Guides

    Hard or soft: source size in lighting prompts

    Shadow edge hardness is set by apparent source size, not by adjectives. A size-and-direction vocabulary that outperforms five style words in every prompt.

    Versely Team9 min read

    "Soft lighting" is one of the most-typed phrases in generative prompting and one of the least specific. Soft is what the light does, not what the light is, and models render things rather than effects. Swap it for the thing itself, a source with a stated size, and the output changes in a way five style adjectives never manage.

    There is exactly one variable behind the hard/soft axis, it is physically simple, and it is the highest-yield lighting instruction you can write after direction.

    One variable sets the shadow edge

    Shadow edges are set by the source's apparent size — how large it looks from where the subject is standing. A large apparent source wraps light around the subject and produces a wide, gradual shadow transition. A small one produces a narrow, abrupt one.

    The word doing the work is apparent. The sun is the largest object anyone will ever be lit by and it produces knife-edged shadows on a clear day, because at that distance it occupies a tiny slice of the sky. Put cloud in front of it and the entire sky becomes the source, apparent size goes to nearly everything, and shadows all but disappear. Same sun, opposite result, one variable changed.

    That gives you a rule you can apply to any source without thinking about it:

    • Big and close → soft. Wide, gradual shadow edges. Forgiving on skin and texture.
    • Small or far → hard. Tight, defined shadow edges. Reveals texture and shape, unforgiving on faces.

    Neither is better. Hard light is what makes a product's edges read and what makes a face look carved; soft light is what makes skin look good and what a beauty campaign is built on. The failure isn't picking wrong, it's not picking at all — and "cinematic lighting" is not picking.

    The size ladder

    The most efficient way to state size is to name a fixture or a natural condition that carries it. Two words of fixture name often encode size, direction and colour temperature at once, which is why they outperform a sentence of description.

    Source Apparent size Shadow edge Prompt phrasing
    Bare bulb, candle, phone torch Tiny Very hard, high contrast lit by a single bare bulb overhead
    Direct midday sun Small Hard, dense direct midday sun, hard-edged shadows on the pavement
    Small spotlight or snoot Small Hard, contained pool a narrow spotlight picking out her face, everything else dark
    Standard window, subject a metre away Medium-large Soft, still directional north-facing window at camera left, subject close to the glass
    Large diffusion panel or softbox Large Soft, gentle falloff a large softbox just out of frame at camera left
    Light bounced off a white wall Very large Very soft, wraps light bounced off a white wall to camera right
    Overcast sky Enormous Almost no visible shadow flat overcast daylight, no visible shadow direction

    Read down that last column: every phrasing names a physical object in a place, and none of them says "soft" or "hard". That is the point. Add the quality word as belt-and-braces if you like, but the fixture carries the instruction.

    "Overcast" describes a real, common, visually distinctive lighting state that turns up in training data labelled exactly that way. "Moody" describes a feeling that several thousand different lighting states could produce.

    Distance is the second half

    Apparent size is size and distance, so distance is a free dial you already have. Moving a source closer makes it softer without changing the fixture. Moving it back makes it harder.

    Two phrasings that use this:

    A large softbox pushed in close, just out of frame at camera left —
    the shadow under her jaw fades out gradually with no defined edge.
    
    The same softbox pulled well back across the room — the shadow under
    her jaw has a defined edge and the far side of her face goes dark.
    

    Distance also controls falloff. A close source drops off steeply, so a background a couple of metres behind the subject goes dark on its own; a distant source lights subject and background nearly equally. If renders keep coming back with a background as bright as the subject, state a close source and let falloff do the work rather than asking for "a dark background".

    Direction, in the same sentence

    Size answers what the shadows look like; direction answers where they fall. They belong in one clause, and the combination is what makes a render commit to a physical setup instead of the sourceless average brightness models default to.

