appeal of film over digital?

[raydm6]

Not sure if the answer below is 100% correct, but if it is (as like Guth alluded to), where do we draw the line?

Do we stop allowing digital images from the major camera manufacturers - and newer smart phones for that matter - which also use AI, on RFF? I know I post a lot of iPhone pics here.

I just asked ChatGPT:

Which digital camera makers are using AI in their software?

Answer:
————————-
Several digital camera manufacturers have begun incorporating AI in their software, enhancing features like image processing, autofocus, scene recognition, and even post-processing. Some of the notable brands using AI technology include:

1. **Canon**:
- Canon has integrated AI in various ways, especially in its autofocus systems (like Dual Pixel AF), face and eye detection, and subject tracking in both cameras and lenses. AI is also used for image optimization and noise reduction in post-processing, such as in their software like Digital Photo Professional.

2. **Sony**:
- Sony uses AI in its Real-Time Eye autofocus and Real-Time Tracking, which automatically detects and tracks subjects with remarkable precision. AI is also utilized in image enhancement through their in-camera processing, as well as in features like the Creative Styles for better image processing.

3. **Nikon**:
- Nikon's Z-series mirrorless cameras leverage AI for subject detection, including faces, eyes, and animals, to improve autofocus accuracy. AI is also used in the noise reduction process and optimizing image quality in varying light conditions.

4. **Fujifilm**:
- Fujifilm has incorporated AI into their film simulations and automatic scene recognition modes. Their cameras adjust settings based on the subject, lighting conditions, and environment using machine learning to optimize exposure, white balance, and focus.

5. **Panasonic**:
- Panasonic's Lumix series uses AI for autofocus and tracking moving subjects, especially in their higher-end models. They also use AI for face and eye detection, as well as in noise reduction and image processing for better low-light performance.

6. **Leica**:
- Leica has been incorporating AI into their M-series and SL-series cameras, especially for autofocus and face detection. They also use AI in post-processing features through their software like Adobe Lightroom (which integrates well with Leica’s RAW files).

7. **Olympus (OM Digital Solutions)**:
- AI is used in Olympus's face detection, autofocus systems, and scene detection to optimize exposure and white balance. Their AI-driven image stabilization system also helps in capturing sharper images in low light.

8. **GoPro**:
- AI is part of the automatic editing features in GoPro cameras, especially with their "HindSight" and "TimeWarp" modes, where AI helps choose the best moments and adjust video stabilization.

Beyond camera manufacturers, AI is also embedded in third-party software and apps used for editing and processing photos, such as Adobe Lightroom (with its AI-powered "Sensei" tools), DxO's AI noise reduction software, and Skylum’s Luminar AI.

Do any of these brands or features interest you?
—————End————-

 

[raydm6]

I also asked:
“Tell me more about AI in photography?”

Answer:
————————-
AI is becoming an increasingly powerful tool in photography, helping both amateurs and professionals streamline their workflows and enhance their images in ways that were once impossible or too time-consuming. Here's an overview of how AI is being used in the photography world:

### 1. **Auto Focus and Subject Tracking**
AI has greatly improved autofocus systems, making them faster, more accurate, and capable of tracking subjects in motion. Cameras can now identify and lock onto faces, eyes, animals, or even specific objects, keeping them in focus regardless of movement. For example:
- **Eye and Face Detection:** AI can prioritize focusing on the eyes or face of a subject, ensuring sharp portraits even in challenging lighting or when the subject moves.
- **Real-Time Tracking:** Sony's Real-Time Eye Autofocus or Canon's Dual Pixel AF use AI to continuously track subjects, whether they’re moving across the frame or shifting in and out of focus.

### 2. **Scene and Object Recognition**
AI can now recognize different scenes or objects and adjust settings accordingly. This is often seen in smartphone cameras, but high-end mirrorless and DSLR cameras are also using similar features:
- **Auto Scene Detection:** AI algorithms can analyze a scene and adjust camera settings like exposure, ISO, and white balance for optimal results. For example, if you're photographing a sunset, the AI will adjust to capture the vibrant colors without overexposing the highlights.
- **Subject Recognition:** AI can identify whether you're shooting a person, animal, or landscape and apply the most suitable settings automatically.

