A Creative Diary: Documenting My AI Art Projects - Portrait of A Chef

   

      Being a culinary enthusiast, I aimed to create a portrait of a chef in a sophisticated kitchen through photography. I used aiTransformer Super Stylizer, which is based on Stable Diffusion, a powerful deep learning, text-to-image model. While Stable Diffusion is generally effective, it occasionally produces distorted or surreal images of human faces or limbs. The distortion happens because Stable diffusion is a generative model that produces realistic images by gradually adding noise to a starting image over a series of steps. This process can produce images with realistic textures and details, but it can also introduce distortions to the image.

      For a new project,  I usually begin with a straightforward representation of the image I envision. This simplistic approach aids in establishing a solid foundation to build upon. If the outcome doesn't meet my expectations, I then introduce more detailed descriptions to the prompt. Usually, with a few attempts—sometimes sooner, sometimes later—I am able to achieve the desired outcome.

      First Prompt: a happy chef in a high end kitchen making dough, a high contrast portrait,  there is a painting of flowers on the wall behind him.

 

        The top-left photo features the desired background painting that I'm looking for. Although the chef is not making dough, I can accept that as long as the high-end kitchen setting is visible. The top-right photo displays an unattractive background flower painting, and the chef's face appears distorted. As for the rest two photos, their backgrounds have no flower paintings.

         With this prompt, I attempted multiple sets but found none of them satisfactory.  I decided to move on to another prompt. In Stable Diffusion, the order of the prompts could make a difference as the first noun tends to receive more focus unless additional weight is assigned. For this new attempt, I placed "high-end kitchen" before "chef"  to prioritize the kitchen setting.

 Second Prompt: in a high end kitchen,  a very happy  a chef making dough,  a painting of flowers on the wall behind him, a high contrast portrait

 

       

       All of the pictures feature a kitchen setting. The bottom-right one has potential if the facial distortion can be removed. I downloaded and saved it for future, see if  I can restore it. I also experimented with another 20 pictures, resulting in one more acceptable outcome.

     However, I am still aiming for a result that surpasses mere acceptability. To enhance the prompt, I included specific details relevant to photography, such as "dramatic lighting", "100mm", "8k", "award winning photography", etc, which can be found in Prompt Builder

  Third Prompt: a profile portrait of  a very happy a chef making dough, in a high end kitchen, a painting of flowers on the wall behind him, a high contrast portrait,  elegance. high resolution,  dramatic lighting, 100mm, 8k, award winning photography

 

    

       The upper-left photo is OK, but despite several attempts, I couldn't get a better result. To refine the prompt, I introduced the specification of "a profile portrait" and switched the lens to 80mm. Additionally, I included additional details such as "cuisine magazine photos" and "DSLR" to see if they can improve the outcome.

Fourth Prompt: a profile portrait, in a high end kitchen,  a very happy  a chef making dough,  a painting of flowers on the wall behind him, a high contrast portrait,  cuisine magazine photos,  dlsr, elegance, high resolution,  dramatic lighting, 80mm portrait photography, 8k, award winning photography,

 

 

      I like the top-right photo for its excellent composition, lighting, color, focus, and storytelling. There are still a few minor imperfections if you look carefully, such as the mismatched buttons—four being white and six being black.  I think those can be overlooked as they are not significant. Generating high-quality images using AI remains a challenging task, and even the most advanced models can produce imperfect results. It's essential to remember that AI-generated images are not real and should be taken with a grain of salt.