Prompt development

Developing the right prompt is key to driving AI engines towards what output you want.

The first prompt I tried was: Hyper realistic photograph of a dog diving underwater for a ball, teeth bared it looks gritty and menacing but actually the dog is just playing

I developed the prompt further: A powerful underwater photograph of a dog diving through dark water, teeth bared, eyes wide, bubbles exploding around its head. The lighting is dramatic and moody, with sharp contrasts between light and shadow, like a cinematic still from a psychological thriller. The water is turbulent, the energy raw and primal, capturing the fine line between play and aggression. Ultra-detailed textures — fur, bubbles, motion blur — evoke intensity and chaos.

And got these results, which are closer to what I am looking for.

I further refined my previous prompt by adding the below phrases, which again yielded a step in the right direction:

• Shot with a high-speed underwater camera
• Moody chiaroscuro lighting
• Film noir underwater aesthetic
• Distorted refractions and turbulent motion
• Gritty texture, 35mm realism
• Emotionally charged, primal energy, cinematic tone

The above images are all by Google Imagen 4.0 Fast which I have worked with a bit already. Other engines will generate images of a different quality. Below is the same prompt in Qwen Image SD.

And this is the same prompt again in Flux Kontext Dev.

I returned to Google imagen 4.0 Fast, slightly tweaking my existing the prompt, adding: side view and dog at a 45 degree angle. Some of these are great and only require a little further development before I experiment with some test prints.