Is the First Generative AI Camera on the Horizon?

Artificial Intelligence

by Omri Keren LapidotJuly 7^th, 2024

The CMR M-1 is a new concept camera. Not yet set for mass marketing, it also requires a somewhat inclusive take on the fundamental idea of a camera. The device is co-developed by SpecialGuestX, an interesting creative technology agency, and 1stAveMachine, a global mixed-media production company. As you’ve probably noticed, no traditional camera manufacturers are on board. Quite a hint regarding the product. So what is the CMR M-1?

The rather simple, minimalistic design of the CMR M-1 Generative AI camera covers a rather complex contraption. The camera uses a FLIR sensor (Front Looking Infra Red) of an undisclosed size (but considering the showcased lenses, I’d expect it to be no larger than Micro Four Thirds). The choice of an IR-capturing sensor is interesting, though no related features are showcased as of now. The CMR M-1 is a massive, box-shaped camera whose design harkens back to 16mm film cameras. This is true for much more than the physical design.

On board generative AI

The AI side of the CMR-M1 is based on Stable Diffusion algorithms, with five different LoRAs available. LoRA, Low-Ranking Adaptation, is a method of fine-tuning Stable Diffusion checkpoints. Without diving too deep, it resembles an AI LUT or film simulation regarding its effect on the workflow and end result.
The LoRAs are loaded onto minimalistic-designed cards. These have a dedicated slot in the camera. Once the LoRA is loaded, the intensity of the generative AI effect is tuned via a large silver-black dial on the right-hand side. Though the CMR-M1 is larger than some powerful desktop computers, generative AI is done via external servers. This brings some challenging questions regarding connectivity that are yet to be answered.

The Camera part of the “Generative AI Camera”

The specs of the CMR M-1 are quite poor compared with the current roster of cameras. Resolution maxes out at an odd 1368×768 with a whopping 12fps frame rate. Yeah, that’s twelve frames per second. No additional info regarding ISO, dynamic range, or color depth. By now, many of our readers might be wondering why we are discussing this seemingly underwhelming camera. Some may speculate that we’re reporting on this simply because AI is currently trending, and who wouldn’t want to capitalize on some juicy SEO traffic? However, I’ll argue that the interesting part about this clickbait-titled “First Generative AI Camera” lies not in its capturing specs but in how it challenges the concept of the camera itself.

What makes a camera?

A camera is quite simple to define. It’s a light-capturing device, able to fix light onto readable media. A camera requires no more than three basic elements:

• A light-focusing apparatus, be it an optical lens or a pinhole
• A light-sensitive surface with the ability to fix the captured image
• A chamber connecting these two in utter darkness. This one gave cameras their name, short for “Camera Obscura.”

The CMR-M1 includes all three; hence, it is considered to be a camera. But it adds something we haven’t seen yet – a generative AI algorithm. Or did we? Aren’t noise reduction algorithms somewhat AI-based? Isn’t automatic white balance based on machine learning? And what about modern auto exposure? What should we call modern autofocus algorithms? Though there are some differences between the level of generative AI offered by the CMR-M1 and these examples, there are also some significant similarities.

SOOC redefined

SOOC, Straight Out Of Camera, is a term used to describe images untouched by software. Used as a declaration of authenticity in an age of Photoshop and Instagram filters, it holds very little truth. Every single ray of light captured by digital or analog media is subjected to rather heavy interpretation before it transforms into a viewable image. Let’s take the common Bayer array as an example.

Each photosite captures either a red, green, or blue channel. The other two values required for the RGB output are calculated, following complex and secretive algorithms every company fiercely guards. Now, let’s add color profiles, another layer of interpretation (or manipulation) made by the camera. Analog workflow is no different in that regard. Our choice of film stock is similar to every digital filter. The single difference lies with timing. We can only choose the film stock before shooting, while digital workflow allows for much more potent post-production. Taking that trait into consideration, I’ll argue that the CMR-M1’s workflow has more in common with analog filmmaking than with its digital counterpart.

A glimpse into the past

“You press the button, we’ll do the rest.” Kodaks’s legendary slogan stood as a major inspiration in the design process of the CMR-M1. In a way, I think the companies who created this concept product managed to recreate this experience to a level you (and I) wouldn’t expect from any device, let alone “The first Generative AI camera.” The raw unpredictability of real-time generative AI may just offer a new level of shock and awe, especially when compared with modern cameras able to capture technically superb footage with little to no skills or know-how on the photographer’s part. It’s not just the vintage design and the tactile card-based LoRAs that provide the gist Kodak intended. It’s something deeper, fundamental, the unique core of the CMR-M1.

A glimpse into the future

As Aaron Duffy, founder and executive creative director of SpecialGuestX, puts it: “Sometimes, to imagine what the future might be like, you have to prototype it.” This quote encapsulates the CMR-M1. It’s a prototype. It provides a glimpse into features that may trickle down into future cameras. While I don’t see this level of generative AI coming to professional cameras in the foreseeable future, I do think some versions of these abilities may find their way into our field. Imagine a camera that can generate a resolution higher than the captured resolution. This will enable more affordable sensors, faster capture speeds, etc. Imagine a camera that can manipulate lighting and requires much less on-location setup. It may sound like sci-fi now, but wasn’t everything we have today sci-fi a couple of decades ago?

The CMR-M1 is only a prototype. No mass market planned, no “brochure features,” no cost cuts. We won’t get our hands on it, yet this camera may give us a glimpse into what cameras may become and what they already are.