How to Make AI Claymation: The Method (+ Real Example)

The short answer: design the clay before the film. Write a short script that captures your story beats, build one reference sheet per puppet showing character turnarounds and expression guides, write a plate that describes plasticine texture, fingerprints and a miniature set with forced perspective, board every shot with camera angles and puppet blocking, then generate one small action per shot keeping movements minimal. Finally, retime the clips to 12 poses per second in your editor to match traditional animation frame rates. That last step is what makes it read as stop motion rather than fluid CG animation.

Real clay remembers every touch—every pinch, press and adjustment—and that physical memory is the first thing an AI model erases when synthesizing movement. Ask a video model for "claymation" and you usually get a glossy toy gliding through a smooth shot with motion blur and interpolated frames: clay in a still, plastic once it moves. The tactile imperfections vanish.

This guide is about the whole handmade look: thumbprints left in the surface, lopsided shapes that shift slightly between frames, a tiny set with visible scale cues, stepped motion with no in-betweens. Our running example is Bertie, an original plasticine mole baker our team designed specifically for this article, and one real shot that took us three takes to get right. We show all three attempts, including the one we got wrong and why it failed.

What you need

Time: aim for a first film of 30 to 60 seconds, one set, one or two puppets. Most of the work is in the sheet and the boards.

Budget: writing is free. Images and video are paid in credits, and the bill depends on how many takes you run per shot. See the pricing and the AI video cost calculator.

Hardware: a computer with a browser. A phone helps if you sculpt a real maquette and want to photograph it.

Level: no animation skills. If you can describe a small, clear action, you can direct a clay shot.

What makes good claymation, before any tool

The model knows what clay looks like in a still—the matte finish, the subtle color variations, the sculptural quality. It knows little about how clay films are made frame by frame, so you supply the grammar: the staging principles, the timing charts, the way animators think in poses rather than motion paths.

1. The hand stays visible. Real plasticine carries fingerprints, tool marks, small dents and seams. Shapes are never perfectly symmetrical. A flawless surface is the fastest way to lose the look, so write the imperfections in on purpose.

2. Everything is small. A clay film is shot on a tabletop: shallow depth of field, soft background blur, thumb-sized props, felt, card and painted wood. If the set looks like a real kitchen, the puppet stops looking like a puppet.

3. Motion comes in poses, not glides. Film runs at 24 frames per second, and the standard practice in clay animation is to shoot "on twos": two frames per pose, so about 12 changes for each second of film (Wikipedia, Clay animation, checked 8 October 2026). That stepped rhythm is the signature of the craft. Video models produce smooth, interpolated motion, so you add the steps back after generation.

4. Simple shapes, strong silhouettes. Round bodies, chunky hands, bead eyes, one or two signature props. Fine detail melts between shots. Shape survives.

5. A locked camera. Stop-motion cameras are mostly tied down, because every move is animated by hand. A static frame looks more like the craft than a sweeping move.

6. Small actions, long reactions. Clay comedy lives in the pause: the cake collapses, the baker stares, the whiskers droop. Give the reaction its own shot and its own time.

Our stop motion technique sheet covers the same craft from the writing and boarding side, breaking down scene structure and shot planning.

Why "in the style of [studio]" fails

The temptation is to put a famous stop-motion studio or film title in the prompt. Don't, for two reasons.

The first constraint is intellectual property rights. Studio names and titles are registered trademarks, and their characters exist as copyright-protected creative works. A puppet design that replicates one cannot be legally owned by you, cannot be sold commercially, and cannot serve as the foundation for an original series you control.

The second constraint is technical in nature. A generative model does not store a single, distinct "studio look" in its training data. Instead, it averages all visual references associated with that studio name, which means you receive the same generic clay face as anyone else who typed identical prompt words, with an added risk of the output drifting toward a widely recognized character. The authentic look emerges from concrete creative decisions you can articulate and specify: matte-finish plasticine material showing fingerprint textures, bead eyes as the facial feature, a miniature physical set environment, and soft practical lighting setups. Describe those production elements explicitly, and the resulting aesthetic becomes uniquely yours.

