How to Prompt Nano Banana Pro for Realistic Lighting

Lighting is the biggest reason an AI image reads as "plastic" or "real." Nano Banana Pro (Gemini 3 Pro Image, released November 20, 2025) is the first generation of Google DeepMind's image stack to expose explicit lighting and camera controls in the prompt surface. Google DeepMind states the model "gives you control over the physics (lighting, camera, focus, color grading) and composition of the image" through prompt text alone (Nano Banana Pro announcement).

Those controls only work if you speak the vocabulary they were trained on. This guide is a compact reference of documented cinematography and lighting language, mapped to Nano Banana Pro's official capability statements. Nothing here is "tested by us." Everything is either Google's own phrasing, established cinematography terminology with a citation, or a documented prompt pattern shared by published guides.

For the broader model overview, see our Nano Banana Pro page and the predecessor Nano Banana reference.

Why lighting language matters in prompts

Image generators learn from billions of captioned images. Photographers and DPs describe frames with a compact technical vocabulary: "Rembrandt with a 4:1 ratio," "blue hour, overcast diffusion," "key right, fill bounce, kicker on the hair." Those words appear in stock metadata, museum captions, and cinematography tutorials throughout the training data, so they act as dense pointers to specific visual outcomes.

"Dramatic lighting on a man's face" gives the model little to anchor on. "Split lighting, single key from camera left at 90°, hard shadow across the right side of the face" maps to a repeatable pattern with its own Wikipedia page, its own SLR Lounge tutorial, and thousands of consistently-tagged reference images.

Google DeepMind's prompting guide frames it directly: "Direct the shot like a cinematographer." Their example fragments include "A low-angle shot with a shallow depth of field (f/1.8)," "Golden hour backlighting creating long shadows," and "Cinematic color grading with muted teal tones."

Claimed vs confirmed: what the model actually controls

Before we get to the vocabulary, here is what Google has officially published versus what is third-party convention:

The third-party-convention terms work because they are established photography terminology, not because Google publishes a "supported lighting modes" list. Treat them as patterns the model has seen in training, not guaranteed APIs.

Core lighting vocabulary

Natural light

Golden hour. Per Wikipedia, "the period of daytime shortly after sunrise or before sunset, during which daylight is redder and softer." Warm color temperature, low sun, long shadows.

Blue hour. PhotoPills defines blue hour as the period just before sunrise or after sunset when "residual, indirect sunlight takes on a predominantly blue shade, and there are no sharp shadows."

Overcast. Clouds diffuse the sun into flat, color-neutral light with low contrast, good for skin and product detail without dramatic falloff.

Studio portrait patterns

Three-point lighting. Classical broadcast and portrait setup (Fiveable cinematography notes): a key (main directional source), a fill (softens shadow side), and a back light (separates subject from background).

Rembrandt. Wikipedia defines it by "an illuminated triangle (also called 'Rembrandt patch') under the eye of the subject on the less illuminated side of the face." Reads painterly and somber.

Butterfly. Light directly in front, slightly above eye level, casting a small symmetrical shadow under the nose. Per SLR Lounge, the standard "glamour" or "Paramount" pattern.

Split. Hard side light at 90°, illuminating half the face. Per SLR Lounge, the most dramatic of the standard portrait patterns.

Dramatic and contrast-driven

Chiaroscuro. Italian for "light-dark." Wikipedia and No Film School define it as a low-key, high-contrast technique producing distinct areas of brightness against deep shadow, the look of Caravaggio and classic film noir. Google DeepMind explicitly lists chiaroscuro as a supported effect on the product page.

Low-key. Wikipedia describes the style as one that "accentuates shadows and high contrast," the foundation of chiaroscuro.

High-key. The inverse: low contrast, soft shadows, abundant fill. Per Adobe, high-key reads as bright, optimistic, commercial.

Atmospheric

Volumetric lighting. Per Wikipedia and StudioBinder, the visible-beam effect when light interacts with particles in the air (smoke, dust, fog). Google DeepMind explicitly lists volumetric lighting as a Nano Banana Pro effect.

God rays / crepuscular rays. A specific case of volumetric lighting: shafts of sunlight breaking through clouds, foliage, or window blinds.

Combining lighting with camera and lens

Lighting alone does not finish a frame. Focal length, aperture, and angle decide how the light wraps. Google DeepMind's announcement describes the ability to "adjust camera angles, change the focus and apply sophisticated color grading" and to vary wide, panoramic, and close-up shots while controlling depth of field. Google's own example combines all of these: "A low-angle shot with a shallow depth of field (f/1.8)."

