelligence”.
I know there is a cost to all technology and literal or figurative “mining” excel at wasting resources, stressing environments but I wasn’t ready to read more doom and gloom at that time. The book went on the shelf.
Fast forward over a year later to my first fossil-fuelled odyssey in three years. A 10,000 km round-trip flight, 10+ hours in the air and I need a book…
ROUND 2: I began Atlas of AI again and re-read those first chapters with much less scrutiny, bias, and a far more open mind. As someone who has paid to use heavy GPU time on Colab to generate frivolous, yet remarkable images, I may need to look in the mirror.
Once I shed my reactionary defence of all things ML I began to understand and even appreciate the author’s point of view. I do believe many corporations will also begin to weigh ethical and environmental costs of Artificial Intelligence.
Today’s announcement by Microsoft to retire facial inference of emotional state is just one instance where I see more companies coming to terms with the ethics of AI. Perhaps guided by this book.
A must-read for anyone practicing ML
I’ve been really impressed with the Merlin App for Android Sound ID.
There have been a few birds that I have not been able to spot or identify visually but using Sound ID I can quickly pull up and confirm which bird I’m hearing.
The Pacific Wren has an amazing song and I noticed for years the calls but I’ve still yet to see one.
Also particular impressive was it’s recognition of the Stellar’s Jay non-screechy vocalizations.