Most seedance glass prompts fail because they ask for a material rather than a behavior. "A glass object" or "refraction effect" tells the model what the surface is made of, not what light does when it meets that surface — how a transparent wall shows what's behind it and reflects what's in front of it at the same time, how a solid crystal bends a straight ray of light into a curved or split path, how a thin pane isolates a single droplet in sharp focus while blurring everything around it. Glass is a light-routing problem the prompt has to solve explicitly, scene by scene, or the model defaults to a flat, slightly shiny surface that reads as plastic.
The five prompts below solve that problem five different ways. One scales a glass structure to house size and carries both transparency and reflection across seven camera setups. One abandons narrative entirely and lets a single sentence of pure light description generate a studio sculpture. One turns three colored glass bottles into a timed, match-cut commercial device. One treats refraction itself — light splitting into a spectrum — as the only subject in the frame. And one pushes refraction to environmental scale, bending light through an entire cavern of ice. Together they show "glass" is a family of distinct optical behaviors, not one visual style.
1. The capsule glass house — transparency and reflection as a single surface's two readings
See the full prompt on Shoty →
"a dreamy yellow capsule-shaped glass house in a lush rain-soaked forest garden, warm orange interior glow, mossy ground, pink flowers, reflective pond, soft mist, cinematic depth of field"
Why this works: This prompt builds its entire seven-shot structure around a single fact about glass: the same pane can read as either a transparent window or a reflective mirror, depending on what's behind the camera and what's behind the glass. Scene 1's wide aerial glide reveals the capsule glowing against the dark forest — here the glass is doing reflection work, catching the warm interior light and throwing it outward so the structure reads as a lantern. Scene 3's side dolly reverses that function: "through the glass walls, revealing the subject seated in the lounge chair" treats the same material as a transparent portal, because the camera is now positioned to see past the glass rather than off it. The prompt never explains this duality in words — it only sequences camera positions so each shot activates a different property of the same surface.
The macro beat is the prompt's most precise instruction: "Close-up of the subject's hand opening the glass door, raindrops sliding down the transparent panel in sharp focus." One sentence encodes three optical facts — the glass is transparent enough to see the hand through it, wet enough that droplets are sliding (gravity plus surface tension, not static texture), and the focal plane is locked to the glass itself, not the hand or background. Naming "in sharp focus" on the panel is what prevents the model from defaulting to a conventional close-up where the glass blurs into bokeh.
The reflective pond in the wide shots does a second job beyond scenery: it doubles the capsule's light output without a second light source in the prompt, since a glowing structure reflected in still water reads as twice as luminous as the structure alone.
Takeaway: When a scene calls for glass across multiple shots, don't describe the glass once — choose, shot by shot, whether that camera position activates its transparent function (seeing through it) or its reflective function (seeing light bounce off it). For macro glass detail, name the focal plane explicitly on the glass surface itself so droplets and condensation stay sharp while everything behind blurs.
2. The suspended glass sculpture — a one-sentence prompt built entirely on light behavior
See the full prompt on Shoty →
"glass sculpture, rain-lit studio, suspended liquid ribbons, slow camera drift, stable center."
Why this works: At five words of actual content, this is the shortest prompt in this set, and it works because every word is doing optical or physical work rather than mood work. "Rain-lit studio" is a specific lighting condition — diffuse, cool, slightly broken by water on a window or skylight — rather than a generic "moody lighting" instruction, and it's exactly the kind of soft, directionless light that makes glass legible: hard single-source light on glass tends to produce one blinding hotspot and total darkness elsewhere, while rain-diffused light lets the model render the sculpture's internal refractions and surface highlights as a gradient rather than a binary.
"Suspended liquid ribbons" is the prompt's load-bearing phrase. It doesn't just describe a shape — "liquid" tells the model the glass is being rendered mid-pour or mid-motion rather than as a finished static object, which is the difference between a glass sculpture that looks cast and cooled versus one that looks caught in the act of forming. "Suspended" adds the constraint that this liquid motion is frozen or slowed rather than falling, which is physically what a glass sculpture of flowing shapes actually is — liquid glass behavior preserved in a solid state. The two words together describe the sculpture's entire conceptual premise in a way that a longer descriptive paragraph about "elegant flowing glass art" would not.