### 3. **Image Enhancement and Post-Processing**
AI is revolutionizing post-processing, either through built-in camera features or through software. Some of the most impressive AI-powered image enhancements include:
- **Noise Reduction:** AI-based software, such as DxO's DeepPRIME or Topaz DeNoise AI, uses deep learning models to analyze an image's noise patterns and intelligently reduce noise while preserving detail and sharpness.
- **Upscaling and Resolution Enhancement:** AI is also used to upscale low-resolution images or enhance detail in areas that would normally suffer from a loss of sharpness or clarity. Tools like Adobe Photoshop’s "Super Resolution" and Topaz Gigapixel AI use machine learning to intelligently add pixels to images, creating higher-resolution results.
- **Color Grading and Style Transfer:** AI-driven apps like Skylum Luminar AI can analyze an image and apply specific editing styles or color grading that would normally take hours to do manually. AI can also recreate particular “looks,” such as mimicking the aesthetics of classic film stocks or even painting styles.

### 4. **Depth Mapping and Bokeh Effects**
AI is increasingly used in computational photography to simulate the effect of shallow depth of field (bokeh) in portraiture, even with cameras that lack large sensors or fast lenses:
- **Portrait Mode (Smartphones):** Modern smartphones, like the iPhone or Google Pixel, use AI to simulate a blurred background effect (bokeh) by recognizing the subject and separating it from the background, even if the camera’s lens is not capable of producing a natural bokeh.
- **Depth Mapping:** Cameras like those from Huawei and Samsung employ AI to create depth maps of a scene, allowing them to apply selective focus and create more refined bokeh or depth-of-field effects.

### 5. **Automatic Image Curation and Editing**
AI is also improving how images are curated, sorted, and edited post-shoot. This is especially helpful when dealing with large volumes of photos, like event photography or social media content creation:
- **AI Curation:** Apps like Adobe Lightroom use AI to auto-tag and categorize photos based on the content, such as recognizing specific people, locations, or objects, making it easier to search and organize large photo libraries.
- **Auto Editing:** AI can help with basic editing tasks such as adjusting brightness, contrast, or white balance automatically. In some cases, it can even apply advanced edits like portrait retouching or sky replacement with minimal user input.

### 6. **AI in Computational Photography (Smartphones)**
One of the biggest areas where AI is having an impact is in smartphones, where computational photography uses AI to enhance camera capabilities beyond the physical limitations of the hardware:
- **Night Mode & Low Light Photography:** AI algorithms analyze the scene and combine multiple exposures to create a bright and noise-free image, even in near-darkness. The Google Pixel, for instance, uses machine learning to produce stunning results in low-light conditions.
- **HDR (High Dynamic Range):** AI helps balance the bright and dark areas of an image, producing more natural-looking results without overexposure or underexposure, especially in scenes with high contrast, like sunsets or backlit subjects.

### 7. **AI in Video**
AI is not just for still images. Video recording has also benefited from AI technology:
- **Auto Framing and Tracking:** Some cameras use AI to track subjects in video, keeping them in the frame or focused, even as they move. This is commonly seen in AI-based cameras for vlogging or YouTube content creators.
- **Stabilization:** AI-based software helps with video stabilization, reducing shaky footage without the need for external gimbals. GoPro’s HyperSmooth, for example, uses AI to keep videos smooth even in extreme conditions.

### 8. **Future Potential**
The possibilities with AI in photography are still growing:
- **Creative Composition Assistance:** AI could help photographers compose shots based on visual rules or artistic suggestions, guiding you to create better images in real time.
- **Real-time Image Enhancement:** Cameras could potentially offer real-time, AI-powered post-processing that gives photographers instant previews of how an image will look after editing.

