Orelon logoOrelon
料金

Claymation Style AI Video: A Practical Workflow Guide

2026年11月7日 · Orelon Team 著

AI動画テンプレートを見る

着想のためにコミュニティ作品をいくつか閲覧し、任意のテンプレートを開いて Orelon で作成を続けましょう。

Learn how to make convincing claymation-style AI video: prompt structure, frame-rate tricks, lighting, clay texture, motion control, and a full workflow.

Claymation is one of the most deceptive styles in animation. It looks handmade, warm, and slightly clumsy, and it is genuinely expensive to produce. A real stop-motion shot at twelve frames per second means twenty-four separate setups for two seconds of screen time, each one nudged by hand, each one lit by the same tiny lamp. That is precisely why AI video tools have become interesting here: the style is costly in labor and cheap in imagination, which is the ideal gap for a generator to close.

This guide is a practical workflow for making claymation-style AI video that holds up on screen. It covers how to prompt for plasticine instead of plastic, how to fake the cadence of hand-moved frames, how to light a world that is supposed to be six inches tall, and where the illusion usually breaks. It assumes you can already generate a clip. Everything below is about directing one.

Why claymation is an unusual test for AI video

Most AI video models are trained toward realism, anime, or glossy commercial polish. Claymation asks for something almost opposite: visible material evidence. Fingerprints in the surface. A seam where an arm was reattached. Dust on the set. A background that is obviously cardboard painted to look like a forest, lit by one desk lamp and a sheet of baking parchment.

That combination is hard because it is internally contradictory. The model must render a world that is clearly fake and make it feel physically real at the same time. It has to keep imperfections consistent from frame to frame, which is where most attempts fall apart. A single prompt can produce a beautiful clay-looking still. A convincing sequence needs that clay to behave like clay for five seconds straight.

There are four things the style depends on, and each one is a separate problem to solve:

  • Material honesty. Matte surfaces, subtle subsurface warmth, tool marks, and a slight tackiness where the clay meets itself.
  • Limited cadence. Stop motion does not move smoothly. It steps, and the steps are visible.
  • Miniature scale. Shallow depth of field, macro perspective, tiny sets, and lighting that behaves like a small soft source very close to the subject.
  • Consistent imperfection. The same fingerprint, the same chipped paint, the same armature wobble across every frame.

If you only solve one or two of these, you get a clip that reads as "CG with a clay filter" rather than claymation. The workflow below attacks all four, in order.

What AI gets right, and what it still fumbles

Surface honesty is the easiest win

Texture is where current models are strongest. Ask for matte plasticine and you will usually get something close: soft highlights, a slightly powdery finish, visible thumb depressions if you describe them. The trick is specificity. "Clay" is too vague and often becomes glossy ceramic or toy plastic. "Matte polymer clay with visible thumbprints and a faintly waxy sheen" lands much closer.

Cadence is the hardest win

Models interpolate. They like smooth, eased motion because that is what most training footage looks like. Stop motion does the opposite: it holds a pose, jumps, holds again. You can request a lower frame rate in the prompt, but the reliable fix is in the edit. Generate smoother motion, then cadence it yourself by removing frames or applying a stepped time effect at 12fps or 10fps. More on that below.

Scale needs to be engineered

Nothing ruins a claymation shot faster than a character that looks twelve feet tall. Miniature scale comes from three cues: a low, close camera angle; very shallow depth of field; and light that falls off quickly, as it does from a small source inches away. Mention macro lens, tabletop set, and shallow focus explicitly.

Prompt architecture for claymation-style video

The five-slot prompt

Instead of writing a paragraph and hoping, structure the prompt into five slots. This makes it easy to iterate on one variable at a time, which is the difference between guessing and directing.

  1. Subject and scale. Who or what, how big it is supposed to be, and how the camera relates to it. "A hedgehog detective, six inches tall, on a tabletop set, shot with a 60mm macro lens."
  2. Material specification. The exact clay behavior you want. Matte, slightly translucent at thin edges, fingerprint texture, armature seams at the joints, painted eyes with visible brush strokes.
  3. Rig and artifact language. Words that imply hand animation: armature visible at the wrist, subtle repositioning marks, slight surface dust, tiny hairs from the set stuck to the clay.
  4. Lighting. Small soft source, close to subject, warm key from camera left, practical fill, visible falloff into darkness, no ambient bounce.
  5. Cadence and camera. "Stop-motion cadence, twelve frames per second, locked-off camera with a barely perceptible drift."