"Slow camera drift, stable center" is a deliberately restrained camera instruction for a subject this visually dense. A sculpture built from suspended ribbons already carries huge internal complexity — refracted light through multiple intersecting glass forms — so a complex camera move on top of it would fight the sculpture for attention. Keeping the subject centered and the motion minimal lets the refraction pattern be the only thing that changes from frame to frame.
Takeaway: For glass subjects where the internal light behavior is the entire point, a short prompt with precisely chosen physical nouns ("liquid," "suspended," "rain-lit") can out-perform a long descriptive one, because every word maps directly to a rendering decision. Pair visually dense glass subjects with the simplest possible camera instruction — a slow, stable, centered drift — so the refraction pattern itself supplies all the visual change.
3. The tri-color bottle montage — glass as a color-coded commercial device
See the full prompt on Shoty →
"Fast close-up montage of three transparent glass bottles in darkness: electric blue, citrus yellow, berry red. A hand reaches for the electric blue bottle. Cold condensation and visible bubbles."
Why this works: This prompt uses glass's transparency as a delivery mechanism for color rather than treating color as a separate instruction. Because the bottles are "transparent," the blue, yellow, and red aren't painted onto an opaque surface — they're the color of the liquid inside, seen through the glass, which means the color reads as a product attribute (this drink is blue) rather than a container attribute (this bottle happens to be blue). That distinction matters for a commercial: a colored liquid implies flavor and ingredients, while a colored bottle implies packaging design. The prompt gets the former by specifying transparent glass as the vessel.
The fourteen-beat timestamped structure assigns each bottle its own three-part arc — reach, "psssht-pop" sound, bubble burst — and runs all three arcs in sequence before the split-screen beat recombines them. "Cold condensation and visible bubbles" on the first bottle is a glass-specific physics cue: condensation only forms on a cold glass surface in contact with warmer air, so naming it tells the model the bottle has just come from refrigeration, which licenses the "psssht-pop" carbonation release that follows. Glass is the only container material where this condensation detail reads clearly on camera — it beads visibly on a hard transparent surface in a way it wouldn't on an opaque can.
The split-screen beat — "blue, yellow and red bubbles move in rhythm" — is the payoff of treating each bottle as its own closed color system for the first ten seconds. Because the three colors were never mixed or shown together until this moment, the recombination reads as a reveal rather than a repeat.
Takeaway: When a product shot needs color to read as the product itself rather than its packaging, put the color inside a transparent glass vessel rather than on an opaque surface. Use condensation as a glass-specific cold cue that justifies the carbonation or pour that follows, and hold multiple colored glass subjects apart before one combining beat so the recombination reads as a reveal.
4. The rainbow crystal tunnel — dispersion as the entire visual subject
See the full prompt on Shoty →
"Rainbow spiral tunnel made of tiny crystals. 5 seconds. Loops perfectly. Pure generative motion."
Why this works: Every other prompt in this set uses refraction as a property of some other subject — a house, a sculpture, a bottle. This one removes the subject entirely and makes dispersion itself the only thing on screen. "Rainbow" directly names what happens when light refracts through a prism or crystal and splits into its component wavelengths; naming the rainbow rather than describing "colorful crystals" tells the model the color is a light-splitting phenomenon, not a paint job, producing a smooth spectral gradient rather than discrete colored facets.
"Spiral tunnel made of tiny crystals" scales the dispersion effect by multiplying the number of refracting surfaces rather than making any single crystal larger. A tunnel of many tiny crystals gives light many more chances to bend and split as the camera travels through it, compounding the rainbow effect along the spiral instead of concentrating it at one object's edges — complexity through repetition of a small unit rather than detail on one large one.
"Loops perfectly" and "pure generative motion" are structural instructions as much as visual ones. A perfect loop requires the spiral's rotation to complete a clean cycle with no visible seam, which constrains the motion to a mathematically regular pattern — the kind real light-splitting follows, rather than organic, irregular movement. Calling the motion "generative" rather than narrative tells the model that no story beat needs to interrupt the pattern; the five seconds exist purely to show the phenomenon sustaining itself.