In essence, AI is making photography more accessible, automating the more technical aspects while allowing photographers to focus on the creative side. Whether through enhanced autofocus, automatic editing, or noise reduction, AI is helping both professionals and hobbyists capture stunning images with less effort and more control. Are there any specific AI features or tools that interest you the most?

—————End————-

 

[santino]

AI is the beginning of the end, not only the end of „photography „. 😗

 

[Godfrey]

It is hard to know just where to draw the line as digital cameras are very much reliant on computers to capture an image. Thus one could say that the files produced by these cameras are computer generated art.

A camera and a computer are two very different things ... There's a big difference between generating data that forms an image and capturing light that is processed into data to form an image.

What about rendering film images with image processing software? Or scanning prints into JPEG files so you can show them on this forum? Are you going to tell me that that is "computer generated art?" I say, "Hogwash."

G

 

[Godfrey]

6. **Leica**:
- Leica has been incorporating AI into their M-series and SL-series cameras, especially for autofocus and face detection. They also use AI in post-processing features through their software like Adobe Lightroom (which integrates well with Leica’s RAW files).

This is fascinating. I didn't know Leica M-series cameras had autofocus and face detection. Or that Adobe Lightroom was Leica software. 🤮

G

 

[peterm1]

It is hard to know just where to draw the line as digital cameras are very much reliant on computers to capture an image. Thus one could say that the files produced by these cameras are computer generated art.

Personally it does not much matter to me as long as the artist (i.e. the photographer) is in control. Technology changes but perhaps (leaving things like A.I. aside) it does not control the outcomes. It simply provides new tools for the artist.

Interestingly this same kind of debate was going on over 100 years ago between the traditionalists who maintained that image making was not art unless it was painted by a "real" painter and the Photo Secessionists / Pictorialists, who believed that cameras and (analog) film were simply new tools in the hands of artists. They argued as follows:

" Pictorialists believed that, just as a painting is distinctive because of the artist’s manipulation of the materials to achieve an effect, so too should the photographer alter or manipulate the photographic image. Among the methods used were soft focus; special filters and lens coatings; burning, dodging and/or cropping in the darkroom to edit the content of the image; and alternative printing processes such as sepia toning, carbon printing, platinum printing or gum bichromate processing." (Quoted from Wikipedia)

Now instead of using the analog methods mentioned above, we are free to use digital technology instead of, or in addition to these methods to achieve the artistic outcomes that we visualize and seek. (And to leave some of the mundane decisions to the firmware in the camera). In other words, the users of analog technology both had, and felt free to use, their suite of tools. We have in addition, our suite of tools and should feel equally free to apply them to achieve the artistic results we envisage. Some of this happens in camera. But we are to a significant extent still in control as to how much control we give to technology. For example, I own a Fuji film camera which has substantial ability to introduce interesting and artistic effects in camera. Or I can turn them off and pretty much replicate those effects in post. I think both are valid.

 

[Retro-Grouch]

I suppose we could go back to 1983, with the introduction of the Nikon FA, to determine when AI was introduced into photography. Its matrix metering relied upon a built-in database programmed with the characteristics of several thousand (randomly chosen?) photographs. It consulted this database to choose an exposure based on those stored characteristics. Rudimentary, by today's standards, but still AI, I'd say.
I would add that I didn't trust matrix metering then, and don't now. I want to be in control, and I want to depend on my expertise, such as it is, for the proper exposure. And when I screw up an exposure, at least it's because of my decision, and I can learn from my bad decision. I'm sure people will jump in here to argue that matrix metering (and automation in general) is the best thing since sliced bread. Fine, but I like to feel that as many decisions as possible in the process are mine. Cranky old guy, here. No apologies for it, either.

 

[Retro-Grouch]

A camera and a computer are two very different things ... There's a big difference between generating data that forms an image and capturing light that is processed into data to form an image.



G

When AI creates an image, it does so by consulting a database of millions of photographically created images and constructing something from that database that uses your verbal prompts to determine its parameters. Possibly, one of those millions of photographically created images used in its training was one of yours, vacuumed up from the internet. So... As they say, "It's complicated."