A few documented patterns:

Portrait, shallow depth. Short telephoto plus wide aperture compresses features and isolates the subject. 85mm at f/1.8 is a long-standing 35mm portrait convention.

Cinematic wide. Low angle, wide focal length, golden hour: a signature look in feature-film stills.

Product macro, hard light. Single hard key with a long macro at narrower aperture, the high-contrast luxury product look.

Editorial beauty. Soft frontal key, slight high angle, neutral background.

Common pitfalls

Over-specification. Stacking five lighting terms ("Rembrandt + chiaroscuro + golden hour + neon noir + volumetric god rays") gives the model contradictory targets. Pick one dominant lighting concept and let supporting details (color temperature, modifier hardness, angle) refine it.

Conflicting cues. "Bright, sunny golden hour with deep chiaroscuro shadows" is a contradiction: golden hour is a natural condition, chiaroscuro is a low-key studio choice. If you want golden-hour color with low-key contrast, say so: "warm golden-hour color temperature, low-key high-contrast scene, single hard backlight."

Vocabulary collisions with subject. "Three-point lighting on a forest at dusk" applies a studio convention to a location condition. Match lighting vocabulary to context: studio terms (three-point, butterfly, split) for portraits and indoor product, natural terms (golden hour, blue hour, overcast) for outdoor, dramatic terms (chiaroscuro, low-key, volumetric) for mood pieces in either.

Forgetting modifier hardness. "Soft" vs "hard" is the most underused lever in AI prompts. The same key position produces wildly different results from a bare bulb (hard, sharp shadows), a softbox (feathered), or a beauty dish (between). Spell it out: single key from camera left, hard bare-bulb quality, sharp-edged shadows.

Skipping the mood anchor. A one-word mood ("intimate," "ominous," "clinical," "nostalgic," "heroic") gives the model a tiebreaker when multiple readings of your lighting prompt are equally valid. Google's own example uses "cinematic" the same way.

A working prompt template

Example:

Every piece traces back to a documented term. Rembrandt is a Wikipedia-defined portrait pattern, 85mm at f/2 is a standard short-telephoto portrait combination, and the structure mirrors Google's "direct the shot like a cinematographer" framing.

Frequently asked questions

Does Nano Banana Pro understand cinematography vocabulary?

Google DeepMind officially documents that the model accepts lighting and camera controls in prompts, including specific effects they name on the product page: volumetric, bokeh, chiaroscuro, day-to-night. Beyond that explicit list, established photography terms (Rembrandt, butterfly, split, golden hour, blue hour) are widely-tagged training-data vocabulary, so they map reliably even though Google does not publish a "supported lighting modes" list.

Can I use f-stop and lens-mm values in prompts?

Yes. Google's own example includes "shallow depth of field (f/1.8)," and combining lens length with aperture (85mm at f/1.8) is a convention echoed in third-party guides like Sider's Nano Banana Pro lighting guide and the WaveSpeed AI tips post. The values are visual references the model maps to, not literal optical simulations.

What is the difference between Nano Banana and Nano Banana Pro for lighting?

The Pro model (Gemini 3 Pro Image) explicitly adds the studio-style lighting and camera controls described above. The original Nano Banana model accepts the same vocabulary in prompts but does not advertise the same level of explicit per-subject focus, depth-of-field, or chiaroscuro/volumetric controls.

Where can I try Nano Banana Pro?

Inside the Gemini app, Google AI Studio, and the Gemini API. The free Gemini app tier is reported to cap at around three generations per day. For higher volume, the Nano Banana Pro studio on gptimg.co wraps the model in a browser interface with credit-based pricing.

Sources

• Nano Banana Pro: Gemini 3 Pro Image model from Google DeepMind — official announcement
• Gemini 3 Pro Image (Nano Banana Pro) — Google DeepMind product page
• Nano Banana Pro image generation in Gemini: Prompt tips — Google's own prompting guide
• Google DeepMind on X — lighting and camera controls announcement
• Rembrandt lighting — Wikipedia
• Chiaroscuro — Wikipedia
• Low-key lighting — Wikipedia
• Volumetric lighting — Wikipedia
• Golden hour (photography) — Wikipedia
• 5 Common Key Light Patterns Every Portrait Photographer Should Know — SLR Lounge
• Three-point lighting — Fiveable Advanced Cinematography notes
• What is Chiaroscuro Lighting — No Film School
• What is Volumetric Lighting — StudioBinder
• Mastering Golden Hour, Blue Hour — PhotoPills
• Low-key vs high-key lighting — Adobe

Last reviewed against source pages: 2026-04-18.