Step 1: Idea and logline

Clay animation aesthetically suits small-scale worlds and narratives with simple, focused desires: one central character, one primary location, one core problem that escalates organically. For a character like Bertie, a suitable premise could be: a mole who operates the smallest bakery in the village attempts repeatedly to bake the perfect cake for a mysterious customer who never arrives to collect it.

Keep your first film project confined to one set environment: every additional location you introduce means another physical miniature that must be constructed, lit consistently, and held stable throughout the shoot.

Step 2: Write the script

Write your story in standard screenplay format, even for a short film: the AI shot breakdown system reads screenplay structure directly and processes it more accurately. 熙熙's built-in screenplay editor is available at no cost on the 免费档 tier.

Two habits make a clay script generate well:

Write physical actions, not inner states. "Bertie presses his paw to his cheek and leaves a floury print" films. "Bertie feels defeated" does not.
Give holds their own line. "He stares at the cake. Hold." is a direction the board can time.

If you come from live action, read our piece on writing animation versus live action to understand the key storytelling differences between the two mediums.

Step 3: Build the reference sheets

Every puppet, set and key prop gets a reference sheet before the first panel: in 熙熙, Characters, Places and Objects, generated with AI or built from your own uploaded images. In clay animation specifically, this reference sheet is where the film is won or lost—it establishes the visual DNA of your entire production—so it has its own dedicated section below.

Here is Bertie's sheet, made with GPT Image 2.5, one of the image models in 熙熙: front view, three-quarter angle, side profile and back view, plus three distinct expressions showing different moods.

Character reference sheet of Bertie, an original plasticine mole baker in a chef's hat and a floury apron with a red pocket, shown from the front, three-quarter, side and back, with three expressions below

The sheet fixes the grey clay with thumb marks, the pink nose, the round wire glasses, the floury apron with its red pocket, the bow at the back. Every shot gets checked against it to ensure visual consistency across your entire film.

Step 4: Write the art direction plate

The plate is one short block of text that describes the material, the set and the light. Attach the same block to every image and video generation in the film, so shot 12 is made of the same clay as shot 1, maintaining material and lighting continuity. A plate for Bertie's bakery looks like this:

Handmade claymation. Matte plasticine with visible fingerprints,
tool marks and slightly uneven shapes. Small black bead eyes.
Miniature tabletop set built from painted wood, card and felt.
Warm practical light from a tiny oven and a window, soft shadows.
Shallow depth of field, background in soft blur. Locked-off camera.

No studio, no title, no artist name—when a frame emerges with that polished, commercial sheen, you immediately recognize which creative boundary the system decided to override.

Step 5: Board every shot

Board before you spend credits on video. The board fixes staging, the puppet's size in frame and every hold. In clay, it also puts the camera at puppet eye level, which is low for a human.

熙熙 uses AI to decompose each scene into individual shots, then produces storyboard panels rendered as grayscale 3D previsualization, allowing you to evaluate camera placement, character blocking, and temporal pacing well before committing to any final textures or surface appearance. For a deeper look at this stage, see the AI storyboard generator.

Step 6: Generate shot by shot

Begin by generating a static keyframe for every shot, with the character sheet reference attached to maintain consistency. Bertie's opening shot—him carefully lifting a slightly lopsided loaf from his compact brick oven—rendered perfectly on model: the same distinctive hat, flour-dusted apron, round glasses, and that characteristic clay texture all matched the reference sheet exactly.

Bertie, the plasticine mole baker, lifts a lopsided loaf from a tiny brick oven in a miniature bakery lit by a warm fire and a sunset window

Then move into the animation phase. Within 熙熙's production canvas, the shots that make up a scene are processed through one of two distinct generation pipelines:

Reference to video: each shot's action becomes a directed prompt, and the characters, places and objects in the shot are attached at the video step.
Storyboard panels to video: the boarded panels of a scene, up to 14, guide the order and composition of the sequence, alongside the references.

