If you've spent any time on Pinterest or Instagram lately, you've probably seen those oddly satisfying videos: a pair of hands sculpting a tiny polymer clay garden inside the cross-section of a fruit, layer by layer, until it turns into this impossibly detailed little world — complete with a pond, a fence, a blossoming tree, and a hand-lettered sign at the end. They're mesmerizing. And now that AI video tools like Sora, Veo, and Kling can generate this kind of footage from a single prompt, I wanted to figure out how to make one reusable prompt instead of writing a brand-new one every single time I wanted a new variation.
That turned into a bit of a rabbit hole. Here's what I landed on, why it works, and how you can use it yourself.
The Problem With "One-Off" Prompts
My first attempts looked like this: describe the diorama, describe the fruit, describe the tree, hit generate. It worked fine — but every time I wanted a different diorama (say, a lime instead of an orange, or a lighthouse instead of a tree), I had to rewrite the whole thing from scratch. That's fine for a single post, but it's a nightmare if you're trying to build a content pipeline — a whole series of these videos for a blog, a TikTok account, or a product catalog.
So the real goal became: one master prompt that can produce a different, complete diorama every time it runs, without me manually swapping words in and out.
The Solution: A Self-Randomizing Prompt
Instead of hardcoding the fruit, the centerpiece, the color palette, and so on, I wrote the prompt so the AI itself is instructed to pick a fresh random combination of elements every time it generates. It's a small shift, but it changes everything — the prompt text never changes, yet the output does.
Here's the actual prompt I use:
| Prompt |
|
A macro timelapse video showing a miniature polymer clay diorama being handcrafted from start to finish inside the cross-section of a fruit. RANDOMIZE THESE ELEMENTS EACH GENERATION (pick a fresh random combination every time, do not repeat the same combination twice): - Fruit base: randomly choose one of — halved orange, halved lime, halved pomegranate, halved dragon fruit, halved grapefruit, halved kiwi, halved watermelon, halved blood orange. - Centerpiece structure: randomly choose one of — blossoming tree, windmill, tiny cottage, lighthouse, treehouse, garden gazebo, stone bridge with a tree, small bonsai pine. - Water feature: randomly choose one of — curved pond, winding stream, small waterfall, still lake. - Path type: randomly choose one of — pebble stepping-stones, gravel trail, wooden plank bridge, mossy stone path. - Border/fence style: randomly choose one of — rustic wood picket fence, bamboo fence, low stone wall, woven twig fence. - Flower/foliage color palette: randomly choose one of — red/white/yellow, pink/purple/lavender, blue/white/violet, orange/yellow/cream. - Scene name for the plaque: generate a short cozy 2-3 word garden/nature name (e.g. in the style of "Citrus Grove Garden", "Moonlit Pond", "Whispering Woods") matching the fruit and centerpiece chosen. STYLE: Soft studio lighting, plain light-grey background, extreme macro close-up, shallow depth of field, hyper-detailed handmade clay texture (visible fingerprints, tool marks, slight sheen), pastel-vivid color palette, cozy stop-motion-like craft aesthetic. BUILD SEQUENCE (show each stage forming progressively, clay being shaped, pressed, and layered by an unseen sculptor's hands/tools): 1. Base shaping — the chosen fruit half is sculpted first, showing its rind texture, pith, and segmented flesh detail forming. 2. Ground layer — a thin layer of moss-green clay grass/soil is pressed flat across the top of the fruit base. 3. Centerpiece — the chosen structure is built piece by piece from the base up, with fine accent details (flowers, tiles, vines, or leaves) added one by one with tweezers. 4. Landscape details — the chosen water feature is poured in glossy blue-tinted resin, the chosen path is laid stone by stone, small rocks are placed, and the chosen fence/border is pressed around the edge. 5. Flora accents — clusters of miniature flowers in the chosen color palette, grass tufts, small bushes, and matching fruit/foliage details are placed individually around the scene. 6. Final signage — a small clay plaque is shaped and the generated scene name is carefully inscribed onto it in a rustic font, then placed at the front. 7. Final reveal — the completed piece is lifted and held in an open palm, rotating slowly to showcase full detail under soft light. CAMERA: Smooth macro dolly-in shots transitioning between build stages, occasional top-down shots, ending on a slow rotating hand-held reveal shot. PACING: Fast-motion timelapse feel with subtle motion blur between steps. NEGATIVE: no plastic-toy look, no low-detail blur, no unrealistic proportions, no extra fingers, no text glitches, no watermark. |
| Copy Prompt |
How to Actually Use It
You don't need any special software or plugin for this — just a text video generator that accepts detailed prompts. Here's the process:
- Copy the whole prompt exactly as written above. Don't edit it — the randomization instructions are what make it reusable.
- Paste it into your video AI tool of choice. This works best with chat-based, instruction-following video models like Sora 2, Google Veo 3, or Kling 1.6+, since they actually parse and act on the "randomly choose" instructions rather than just pattern-matching the words.
- Set your duration — 15 to 30 seconds tends to give the best sense of a real timelapse without dragging.
- Generate. The tool will pick its own fruit, centerpiece, water feature, path, fence style, color palette, and even invent a matching name for the little plaque at the end.
- Run it again for something new. Since the fruit, structure, and colors are chosen fresh each time, you'll get a different diorama on the next generation — no editing required.
- If your results start repeating, add one line at the very end of the prompt: "Vary randomly from previous generations." This nudges tools that default to similar outputs into producing something visibly different.
One honest caveat: this self-randomizing approach works best on newer, instruction-following video models. Older or more rigid diffusion-based tools sometimes ignore embedded instructions like "randomly choose" and just default to the first option listed. If you're using one of those, you're better off manually swapping in different fruits and objects yourself rather than relying on the AI to randomize for you.
Why This Actually Matters
It might seem like a small trick — just phrasing a prompt differently — but the impact on workflow is bigger than it looks, especially if you're publishing content regularly.
It saves real time. Instead of writing and rewriting a full prompt every time you want a new video, you write it once and reuse it indefinitely. For a blog, a social account, or a small shop selling clay-inspired products, that's the difference between spending twenty minutes per video and twenty seconds.
It keeps your content visually consistent but never repetitive. The style, lighting, pacing, and build sequence stay locked in every single time — so your whole series feels like one cohesive collection — while the fruit, structure, and color palette shift with each generation, so nothing looks copy-pasted.
It scales. If you ever want to batch-generate ten or twenty of these for a portfolio, a product mockup set, or a themed video series, you can run the exact same prompt repeatedly and get a naturally varied set of outputs, rather than manually building twenty separate prompts by hand.
It's a small example of a bigger idea. Once you start writing prompts that instruct the AI to make its own creative decisions within guardrails you set — rather than you deciding every detail upfront — you unlock a much more efficient way of working with these tools. This same structure (fixed style and sequence, randomized content variables) can be adapted for all kinds of other recurring content: miniature room dioramas, seasonal desserts, product unboxings, you name it.
At the end of the day, the appeal of these clay diorama videos is that they feel handmade and personal — a little bit different every time, just like a real craftsperson's work would be. A prompt that mimics that same natural variation, instead of spitting out the same result on repeat, just feels truer to what makes the format charming in the first place.