From Silver Halides to Binary Code – an Interview with ChatGPT

Intro:
I wrote this because I’m concerned about how people sometimes react about Film Simulation Recipes,
about the authenticity and accuracy of recipes compared to real film stock, especially in the Fujifilm community.

I’m also worried about how some blog writers criticize AI-generated recipes made by ChatGPT, or commenting or detracting from the accuracy of simulation recipes in general.

I want to know if these criticisms are fair and give ChatGPT a chance to share its thoughts about Film Simulation Recipes in general.

The Interview

My first question was whether ChatGPT could generate an accurate recipe

Here ChatGPT actually indicates that it was not developed for this, but also that with the necessary data it would be possible.
Data could, for example, be a certain film stock, if he recognizes that specific film stock, what would his reaction be ?

And indeed, without showing the full recipe here, ChatGPT has responded positively, and provided a detailed explanation, both about the characteristics of the film stock and the settings on the different cameras.

In 2023, someone writes an article and talks about a film simulation recipe created by ChatGPT based on a specific type of stock film. He writes that photos taken with this recipe aren’t at all comparable with his own photos taken with the actual film and scans of them.

Because this was not an isolated case, but also with a very well-known maker of Fujifilm simulations, I would rather not mention the names and websites of the authors here. The comments and decisions they made are entirely their own. I just want to see if they are based on good research, facts and ultimately if they are true.
So I continue my questions, and ask him about this incident.

Again, ChatGPT admits that it is not designed for this, and also has no visual capabilities to be able to check whether his film recipe has been made correctly. But he says it does offer opportunities to experiment further with the recipe.

I myself have of course read the blog article and seen the comparisons made by its author. What ChatGPT didn’t know is that the article writer compared photos taken in a completely different place and lighting conditions, not to mention the difference in subject; dark buildings, hard shadows, grass and trees in the new digital photos – Analog photos taken on the coast, very light buildings and background, people in colorful clothing, blue sky, etc. -> completely different from each other.
What does ChatGPT think about that ?

Transcribed text from ChatGPT:

“As an AI language model, ChatGPT does not have thoughts, beliefs, or consciousness, so it does not hold opinions about whether a comparison is fair or not. However, I can provide some insights into how the comparison might be perceived:

From a technical standpoint, comparing photos taken with a film simulation recipe generated by ChatGPT with completely different photos taken with the real film stock and scans thereof may not be considered a fair comparison. This is because the photos being compared are likely to have different subjects, lighting conditions, compositions, and other variables that can significantly influence their appearance.

To accurately assess the effectiveness of a film simulation recipe, it’s essential to compare photos taken under as similar conditions as possible. This ensures that differences in appearance can be attributed primarily to the film simulation settings rather than external factors.

However, from a practical standpoint, such a comparison may still provide some insights into the capabilities of the film simulation recipe. While the comparison may not be ideal, it could still offer a rough indication of how well the simulation replicates certain characteristics of the real film stock under different conditions.

Ultimately, the validity of the comparison depends on the specific goals and context of the evaluation. For a more rigorous assessment of the film simulation recipe’s accuracy, it’s advisable to compare photos taken under similar conditions with both the simulation and the real film stock.”

Even if people compare apples and lemons as if they were the same, ChatGPT still sees the good in it – I would have given a slightly more nuanced response 🙂
Shows what a good guy ChatGPT is, doesn’t it ?

So I also ask him whether the comparison we photographers make with real photos and film is actually a good comparison?

Transcribed text from ChatGPT:

“You’ve raised an important point about the reliability of using photos taken on real film stock as a means of comparison. While comparing film recipes with photos taken on real film can provide valuable insights, it’s essential to consider the inherent variability and complexity of the analog film process. Here are some factors to consider:

1. Chemical Processes: The entire workflow of analog film photography involves multiple chemical processes, including exposing the film, developing it, and printing it onto light-sensitive paper in a darkroom. Each of these processes introduces variables that can influence the final outcome, including temperature, agitation, and chemical concentrations. As a result, variations can occur between individual film rolls, batches of chemicals, and printing sessions, even when using the same film stock.
2. Subjectivity in Development: The development process of analog film is not only chemical but also subjective. Different photographers may develop the same film stock differently to achieve desired aesthetic effects or compensate for specific shooting conditions. This subjectivity introduces another layer of variability that can affect the appearance of the final images.
3. Variability in Scanning: When analog film is scanned digitally, additional variables come into play, such as scanner settings, resolution, and post-processing techniques. The choice of equipment, scanning resolution, and color adjustments can all influence the appearance of the digital scans, leading to variations in color rendition, contrast, and overall quality.

Given these factors, relying solely on photos taken on real film as a basis for comparison may not always provide a reliable or consistent standard. While real film photos offer tangible results that can be visually assessed, they are subject to inherent variability and subjective interpretation throughout the analog workflow.

To mitigate these challenges and ensure a more objective comparison, photographers may consider the following approaches:

• Standardized Processes: Establishing consistent and standardized procedures for film development, printing, and scanning can help reduce variability between photos taken with real film.
• Reference Images: Using reference images or standardized test charts as benchmarks for comparison can provide a more objective basis for evaluating film recipes and digital simulations.
• Controlled Experiments: Conducting controlled experiments where photos are taken under identical conditions with both real film and digital simulations can help isolate the effects of each variable and provide more accurate comparisons.

By carefully considering these factors and approaches, photographers can make more informed assessments of film recipes and digital simulations, taking into account the complexities of the analog film process.”

Well, that gives one something to think about, doesn’t it ?

I ask him where we humans go wrong – and his answer is actually what I think about it too:

Transcribed text from ChatGPT:

“When comparing photos taken with real film and digital film simulations, regarding the accuracy of the film simulation recipe, two common mistakes photographers make are:

1. Expecting Exact Replication: One major mistake is expecting digital film simulations to perfectly replicate the look of real film under all conditions. While film simulation recipes aim to mimic the visual characteristics of specific film stocks, they may not always achieve exact replication due to differences in the underlying technologies and processes. Expecting digital simulations to perfectly match the nuances of analog film in every scenario can lead to disappointment and unrealistic expectations.
2. Neglecting Contextual Differences: Another common mistake is neglecting contextual differences between photos taken with real film and digital simulations. Factors such as shooting conditions, lighting environments, and subject matter can significantly impact the appearance of the final images. Comparing photos taken with real film and digital simulations under different conditions without accounting for these contextual differences can result in unfair or inaccurate assessments of the accuracy of the simulation recipe.”

Basically, ChatGPT doesn’t have feelings or personal views. It just processes a ton of information and responds to whatever you ask it. Many people, like reporters and students, use ChatGPT and generally find it reliable. It’s also given some advice on handling film simulation recipes. So, I hope you can take something useful from this as a reader.

Credits:

This article was written by Marc R.
While I primarily speak Dutch and have some knowledge of English and a little French, ChatGPT helps ensure my writing is grammatically correct. I often mix Dutch and English in my drafts, and ChatGPT steps in to translate and correct.

Header and article picture was created with AI on Freepik.com and Gimp.

La Gallerie

The photo gallery of Open Source Photography, Olympus micro 4/3 system, Vintage Lens Photograpy, Film Simulation, PictureFX, HDR – Photographer : Marc R.