Seedance 2.5 is the default video model: clips from 4 to 30 seconds, up to 1080p, up to 30 image references. Seedance 2, Veo 3.1, Kling 3.0, MiniMax H3 and Grok Imagine Video are also available, shot by shot.

Rules that save credits in clay:

One action per shot. "He lifts the loaf" is a shot. "He lifts the loaf, turns and shelves it" is three.
Small, deliberate moves. Wide, fast motion is where clay melts and limbs stretch.
Draft cheap, finish sharp. Pay for 1080p only on the take you keep.

When a still is almost right, fix it before animating. Crop, Angle (a new camera angle on the same frame) and Edit turn a wrong prop into one image edit, not a new video.

Step 7: Voice, music and sound

熙熙 does not generate voices, music or sound effects—these audio elements remain outside its scope. Record dialogue tracks yourself, hire a voice actor, or integrate a text-to-speech service before assembling the final cut. Secure licensed music or collaborate with a composer. Because clay carries such a strong tactile presence, foley work becomes especially important: the wet squish of a paw pressing into dough, the metallic clink of a baking tray being set down. Our sound workflow for AI films walks through the complete audio pass in detail.

Step 8: Edit, retime and export

熙熙 is not a finishing tool—it does not handle editing or exporting your final film. Once you have generated and kept the clips and audio you need, you will bring those assets into a professional editing application such as DaVinci Resolve or Premiere Pro to assemble, color-grade, and export your completed project.

This is where the claymation look is finished. Generate at the model's native frame rate, then hold every pose for two frames: 12 new images per second on a 24 fps timeline. A posterize-time or frame-hold effect does it per clip. The glide turns into the stepped rhythm of stop motion, and the clay starts to look moved by hand. The video frames calculator does the arithmetic for any duration and frame rate.

For 16:9 and 9:16 versions, the recommended workflow is to generate both aspect ratios as separate renders rather than cropping one from the other, ensuring that composition and framing are optimized for each format.

Character consistency in clay: the section that decides everything

Every generation reinvents the character. Describe "a grey clay mole in a chef's hat" in ten prompts and you get ten different moles. A film needs one specific puppet.

Clay animation makes maintaining consistency significantly harder: a real physical puppet exists as one tangible object, so if details like the apron pocket suddenly switch sides between frames, the audience immediately recognizes that it cannot be the same puppet. The underlying technical challenges and solutions are explored in depth in character drift in AI video.

The rule: attach, don't redescribe

Once a character has a sheet, stop describing his look in prose and attach the sheet. The prompt only says what changes in this shot: pose, expression, action. Bertie's second shot shows exactly why.

Bertie's worried shot: three takes, one lesson

The shot we were working on: Bertie stands staring with a worried expression at a cake that has collapsed and sits deflated on the kitchen counter. We attached the same character reference sheet each time we generated, across three separate takes.

The same claymation shot of Bertie the mole baker worried at a collapsed cake in three takes: in the first two he wears round glasses, in the third his face is bare

Take 1. Our prompt redescribed Bertie in prose on top of the attached sheet: "pink nose, round glasses, apron..." The model followed the words. The result is on model, but only because every word happened to agree with the sheet.

Take 2. We removed the description and wrote: "the character from the reference sheet, exactly as he appears there and nothing added." Glasses, pocket, hat and nose all came from the sheet alone. On model again, with a prompt that carries only the action.

Take 3. By this point we had somehow convinced ourselves that Bertie's design featured a bare face, so we wrote the prompt "bare face with NO glasses, small black bead eyes." The model followed the written instruction exactly and removed the glasses from the generated frames. But when we look back at the reference sheet, Bertie clearly wears round wire glasses in every single view. Our working file for this particular take was even confidently named "fixed". It turned out to be the one take that drifted off model.

Three lessons, for the price of three image generations on one shot:

Prose overrides the sheet. When the words and the image disagree, the words tend to win. Every description you add is a second source of truth that can be wrong.
Check against the sheet, not your memory of it. We were sure, and we were wrong. Put each take next to the sheet before you keep it.
Name a detail only after reading it off the sheet. Naming a detail the model keeps getting wrong is a good fix. Naming one you remember wrong breaks a shot that was fine.

