Generative AI (GenAI) is a form of Artificial Intelligence that may create a wide selection of images, video and text. To speed up the robot rebellion I chained two GenAI models together to construct a camera which describes the present scene in words, after which uses a second model to create a latest generated stylised image. Let me introduce GenPiCam — a RaspberryPi based camera that reimagines the world with GenAI.
The heavy processing and true smarts of this project is handled by Midjourney — an external service using machine learning-based image generators. GenPiCam makes use of two Midjourney capabilities
- Describe which starts with an existing photo and creates a text description prompts for the image.
- Imagine which converts natural language prompts into images
Between these two steps I allow of a level of creative input, so the GenPiCam camera has a dial to tweak the sort of the ultimate image. This essentially becomes a filter, adding an “anime”, “pop-art” or “futuristic” influence to the generated image.
Sure — here’s the two minute summary
The initial photo image is taken with a Raspberry Pi Camera Module. An external camera shutter (pushbutton connected to the Raspberry Pi GPIO pins) when pushed takes a still image and saves the photo as a jpeg image.
The photo is uploaded to Midjourney which starts with an existing photo and creates a text description prompts for the image. For the curious, I’m using some very inelegant bot interactions with PyAutoGUI to regulate the mouse and keyboard (as there’s no API) — let this be an example of code you must never write.
Midjourney’s describe tool takes a picture as input, then generates text prompts. This can be a pretty clever service, reversing the standard strategy of “text to image” by doing the reverse, starting with the photo after which extracting text to explain the essence of the image. Here is Snowy, but Midjouney has a way more expressive description.
black cat laying on bed under yellow blanket, within the sort of berrypunk, irridescent, glimmering, unpolished, symmetrical, rounded, chinapunk — ar 4:3
The describe function actually returns 4 descriptions based on the image, but GenPiCam arbitrarily selects the primary description.
Now for the fun part. We will take that text prompt, and use it to create a brand latest image with Generative AI with a latest call to Midjouney imagine. Here’s a image generate from the previous text prompt.
GenPiCam has a variety switch to update the prompt with stylistic instructions.
This can be a 12 way rotary switch connected to the Raspberry Pi GPIO pins. By reading the present “artistic selection” GenPiCam will add a prefix equivalent to “ to the text prompt. A couple of of the opposite style prompts include
- Anime style
- Hyper Realistic, whimsical with vibrant hat and balloons,
- Blurry brushstrokes,
- Futuristic, in an area station, hyper realistic
Let’s see the before and after “pop-art” images for Snowy.
The ultimate image is a created using the Pillow Python imaging library, and is comprised of
- Initial photo taken by the Raspberry Pi camera module, resized on the left
- Final Midjouney image — the primary of 4 images is chosen, composited to the suitable
- Text prompt — against a colored background and icon signifying the style mode
Here’s the identical process, but adding the text
Although the image on the suitable is a creation from Generative AI, there’s still still a way of disappointment coming through Snowy’s judgmental eyes.
I had a lot fun constructing the GenPiCam camera — and this was an interesting path for exploring prompt engineering for Generative AI. The higher photos were those which had a straightforward composition — essentially images that were easy to place words to. For instance, this scene is straightforward to explain with a color and definitive objects.
Nevertheless, there have been some very strange results while describing more unique scenes. I discovered the outline of a classic Australian cloths line created a peculiar image.
One in all my favourite reimagined images was the identification of my laser mouse. It seems a laser mouse has multiple meaning resulting in a striking result.
The least stylish a part of GenPiCam is the hardware which I rapidly assembled. If you need to construct your individual reality distorting camera, you’ll need the next.
It isn’t essentially the most beautiful of builds — but I’ll just excuse this as being highly functional
The GenPiCam has been a fun solution to explore Generative AI, transforming photos into stylised (and sometime surprising) images.
… [Trackback]
[…] Here you can find 8788 additional Info on that Topic: bardai.ai/artificial-intelligence/genpicam-generative-ai-cameraim-bored-can-i-get-a-videothe-photographic-processgenerative-ai-images-learningsthe-hardwaresummary-code-credits/ […]
… [Trackback]
[…] Find More here on that Topic: bardai.ai/artificial-intelligence/genpicam-generative-ai-cameraim-bored-can-i-get-a-videothe-photographic-processgenerative-ai-images-learningsthe-hardwaresummary-code-credits/ […]
… [Trackback]
[…] Find More to that Topic: bardai.ai/artificial-intelligence/genpicam-generative-ai-cameraim-bored-can-i-get-a-videothe-photographic-processgenerative-ai-images-learningsthe-hardwaresummary-code-credits/ […]
… [Trackback]
[…] Read More to that Topic: bardai.ai/artificial-intelligence/genpicam-generative-ai-cameraim-bored-can-i-get-a-videothe-photographic-processgenerative-ai-images-learningsthe-hardwaresummary-code-credits/ […]
sweet jazz
relaxing music
… [Trackback]
[…] Read More to that Topic: bardai.ai/artificial-intelligence/genpicam-generative-ai-cameraim-bored-can-i-get-a-videothe-photographic-processgenerative-ai-images-learningsthe-hardwaresummary-code-credits/ […]
… [Trackback]
[…] Information on that Topic: bardai.ai/artificial-intelligence/genpicam-generative-ai-cameraim-bored-can-i-get-a-videothe-photographic-processgenerative-ai-images-learningsthe-hardwaresummary-code-credits/ […]