 

[Freakscene]

I suppose we could go back to 1983, with the introduction of the Nikon FA, to determine when AI was introduced into photography. Its matrix metering relied upon a built-in database programmed with the characteristics of several thousand (randomly chosen?) photographs. It consulted this database to choose an exposure based on those stored characteristics. Rudimentary, by today's standards, but still AI, I'd say.
I would add that I didn't trust matrix metering then, and don't now. I want to be in control, and I want to depend on my expertise, such as it is, for the proper exposure. And when I screw up an exposure, at least it's because of my decision, and I can learn from my bad decision. I'm sure people will jump in here to argue that matrix metering (and automation in general) is the best thing since sliced bread. Fine, but I like to feel that as many decisions as possible in the process are mine. Cranky old guy, here. No apologies for it, either.

The algorithm in the Nikon FA metering is not AI. It simply chooses the similarity between the scene you point the camera at and a series of metering patterns and their ‘optimal’ metering outcomes. It cannot add information from the photographs it takes. A child can understand the algorithm. The difference between it and large image models - even though these are not AI, no matter what anyone says - they do not think, they pattern match and have no capacity for self-discrimination - is acquisition. The FA metering cannot grow, unless those patterns are added manually by a user, which, I should add, the hardware does not have room for.

 

[Retro-Grouch]

The algorithm in the Nikon FA metering is not AI. It simply chooses the similarity between the scene you point the camera at and a series of metering patterns and their ‘optimal’ metering outcomes. It cannot add information from the photographs it takes. A child can understand the algorithm. The difference between it and large image models - even though these are not AI, no matter what anyone says - they do not think, they pattern match and have no capacity for self-discrimination - is acquisition. The FA metering cannot grow, unless those patterns are added manually by a user, which, I should add, the hardware does not have room for.

I'll defer to your greater knowledge, and I see how my use of the term was in error. Nevertheless, my aversion to matrix metering stands!

 

[Coldkennels]

Not sure if the answer below is 100% correct, but if it is (as like Guth alluded to), where do we draw the line?
[...]
I just asked ChatGPT:

Which digital camera makers are using AI in their software?
[...]
6. **Leica**:
- Leica has been incorporating AI into their M-series and SL-series cameras, especially for autofocus and face detection.

Ah yes, the M series, well known for its face detection autofocus.

Why anyone asks ChatGPT and expects a reliable answer on anything absolutely boggles my mind.

Edit:

This is fascinating. I didn't know Leica M-series cameras had autofocus and face detection. Or that Adobe Lightroom was Leica software. 🤮

Ah, Godfrey beat me to it. But the point remains!

 

[Guth]

A camera and a computer are two very different things ...

Not when it comes to digital cameras. If the camera contains a sensor to capture light, then the camera is essentially a computer. Basically you have a computer with a lens attached. We call them cameras because that is their primary function, but they are very much computers and without those computers they simply don’t work. A number of film cameras also contained computers, but not all film cameras were/are totally dependent on computers as digital cameras are.

There's a big difference between generating data that forms an image and capturing light that is processed into data to form an image.

I understand what you are trying to get at, but once a sensor is involved it is all data. That’s why it is called digital photography. Yes, there is a difference, but in the end it’s all just 1s and 0s that are being processed by computers.

What about rendering film images with image processing software? Or scanning prints into JPEG files so you can show them on this forum? Are you going to tell me that that is "computer generated art?" I say, "Hogwash."

In effect, yes. Once a person scans their film, they have entered the realm of digital photography. Like it or not, it IS computer generated. Here’s a challenge for you: Let’s see what you are able to produce with your image processing software without a computer to run it on. For that matter, let’s see what you can produce with your digital cameras with their image processors (computers) removed. Without the computer to generate the data my guess is that you will get nothing.

Like I pointed out in my initial post, I am fine with others using whatever they enjoy using. I’m not an analog purist as I also noted that I use the digital camera in my smartphone (which is a computer of course). I also use image processing software and film scanners (which are also computers). I am not trying to bag on myself or anyone else that has embraced digital photography in any form. I’m not trying to tell you what to do, nor did I attack you in my writing. So why are you so upset over what I think?