Take 3 of the worried shot: Bertie the clay mole with a bare face and no glasses, staring at a collapsed cake

The claymation character kit

Build this for each recurring puppet before the first panel:

A turnaround: front, three-quarter, side and back, same pose, same light, plain background. Use the Angle tool to fill a missing view instead of describing it.
An expression sheet: neutral, worried, delighted, plus whatever the script needs most.
Accessory and costume close-ups: glasses, whiskers, the pocket, the bow. Small, high-contrast details are the first to appear or vanish, as Bertie showed.
A scale line-up: every puppet side by side and next to one key prop. Models forget relative size as soon as two characters share a frame.

Lock the clay finish too: a puppet that turns glossy has turned into plastic.

Use your own clay if you have it

If you sculpt, make a real maquette, even a rough one, and photograph it from four sides in even light. Upload the photos as the character's reference images. Your fingerprints are then real fingerprints, and the proportions come from an object that cannot drift. Nano Banana 2 accepts up to 14 reference images and Seedance 2.5 up to 30: enough for a puppet from several angles, a second puppet, the set and a prop.

The drift test

Before the second half of a scene, put shot 1 and the latest shot next to the sheet. Same accessories, same pocket side, same clay? Drift compounds when you reuse a generated frame as the next reference instead of the sheet: a copy of a copy. Always go back to the sheet. The full method for boards is in character consistency in AI storyboards, and the AI character consistency sheet helps you build a sheet from a description.

Try it free

Try 熙熙 for free on your script

It is free. Import your existing project, get a clearer view of your outline, and regain control of your story structure in minutes.

Start Free

One shot, taken apart

Bertie's bakery shot, layer by layer. We made it as a single still for this guide, so the script line and board show how it would sit in a full film.

The script line: "Bertie hauls a lopsided loaf out of the oven and beams at it."

The board: a medium shot at puppet eye level, Bertie centre left, the glowing oven on the right. The oven is the light source, so it sits in frame.

What gets attached: Bertie's sheet. His hat, apron and glasses do not need to be written in the prompt.

What we check: hat, glasses, pink nose, red pocket, stubby feet. Then the material: thumb marks, an uneven loaf, a set built from wood and card, a soft background. We would reject glossy skin, a perfectly smooth loaf or a missing accessory.

What comes next in a film: the still becomes the first frame of the video shot, and the clip is retimed on twos in the edit. We did not animate Bertie for this guide.

How long it takes and what it costs

We won't quote a fixed price. It depends on three numbers only you control: how many shots, how long each runs, and how many takes you need before one matches the sheet. Bertie's worried shot took three image generations. Budget it this way:

Board the whole scene to get the real shot count.
Generate one full scene and count your own takes per kept shot.
Multiply by the number of scenes with the AI video cost calculator.

Writing is free on the 免费档, paid plans are in the pricing section, and for a broader view, read AI film production costs. For teams evaluating claymation workflows alongside other AI animation tools, the pricing page breaks down credits, rendering time, and export limits across tiers. For teams evaluating claymation workflows alongside other AI animation tools, the pricing page breaks down credits, rendering time, and export limits across tiers.

Common mistakes, and how to fix them

1. Generating before writing. Clips with no story become a mood reel. Fix: logline, script, sheet, boards, then video.

2. Redescribing the puppet in the prompt. When the words disagree with the sheet, the words win, as our take 3 showed. Fix: attach the sheet and keep the prompt to the action.

3. Trusting your memory over the sheet, and keeping the first take. We wrote "NO glasses" for a mole who wears glasses. Fix: open the sheet and compare every take side by side before you keep it.

4. A clean, perfect surface. No fingerprints, no seams, a glossy finish: it reads as plastic or CGI. Fix: write the imperfections into the plate and reject glossy takes.

5. Leaving the motion smooth. The clip moves like a toy commercial. Fix: retime to 12 poses per second in the edit.

6. A life-size world. Real-scale rooms and deep focus kill the miniature. Fix: tabletop materials, shallow depth of field, a camera at puppet eye level.