Takeaway: When refraction or dispersion is the subject rather than a property of some other object, name the optical phenomenon directly ("rainbow," "dispersion," "spectral split") rather than describing colorful materials, and scale the effect by multiplying small refracting units rather than enlarging one. For a loopable generative pattern, specify the loop and the absence of narrative explicitly — it tells the model to prioritize mathematical regularity in the motion over story logic.
5. The FPV ice cavern — refraction at environmental scale
See the full prompt on Shoty →
"FPV shot: The camera rockets at high speed through a massive ancient ice cavern at twilight, surrounded by towering crystalline walls and frozen waterfalls."
Why this works: Where the rainbow tunnel compresses refraction into a tiny, repeated unit, this prompt scales the same optical family up to architectural size and fuses it with speed. "Towering crystalline walls and frozen waterfalls" establishes ice as a refracting material at a building-sized scale — not a single prism but an entire environment made of the substance that bends light — and the FPV camera's high speed through that environment means the refraction pattern is changing constantly as the viewing angle sweeps across thousands of irregular ice surfaces per second. Scale and velocity compound each other here: a slow camera through a static ice cavern would let individual refractions resolve into distinct sparkles, while a rocketing FPV pass blurs those thousands of micro-refractions into the "shimmering ice particles" and "sparkling refractions" the prompt names directly.
"Ethereal blue liquid light… colliding in explosive crystalline splashes" introduces a light source distinct from the ice itself — a glowing fluid that interacts with the crystal structure rather than sitting inside it. This is the prompt's most ambitious refraction instruction: light that is simultaneously the illuminant (it glows) and the thing being refracted (it "spirals and condenses… forming the glowing 3D word"), passing through the crystalline ice walls that scatter and re-bend it. Naming "sparkling refractions" on the frost patterns ties the environmental ice and the magical light back into one optical system, rather than treating the ice as scenery and the light as an unrelated overlay.
The camera's aggressive weaving "between massive ice pillars" gives the effect spatial variety a straight-line pass wouldn't: each pillar is a distinct crystalline mass with its own fracture pattern, so changing the camera's lateral position relative to each one produces a different refraction signature as it passes.
Takeaway: To scale refraction from an object to an environment, name the architecture in refracting-material terms (crystalline walls, ice pillars) and pair it with camera speed — fast movement through many irregular refracting surfaces compounds into shimmer and sparkle that a slow pass would resolve into separate, duller highlights. When a glowing effect needs to interact with a refracting environment, specify that the light is both the source and the thing being bent, and let the camera weave past multiple refracting masses so each contributes a distinct angle instead of repeating one.
What these five glass & refraction prompts have in common
- Glass is at least two optical functions, not one material. The same surface reads as transparent or reflective depending on camera position and what's behind each side of the pane — sequence your shots to choose which function is active, rather than describing "glass" once for an entire scene.
- Name the specific optical word, not the material. "Rainbow," "dispersion," and "refraction" tell the model what light is doing; "crystal" or "glass" alone only tells it what something is made of.
- Condensation, droplets, and liquid motion are glass-specific physics cues that read clearly on a hard transparent surface — use them to justify temperature, freshness, or a just-poured state without a separate instruction.
- Scale refraction by multiplying small units or enlarging the environment, not just the one object. A tunnel of tiny crystals and a cavern of ice pillars both compound a simple optical effect into something that reads as complex.
- Pair visually dense refraction subjects with simple, stable camera moves, and reserve camera speed or aggressive weaving for scenes where more angle changes are the explicit goal — matching camera complexity to subject complexity keeps the refraction pattern legible instead of chaotic.
- A glowing light source can be both the illuminant and the thing being refracted — naming that dual role lets a magical or VFX light interact with a crystalline environment as one optical system instead of two unrelated layers.
For adjacent techniques, see 5 Seedance VFX & Special Effects Prompts for how light sources trigger and scale effects, and 5 Seedance Jewelry Prompts for macro reveal lighting on small reflective objects. The 5 Seedance Product Video Prompts post covers commercial shot structure, and How to Write Seedance 2 Prompts covers the general principles behind all five techniques above.