 

[Godfrey]

I suppose we could go back to 1983, with the introduction of the Nikon FA, to determine when AI was introduced into photography. Its matrix metering relied upon a built-in database programmed with the characteristics of several thousand (randomly chosen?) photographs. It consulted this database to choose an exposure based on those stored characteristics. Rudimentary, by today's standards, but still AI, I'd say.
I would add that I didn't trust matrix metering then, and don't now. I want to be in control, and I want to depend on my expertise, such as it is, for the proper exposure. And when I screw up an exposure, at least it's because of my decision, and I can learn from my bad decision. I'm sure people will jump in here to argue that matrix metering (and automation in general) is the best thing since sliced bread. Fine, but I like to feel that as many decisions as possible in the process are mine. Cranky old guy, here. No apologies for it, either.

That is "pattern matching" not "artificial intelligence" ... The hallmark of artificial intelligence is adaptive learning; the Nikon FA never "learned" anything at all, it simply ran through a very simple algorithm each time it was powered up.

I wrote a similar algorithm as part of my Mathematics degree final project work in 1980 as I was focusing on statistical pattern recognition.

Please do not make the mistake of assigning the label of AI to every bit of computation. My older brother was a researcher in AI for 23 years and his combined assessment was that vast amounts of what is called AI has nothing to do with AI from the technical sense. Most of what everyone is calling AI is simply a modest, simple bit of programming, not adaptive learning.

G

 

[Godfrey]

Not when it comes to digital cameras. If the camera contains a sensor to capture light, then the camera is essentially a computer. Basically you have a computer with a lens attached. We call them cameras because that is their primary function, but they are very much computers and without those computers they simply don’t work. A number of film cameras also contained computers, but not all film cameras were/are totally dependent on computers as digital cameras are.

Absolutely not. A digital sensor is called that because it senses photons and encodes them as a sequence of ones and zeros ... digits ... which are read out and stored. It's just a little machine with a pile of switches that retain the state of on and off to represent the photons.

Calling a camera a computer in the manner in which you do is just absurd. By your rationale, my toaster is also a computer, as is my refrigerator. Really.

I understand what you are trying to get at, but once a sensor is involved it is all data. That’s why it is called digital photography. Yes, there is a difference, but in the end it’s all just 1s and 0s that are being processed by computers.

There's little difference between chemical potentials and electrical potentials. They are both data. Film data is expressed as images by shining light through the gelatin and looking at the shadows it casts. Electrical potentials are expressed by using sensing that relays the state into bright and dark spots in a raster device that forms the image.

It is obvious that you dislike anything having to do with your notion of "computing". I see no reason to participate further in this nonsensical discussion.

Sayonara.
G

 

[CMur12]

I have great respect for digital photography and for computers, and I have no problem seeing a digital camera as primarily a computer. In fact, that's how I see it and I see nothing negative therein. It starts with digital data (albeit triggered by photons) and it ends with digital data.

- Murray

 

[Retro-Grouch]

That is "pattern matching" not "artificial intelligence" ... The hallmark of artificial intelligence is adaptive learning; the Nikon FA never "learned" anything at all, it simply ran through a very simple algorithm each time it was powered up.

I wrote a similar algorithm as part of my Mathematics degree final project work in 1980 as I was focusing on statistical pattern recognition.

Please do not make the mistake of assigning the label of AI to every bit of computation. My older brother was a researcher in AI for 23 years and his combined assessment was that vast amounts of what is called AI has nothing to do with AI from the technical sense. Most of what everyone is calling AI is simply a modest, simple bit of programming, not adaptive learning.

G

Thanks for clarifying. Freakscene already pointed out the error in my thinking, too. Mea culpa!

 

[Freakscene]

That is "pattern matching" not "artificial intelligence" ... The hallmark of artificial intelligence is adaptive learning; the Nikon FA never "learned" anything at all, it simply ran through a very simple algorithm each time it was powered up.