Three worked examples

A quiet character shot:

Tabletop stop-motion shot of a small clay fox sitting on a painted cardboard hill. Matte plasticine fur with visible thumbprints, seams at the shoulder joints, one eye slightly off-center. Warm key light from a small softbox inches away, fast falloff, deep shadow behind. Macro lens, shallow depth of field, dust motes in the air. Twelve frames per second, locked camera, slight frame-to-frame jitter.

A product-adjacent shot:

Stop-motion claymation advertisement: a matte clay coffee cup on a miniature wooden table, steam made of stretched cotton wool, tiny fingerprints on the rim. Soft window light from the left, warm bounce from a card just out of frame, shallow focus, visible set edge at the top of the shot. Hand-moved cadence, gentle 12fps stepping.

A motion shot:

Claymation chase: a clay rabbit sprinting across a miniature kitchen floor, armature visible at the knee, clay stretching slightly at the ankle on each stride. Overhead practical lamp, hard shadow, painted tile backdrop with visible brush texture. Stepped motion at twelve frames per second, slight camera shake as if the tripod was bumped.

Negative prompts that actually matter

Most disappointing claymation output fails for the same handful of reasons, so name them directly in the negative field: glossy CG render, airbrushed surface, motion blur trails, smooth 60fps interpolation, plastic toy sheen, photoreal skin, digital perfection, watermark, floating objects, inconsistent character design.

If your tool does not support negative prompts, invert the language in the main prompt: "matte, not glossy; stepped, not smooth; handmade, not rendered."

Pre-production: designing for a plasticine budget

Real stop motion is constrained by what a person can physically build. Those constraints are the style. If you design freely, you get something that could never have been built, and the illusion dies quietly.

Set design rules

Build small. A set should feel like it fits on a table: two or three planes of depth, a painted backdrop, and a foreground element like a teacup or a stack of books to establish scale. Avoid wide establishing shots; a claymation world reveals itself through close, curious framing, not aerial sweeps. When you need a sense of place, use a slow pan across a miniature background instead of a big reveal.

Character design rules

Simplify until the silhouette reads at a glance. Real clay characters have oversized heads, limited facial articulation, and hands that are basically mittens because fingers are painful to animate. Give your AI characters the same economy: three or four visible facial features, blunt hands, simple clothing shapes with visible sculpting seams. Then lock that description and reuse it word for word in every shot so the design does not drift.

Shot discipline

Write a shot list before you generate anything, and keep each shot under five seconds. Short shots hide continuity problems, match how stop motion is actually edited, and let you regenerate a single beat without rebuilding a whole sequence.

The end-to-end workflow

Step one: build a style bible. Generate eight to twelve still images of your character and set using the AI image generator. Pick the two that best capture the material feel. Save the exact prompt text. These frames become your reference for every clip.

Step two: keyframe before you animate. For each shot, generate a still of the first frame and a still of the last frame. If both look like they belong on the same tiny set, the clip has a chance. If they do not, no amount of motion prompting will save it.

Step three: image-to-video, not text-to-video. Feed your approved still into the AI video generator as the first frame and describe only the motion: "the fox turns its head slowly to camera, set stays static, cadence stepped." Locking the look in the image step removes half the variables.

Step four: generate three versions per shot. Even strong models produce one clip in three that has the right energy. Cheap insurance: always take three takes.

Step five: cadence and finishing pass. Bring everything into your editor. Apply a stepped time effect at 12fps, add a touch of grain, and let a few frames duplicate deliberately. Add tiny sound design: set creaks, clay squishes, room tone, a distant lamp hum. Sound sells stop motion more than picture does.

Step six: assemble and check continuity. Cut in order, then watch once with the sound off. Anything that reads as CG usually reveals itself in the first three seconds of a shot.

Making the camera feel hand-moved

A stop-motion camera is a physical object on a physical rig. It gets bumped, it drifts, it is re-focused by a human hand between frames. AI models tend to produce camera moves that are too confident, too smooth, too cinematic.

Fix it in the prompt with restraint language: "locked-off camera, barely perceptible drift, one small refocus mid-shot." Avoid words like sweeping, dynamic, epic, or cinematic push-in. If your tool supports camera controls, choose the smallest available value and dial it down further. When you do want movement, prefer a miniature dolly or a slow hand-cranked pan, and keep the move shorter than the shot.

Lighting a miniature world

Think like a tabletop photographer, not a cinematographer. Light comes from small sources placed inches from the subject, which means fast falloff, hard shadows at close range, and almost no ambient fill. The background card is lit separately and often looks it, which is part of the charm.

