How to create realistic images using Generative AI

November 23, 2023

Generative AI models for image creation, such as Midjourney and Dall-E or Bing/Copilot, are increasingly making it possible to produce images and photographs with ever greater realism and technical and aesthetic quality in seconds. And in this creative process, chemical photographic film plays an influential role in the result of digital images produced with generative AI.

In fact, photographic film has a significant and visible influence on today's digital photography. Many digital cameras, filters such as those in Instagram, or settings in programs such as Adobe Lightroom allow us to emulate aesthetic qualities that have their origins in conventional photographic film, including its imperfections and limitations.

Over the course of a century and a half, the resulting photographic image, in technical and aesthetic terms, as well as in terms of style and character, has depended largely on the film chosen.

Factors such as the film manufacturer, its purpose and technical characteristics, the manufacturing process or the materials and chemicals used have left characteristic traits in millions of images throughout the history of photography. Traits that can be applied to images created by generative AI.

In generative AI, indicating a type of chemical film helps to simplify the process of generating realistic and more natural-looking images. It also allows you to develop a personal style in image creation, or to have more control over the technical and aesthetic result by taking advantage of the type of scene and the most recognizable characteristics of that film.

The impact of chemical film on generative AI

Generative AI uses text instructions (prompts) to create digital images. By specifying photographic film types in theprompt it is possible to influence to a greater or lesser extent the style and aesthetics of the generated images, to narrow down the reference models, and to implicitly mandate a realistic image type.

A prompt that includes the Fujifilm Velvia 400 style, for example, will generate realistic images with vivid colors and sharp contrasts without the need to indicate this in the instruction, since these are characteristic features of photographs taken with that film.

Fujifilm Velvia 400 style

However, the same prompt tacked to Kodak Tri-X style will produce a black and white image without prompting, resulting in a digital image with the high contrast, fine grain and sharpness characteristic of that film.

Kodak Tri-X film style

The generative AI can interpret the type or style of film and take it into account when generating the image, simulating its visual style and characteristics.

Simplification of image creation

Mentioning chemical film styles in the generative AI thereby saves specifying other parameters that are already implicit in the mentioned film style: colors and tones, saturation, grain, dynamic range and contrast, texture, and even the overall quality of the generated image.

Although not necessary, describing these or other parameters (such as the type of light or camera angle) does help the model to interpret the type of image and respond to the request more accurately.

Kodak Professional Ektachrome E100 style

Another example would be to indicate on the prompt Kodak Professional Ektachrome E100 style, a film especially appreciated for capturing scenes that require high precision in the reproduction of bright, vivid colors with great color accuracy and fine details, even in low light situations.

Polaroid style

Also very characteristic is the result of adding Polaroid style to the prompt, the film that already in the 1970s made it possible to have the photographs on paper within seconds of taking them without going through the development lab.

Kodak Portra 400 style

Kodak Portra 400 film is particularly appreciated for its versatility, as it can be used in a wide variety of lighting conditions, including low-light situations. Additionally, it offers an exceptionally wide tonal range, allowing for images with great detail in both highlights and shadows.

Kodachrome 64 film

Kodachrome 64, by Kodak, was a photographic film known for its color accuracy and low sensitivity (ISO 64) well suited for high light conditions. It was manufactured between 1935 and 2009.

In 1984 Steve McCurry used Kodachrome 64 film to take his iconic photo of Sharbat Gula, "the Afghan girl," which made the cover of National Geographic magazine in 1985. This image stands out for the intensity of its colors. In the context of generative AI, Kodachrome 64 would be a good choice for generating a realistic digital portrait with intense natural colors.

Other photographic parameters influencing generative AI

In addition to the choice of photographic film, there are other parameters and camera settings of conventional photography that can also influence the appearance and final result of images generated with AI:

ISO sensitivity: Mentioning a high ISO in the prompt may result in images with more noise or grain, similar to photographs taken in low light conditions or at high ISO.
Aperture: Including a wide aperture such as f/2 at the prompt is likely to generate images with a blurred background, which will emphasize the main subject. On the other hand, a small aperture, such as f/11, will result in a greater depth of field and a flatter, more focused overall image.

A wide aperture produces shallow depth of field.

Shutter speed: Specifying a high or low shutter speed (or numerically) will affect motion capture. This can result in images that freeze the image or images that instead 'capture' motion, as in a long exposure photograph.

Changing shutter speed.

White balance and contrast: These parameters also influence the appearance of the image. A warm white balance will give more yellow tones, while a cool white balance will give more blue tones. High contrast will produce darker shadows and brighter highlights, while low contrast will soften shadows and bright highlights.
Saturation and HDR: High saturation will generate more vivid and intense colors, while low saturation will result in duller colors. HDR can result in more detail in highlights and shadows.

Dynamic range and selective focus control.

In addition to these parameters, the generative AI can also interpret and take into account other details related to the photograph, such as the type of lens (wide angle, telephoto, fisheye, macro), lighting, including use of flash, selective focus, camera angle or distance in meters, among others. These adjustments provide more information to the model, resulting in greater control over the final result.

Beyond digital photography, chemical photographic film maintains its influence in the creation of digital images through generative AI. The choice of a photographic film style in generative AI prompts influences the type and appearance of the final image, while defining key technical and aesthetic aspects, providing greater control and involvement in the final outcome of the generated images.

🎞️ All images in this post are generated using AI (Dall-e + Magnific.ai) to demonstrate possible results, although the same concept can be applied to other generative AI models. OpenAI's Dall-e can also be used through Bing and Copilot from Microsoft, as well as ChatGPT.