I wrote a similar algorithm as part of my Mathematics degree final project work in 1980 as I was focusing on statistical pattern recognition.

Please do not make the mistake of assigning the label of AI to every bit of computation. My older brother was a researcher in AI for 23 years and his combined assessment was that vast amounts of what is called AI has nothing to do with AI from the technical sense. Most of what everyone is calling AI is simply a modest, simple bit of programming, not adaptive learning.

G

Indeed. The current large model systems such as Chat GPT are not really learning themselves, they are just large models that are trained. So far, as Brian Reid wrote on an exam for Doug Lenat “AI is bullshit” seems to remain true.

 

[sojournerphoto]

The bottom line in all art is the provenance of the art object. Cleaning dirt (and other 'noise') out of an image in rendering it to enhance the capture has been done since the beginning of Photography, so I see little issue with that. Photographic compositing (merging photographic images into a coordinated whole in as seamless a way as possible) is also a perfectly acceptable photographic endeavor ... see the work of Jerry Uelsman, the master of this kind of photographic art.

The key is that all of these forms of art include provenance that testifies to what the image(s) are about (or "intent by the artist" to use another description), when they were captured, and how they were rendered. You can't do that if your imaging process is "tell the AI software to generate a smiling woman with a baby in her arms that looks like Aunt Darlene" ... Provenance in that case can say what the instructions were, what AI tool was used, who dreamed up the idea for the image, etc, but it CANNOT say that the image is a photograph. It's computer generated art, that's all.

If you have AI tools that can take your photograph, isolate the key subject you wanted, and then composite it into a lunar landscape setting ... Well, I can accept that the provenance in that case can articulate some of the inclusions to the art work, but the art piece is still not a Photograph ... it's a painting which includes a photographic element.

That's not what I produce or am interested in producing, but I've seen stuff like that which is pretty nicely done.

The important thing is that the provenance of whatever art piece or photograph you make is clear and honest. If it isn't then the piece is not credible, whether or not it might be considered a photograph or computer graphics.

Commercial art, whether photographs or CGA ... the intent is always to display or promote something, usually for profit. So the provenance of an image used for this kind of purpose is rather simplistic. Much more critical is the provenance of images used for documentarian and forensic purposes. Nothing "generated" by AI can be truly documentary in nature, it can at best be an artist's simulation/recreation.

G

The last statement here is, I think, important. Photographs still tend to be represented as documentarian. Clearly, that is less and less the case - see the Photo of the Year samples linked up thread. The consequence is a justified loss of trust in photography to represent the real world. Even though we all know that straight photography can be manipulative, I think this is a loss for several reasons:

- there is great beauty in what is around us, whether at home, on the streets or out in nature. By constantly seeking to enhance we lose appreciation for the real and come to need the synthetic.

- losing faith in the ability to document leaves us without any sense of the objective and without the means to ascertain what is real, what happened.

- when we lose our notion or ability to discern what is true every man’s and woman’s opinion becomes their own truth, pushed this way and that by the lies and dishonesties that surround us all. We see this now in many areas of life and it’s not helping.

 

[sojournerphoto]

Indeed. The current large model systems such as Chat GPT are not really learning themselves, they are just large models that are trained. So far, as Brian Reid wrote on an exam for Doug Lenat “AI is bullshit” seems to remain true.

Yep, the probabilistic models fail when the data they are fed is so full of… nonsense. In my field you can’t (safely) ask an LLM anything unless you already know the answer.

 

[Freakscene]

Yep, the probabilistic models fail when the data they are fed is so full of… nonsense. In my field you can’t (safely) ask an LLM anything unless you already know the answer.

Students always wondered how I could tell instantly without the online tools that they had used a synthetic model to write their work. Some of them could never quite grasp that “it will be wrong in predictable ways” which I told them in lectures repeatedly was actually a fact. Even after some of them learned that the online checking tools could be fooled by outputting from ResearchRabbit, paraphrasing the text in ChatGPT then checking and rewording through CoPilot, they did not understand that the factual errors remained and were completely clear.