If your AI image prompts end with "photorealistic, ultra detailed, 8K," you are generating sharp renders, not photos. A new breakdown from DEV.to contributor Yash Patidar argues that resolution keywords are a dead end for realism. Instead, the author suggests that a photo is defined by a series of technical decisions: lens choice, light source, camera position, and intentional imperfections. This shift in vocabulary is what separates the plastic, uncanny valley look from genuine photographic texture.

The Lens and Aperture Decision

The article details how specific focal lengths alter the narrative of an image. A 35mm lens places the viewer "in the scene" for street or travel photography, while an 85mm lens is the standard for flattering portraits with soft backgrounds. The author cites examples from the Promptclay collection, noting that a "Moody Editorial Interior Portrait" prompt explicitly starts with "Portrait shot on 85mm f/1.4." Aperture settings are equally critical; low f-numbers (f/1.4 to f/2.8) create shallow depth of field, whereas street photography often requires f/8 to keep the scene sharp. One decisive-moment prompt even uses "shallow depth of field" in its negative prompt to force a specific look.

Lighting and Positioning Are the Real 'Secret Sauce'

Vague terms like "good lighting" are useless to image models. The source material emphasizes naming the source and direction, such as "window light from the left" or "golden hour sun behind them." A standout example provided is a prompt that specifies: "Warm amber low-key interior light from background pendant lamp, subject's face lit partially by cool window light, two-source mood, chiaroscuro balance." This level of specificity creates a "two-source" environment that mimics real-world interiors. Camera position is another key factor; for pets and children, the author recommends "shot at the dog's eye level" to avoid the unnatural, downward-looking perspective common in AI generations.

Embracing Flaws and Negative Prompts

To break the "perfect" AI look, the article advocates for adding technical flaws. Specific keywords like "ISO 400 film grain in shadow areas," "slight halation on highlights," and "visible pores" add the texture of real film or digital sensors. The author suggests an "avoid list" to steer the model away from common AI artifacts, including "HDR," "plastic skin," and "perfect symmetry." A fill-in-the-blank template is provided, combining these elements: "Real photograph of [subject], Shot on a [lens] at [aperture], [light source], [texture detail], Avoid: smooth skin, HDR, studio lighting."

Key Takeaways

  • Specificity beats resolution: Keywords like "85mm f/1.4" are more effective than "4K" or "8K."
  • Light is directional: Always specify the source (e.g., "window light") and time of day.
  • Imperfection is realism: Adding film grain, motion blur, or imperfect framing reduces the "render" look.
  • Use negative prompts: Explicitly banning "plastic skin" and "HDR" helps maintain photographic integrity.

The Bottom Line

Resolution is a technical metric, not an artistic one. Stop asking AI for pixels and start asking for perspective, lighting, and imperfection.