A reliable lighting recipe in prompt form: warm key from camera left at a low angle, hard shadow to the right, a bounce card just out of frame, the backdrop slightly brighter than the subject, and darkness beyond the set edge. Add atmosphere sparingly. Real sets are dusty, so a few visible dust motes or a faint haze near the lamp is authentic; heavy volumetric fog is not.

Editing, cadence and finishing

The edit is where most AI claymation is won or lost. Three cheap moves do most of the work:

  • Step the time. A posterize-time or frame-drop effect at 10 to 12 frames per second instantly reads as stop motion. Test 12 for character work and 10 for anything that should feel older or rougher.
  • Add a beat of imperfection. Duplicate one frame every second or two, or offset the exposure by a hair. Human animators introduce this naturally when they nudge a puppet.
  • Grade warm and slightly flat. Real sets are lit practically with no fancy highlight rolloff. Pull contrast back a little, warm the midtones, and keep the blacks from going fully crushed.

Sound deserves equal attention: a clay squish on contact, a low room hum, a soft click on hard landings. Add a slight ambient room reverb to everything and your miniature reads instantly.

Mistakes that break the illusion

Mistake What it looks like Fix
Glossy material Toy plastic or vinyl sheen Specify matte plasticine plus fingerprint texture; add gloss to negative prompt
Smooth motion CG float, eased movement Step the time in the edit; request 12fps cadence
Wrong scale Character looks life-size Macro framing, shallow focus, fast light falloff, visible set edges
Design drift Character changes between shots Lock a reference still and reuse identical prompt text
Overly complex sets Impossible, expensive-looking worlds Two or three planes of depth, painted backdrop
Too-long shots Continuity errors accumulate Keep every shot under five seconds
Missing sound Reads as a render Room tone, squishes, set creaks

Choosing between AI claymation and real stop motion

AI is the right call when you need volume, speed, or a style test: a social campaign with twelve clips, a pitch deck animatic, a music video where the look matters more than the physics. It is also excellent for previsualizing a real stop-motion shoot, because you can test character silhouettes and set layouts before anyone touches clay.

Real stop motion remains better when the material itself is the message: a hero brand film, a title sequence where the audience should feel the fingerprints, anything where tactile authenticity is the product. A useful hybrid: use AI to design and previsualize, build the key hero shot practically, then use AI for inserts and transitions.

One practical decision rule. If a viewer will watch a single clip once, on a phone, AI will almost always be indistinguishable. If they will watch it twice and look closely, your cadence, dust, and seams need to be right. Both are achievable; only one requires you to slow down.

FAQ

Can AI actually generate claymation-style video?

Yes, convincingly, with the right workflow. Texture is the easy part; cadence and consistent imperfection are the hard parts. Lock your look with still images, generate short clips from those stills, then step the frame rate in the edit.

What frame rate should claymation-style AI video be?

Generate at your tool's normal frame rate, then step it down to 10 to 12 frames per second in the edit. Twelve reads as modern studio stop motion; ten reads as older, rougher, more handmade. Anything above 15 loses the effect.

How do I stop AI clay from looking like glossy plastic?

Use material language rather than style language. "Matte polymer clay, visible thumbprints, waxy low sheen, armature seams at the joints" works far better than "claymation style." Add glossy CG render and airbrushed surface to your negative prompt.

How long should each shot be?

Two to four seconds is the sweet spot. Longer shots invite design drift and physics errors, and real stop-motion editing is built from short, punchy shots anyway.

Do I need sound design for a claymation clip?

It changes the result more than any visual tweak. Add room tone, a soft squish on contact, and a small mechanical click on hard landings. Without sound, viewers read the clip as a render.

Can I keep a character consistent across shots?

Yes, if you treat it like a puppet. Save one approved reference frame, reuse identical descriptive text every time, and never paraphrase the character description. Paraphrasing is the main cause of drift.

Start sculpting with Orelon

Claymation rewards planning more than any other AI video style, because every shortcut shows up on screen as a smudge of CG. Build your style bible as stills first, keep your shots short, generate three takes, and step the frame rate before you judge the result.

Orelon is built for exactly this kind of shot-by-shot directing: generate a reference frame in the AI image generator, turn it into motion in the AI video generator, and keep your prompts organized with the prompt library so every take stays on-model. If you want a running start, browse the video templates for tabletop and miniature setups, or read more production workflows on the Orelon blog. Bring a small world and a steady hand, and let the tool do the twenty-four setups.