    The four directions worth knowing by name, with a size attached:

    • 45 degrees, front-side — the workhorse. a large source at 45 degrees to camera left, slightly above eye level. Soft version is a portrait key; hard version gives you the classic triangle of light on the shadow cheek that gets called Rembrandt lighting.
    • Side, 90 degreesa hard source directly from camera left splits the face into lit and unlit halves. Dramatic and easy to overdo.
    • Backa small hard source behind her, rimming the edge of her hair and shoulders. This is rim light, and stating that it is behind and small is the whole instruction.
    • Topa large source directly overhead gives you the flat, institutional look of a supermarket or an office, which is exactly right when that is the scene.

    The lighting prompt formula covers the full source–direction–quality structure these clauses drop into. Source size is the part of it people skip, and skipping it is why so many prompts come back looking evenly lit from nowhere in particular.

    The adjectives to cut

    One concrete lighting descriptor beats five style adjectives — and the corollary is that the five adjectives are costing you.

    Cinematic, dramatic, moody, atmospheric, beautiful lighting. None specifies a source, a size, or a position. What they mostly do is bias the grade toward lower saturation, more contrast and crushed blacks, so a prompt loaded with them returns something that looks processed rather than something that looks lit. The temptation is then to add more, which deepens the grade while leaving the light as sourceless as it was.

    Cut all five and spend the words on one clause: a large softbox at 45 degrees camera left, close to her, the rest of the room falling into shadow. That clause does more than the adjectives did, and it survives being copied verbatim into the next shot, which they never do.

    Worth knowing where lighting is an actual input rather than prose: a small number of models expose an explicit lighting-control capability — the Seedance 2.0 family carries one — and there it behaves like any other parameter across takes. Everywhere else, prompt adherence on lighting is a function of how physically you phrased it.

    Checking it on the render

    Two things to look at, and neither is overall brightness.

    The shadow edge. Find the shadow under the chin, the nose, or a product's base and look at how many pixels the transition takes from lit to unlit. Wide and gradual means the model gave you a large source. Narrow and abrupt means small. If you asked for a softbox and got a knife edge, the size instruction didn't land and the fix is to name the fixture more specifically, not to add "soft" again.

    The specular highlight. Shiny surfaces reflect the source, so the highlight is a picture of it. A large source gives a broad, dim, soft-edged highlight; a small one gives a bright pinprick. On glass, metal or skin this is the fastest read available, and on product work it is more diagnostic than the shadows — which is why product photography without a studio turns on getting it right.

    The failure worth naming: shadow edges that disagree within one frame. Soft under the chin, hard under the hand, background shadow going the other way. That is light assigned per region rather than from one place. Name one dominant source and let everything else be fill. If you need the same product under three setups without the product itself moving, that is a relight operation rather than a re-prompt, covered in relighting product shots.

    FAQ

    What actually makes light hard or soft?

    The apparent size of the source as seen from the subject. Large and close gives wide, gradual shadow edges; small or far gives narrow, defined ones. The clearest demonstration is the sun: enormous, but so distant it occupies a tiny part of the sky, so it throws hard shadows — until cloud cover turns the whole sky into the source and the shadows vanish.

    Is it better to write "softbox" or "soft lighting"?

    "Softbox", almost always. A fixture name carries size and usually implies a colour temperature too, so it is a compound instruction in one word. "Soft lighting" names the result and leaves the model to invent a setup that might produce it. Name the fixture, state where it is, and add the quality word only as belt-and-braces.

    Why does adding more lighting adjectives make renders look worse?

    Because they stack a grade rather than a setup. Cinematic, moody, dramatic and atmospheric all pull toward lower saturation and higher contrast without specifying where any light is, so the image gets darker and more processed while remaining sourceless. Replace all of them with a single physical clause: one source, one size, one direction.

    When should I deliberately choose hard light?

    When texture and shape are the subject. Hard light is what makes a product's edges read, what makes fabric weave and machined metal legible, and what carves a face into planes. It is the wrong pick for beauty and skin work, where a small source gives every pore its own shadow. The quick test: if the thing you are selling is a surface, go small and far; if it is a face in a flattering context, go large and close.