Your Black-and-White Prompt Should Not Mention Color
Quick Tip: Copy the prompt with the button above — paste it into Midjourney as-is (--ar and --style raw included). On DALL-E, Flux, Stable Diffusion, Ideogram, Leonardo AI or Firefly, keep the descriptive text and drop the -- flags. Try the same scene at two different aspect ratios: composition and light falloff change more than you'd expect.
Why B&W Film Noir Prompts Fail When You Add Color Words
Conventional advice says to make a rainy window cat prompt more cinematic by adding neon glow, amber eyes, and warm 3200K color. That advice is wrong for black-and-white film noir. Those words are color instructions, and they fight the grayscale instruction. The model receives conflicting signals: monochrome film stock on one side, chromatic atmosphere on the other. The result is often muddy, washed-out, or secretly colorized. If you want a true B&W film noir aesthetic, stop smuggling color into the prompt and start translating atmosphere into luminance.
The better move is to keep the mood but remove the color. 'Neon sign glow at 3200K' becomes 'bright out-of-focus light sources blooming through fog.' 'Luminous amber eyes' becomes 'eyes catching a hard specular highlight against deep shadow.' The visual function survives in grayscale: a soft orb of light, a sharp catchlight, rain streaks on glass. What disappears is the contradiction. A cinematic rainy window cat prompt should describe contrast, texture, and shape, not temperature. Once I translated those terms, the B&W film noir look finally held together.
The Glass Is the Real Subject: Layered Depth Construction
What makes this image work is not the cat — it's the rain-streaked window acting as a physical intermediary between viewer and subject. The prompt constructs three distinct depth planes, and each one is assigned a sharpness value. The foreground glass carries partially focused droplets and condensation trails. The midground holds the cat at the razor-thin focal plane, specifically the eyes. The background dissolves entirely into bokeh: streetlamp glow, the fedora silhouette, wet reflections.
This layering works because the prompt uses spatial prepositions as technical instructions. Phrases like "between viewer and subject" and "depth-of-field falloff" are not decorative — they tell the model how to allocate focus across the frame. Without that spatial ordering, AI models tend to collapse everything onto a single plane: the droplets, the cat, and the background all fight for sharpness, and the result reads as a flat, busy texture rather than a photograph taken through something.
Compare this with the common alternative: "cat behind rainy window." That phrasing gives the model a subject and a condition, but no optical logic. The model must guess which layer is sharp, and it frequently guesses wrong — sharpening the water droplets and softening the cat's face, which inverts the emotional hierarchy of the image. The eyes carry the narrative; the glass carries the texture. The prompt must say so explicitly.
Film Stock as a Complete Grading System
"Pushed Ilford HP5 Plus 1600 ISO" is doing more work in this prompt than any other single phrase. Film stocks are not aesthetic filters — they are complete imaging systems with characteristic grain structures, contrast curves, and highlight behavior. Pushing HP5 (exposing it at a higher ISO than its rating and compensating in development) is historically known for three things: increased contrast, prominent clumping grain, and a distinctive milky lift in the shadows. By naming the stock and the process, the prompt imports all three behaviors at once.
The pairing with "Leica M3, 50mm Summilux f/1.4 wide open" reinforces the same system from the optics side. A fast 50mm shot wide open produces exactly the razor-thin depth of field and gentle optical softness the noir look demands — and it gives the model a reason for the background to dissolve rather than simply blur. "Shot on" language works because it bundles physics: focal length, aperture, and rendering character arrive together instead of as disconnected requests.
The generic alternative — "35mm film grain, shallow depth of field" — describes outcomes without causes. Models handle outcome-only descriptions by applying a superficial approximation: uniform noise overlaid on the image, Gaussian-style blur on the background. The difference is visible in the highlights. Real pushed film blooms; fake grain just sits on top of the pixels. For more on how material specificity drives results, the same principle appears in this ultra-detailed cyberpunk singer prompt and in this deconstructed portrait sketch breakdown, where medium and process terms carry the entire style.
The Silhouette: Narrative Depth Without Competing Focus
The fedora-wearing figure under the streetlamp is the most delicate element in the prompt. Handled wrong, it hijacks the composition — a second subject pulling the eye away from the cat. Handled right, it transforms the image from a pet portrait into a scene from a story: someone is out there, and the cat is watching, and we are watching the cat watch.
The mechanism that keeps it subordinate is the phrase "aerial perspective" combined with "heavy atmospheric fog." Aerial perspective is a classical painting term describing how distant objects lose contrast and detail as atmosphere accumulates between them and the viewer. Instructing the model to apply it to the silhouette guarantees the figure renders as a tonal shape — a dark mass against glowing fog — rather than a detailed person. The blur from the wide-open aperture does the rest.
This is a transferable technique for any cinematic composition: background story elements should be described with their atmospheric treatment attached, never as standalone subjects. If you're building moody character-driven frames, the same subordination logic appears in this street Joker portrait prompt, where environmental chaos stays behind the focal plane. For experimenting with variations of this noir setup, Midjourney handles the film-stock language particularly well, while Leonardo AI offers useful fine-tuned models for monochrome photography styles.
Where to Take This Next
The natural next experiment is to swap the intermediary layer while keeping the structure. Replace the rainy window with a fogged café glass, a scratched train window, or a wrought-iron fence — the three-plane system (textured foreground, sharp subject, dissolved narrative background) remains identical, but each intermediary brings its own emotional register. You can also push the silhouette further into abstraction: remove the fedora and trench coat cues and describe only "a figure-shaped absence of light in the fog," which produces an even more ambiguous, unsettling read. Each iteration teaches you how much narrative a background can carry before it starts competing with the eyes.
Key Principle: In black-and-white prompts, delete every hue and color-temperature word and replace it with tonal language — "blown highlights," "crushed blacks," "specular bloom" — so the model commits to grayscale instead of hedging into muddy desaturated color.