7. A studio name in the prompt. An averaged look and a design you cannot own. Fix: describe the clay, the set and the light.

Which tool to use

Kind of tool	Good at	Where it breaks for claymation
A claymation AI generator or clip app on its own	One charming clip from a prompt	No memory between clips, so the puppet changes. Script, boards and edit live elsewhere.
All-in-one video apps and "clay" filters	Turning a photo or footage into a clay look	Little control over staging, the same generic clay as everyone else, hard to hold a cast across multiple shots or scenes across multiple shots or scenes
Real stop motion with a camera	Real clay, real fingerprints, total control	Every second is many hand-moved poses, plus a built set and lights
A production suite: script, storyboard and video in one project	Holding puppets, sets and decisions from page to frame	You still write, board, choose the takes and retime in the edit—full authorial control over character consistency and scene blocking—full authorial control over character consistency and scene blocking

We built 熙熙 to be that last row: the puppet you design on day one is still the same puppet in shot 30, maintaining character identity and design choices throughout your claymation project. There is no claymation page on the site yet. The closest are the AI 3D animation generator, for stylised characters and sets, and the AI family movie generator. For neighbouring methods, read how to make 3D animation with AI and how to make an AI cartoon.

Proof: what this method produces

Bertie, a test by the 熙熙 team. One original clay character, one sheet, one bakery shot and one shot in three takes, made with GPT Image 2.5 and shown above as they came out, including the take we got wrong. It is a test, not a released film—a real glimpse into iteration, puppet consistency, and set continuity in claymation production.

Lost Garden serves as the long-form proof of this identical pipeline's capabilities. Spanning two complete episodes and delivering 38 minutes of polished anime content, this project was executed single-handedly by one creator working within 熙熙. The output is genuine anime rather than clay animation, yet the underlying methodology remains consistent: script development, reference sheets, storyboard construction, with one deliberate beat allocated per shot. For those seeking deeper insight, the comprehensive case study is available: how I made a Japanese-style anime with AI.

FAQ

Can I make AI claymation for free?

The complete script can be written at zero cost using 熙熙's 免费档. Reference sheets, storyboard generation, and video production consume credits under a paid subscription tier, while voice recording, music composition, and editing are handled through separate tools—several of which offer free tiers. Notably, the writing phase and the foundational plate work, which collectively determine the majority of the visual aesthetic, require no financial investment.

What is the best AI for claymation?

No single model wins every shot. We made Bertie's stills with GPT Image 2.5. For video, Seedance 2.5 is our default because it takes up to 30 image references, which is what holds a puppet on model. Test the same board on two models and keep the one that respects the sheet.

How do I make AI video look like real stop motion?

Write fingerprints, uneven shapes and a miniature set into the plate. Keep the camera still and the actions small. Then retime the clip in your editor so each pose holds for two frames, 12 poses per second, the standard rhythm of clay animation.

Can I use photos of a clay figure I sculpted?

Yes, absolutely. Photograph your physical maquette from the front angle, three-quarter view, side profile, and back in evenly distributed lighting, then upload these photographs as the character's reference images within the system. A tangible, real-world object provides the model with consistent proportions and authentic surface texture that remain stable across generated frames.

Can I make a claymation series or a music video for YouTube?

Yes. A series works because the same sheets carry over between episodes. For a music video, the track comes from you or a licensed source: 熙熙 does not generate music. YouTube's AI disclosure label targets realistic content, and clearly animated content generally does not need it (YouTube Help, checked October 2026). For monetisation, YouTube's "inauthentic content" policy excludes mass-produced, template-made videos (YouTube monetisation policies). An original puppet with an original story is the opposite of a template.

Start with the puppet

Write the logline and the first scene tonight. The free 免费档 is enough: open 熙熙. When the script is ready, build the sheet, write the plate and board every shot.

For understanding the complete fiction production pipeline that extends beyond clay animation techniques, consult the detailed guide: how to make an AI film.