Things are getting crazy. AI just one-shotted this claymation cartoon using code and not an image model. My 11 yo son gave the story. Full prompt and model info in next comments.

I used Opus 5.5 (medium) on Claude App/Code. Gave it access to a folder on my computer (note: Blender already installed on my computer, ElevenLabs connected to my Claude account). Here's the full prompt. Credits to my son for the story. PROMPT: Make a claymation cartoon with this story. Use elevenlabs for voice but not video. Use code to create it. Use my computer to render. Comfy the couch was made on september 12th 1996 in a mass production company. Comfy, (as we may call him) was especially comfy but was only reconize as super comfy when he came to his first owner who named him burger the cat. Soon he was sold to a collector for 100,000 dollars. That collector didn’t like naming things and called him couch. Now little did the collector or the first owner know, he was alive. It was due to some malfunctioning in the machines. Now, comfy didn’t like the collector because the collector put him in a public lobby and many people sat on him. When a three year old toddler peed on him he decided it was enough! He made himself super hard whenever somebody sat on him. Soon, the collector washed him and put him in a box. After a while of complete darkness, some rattling, and muffled talking comfy ended up in a nice family’s house. What really happened was the collector decided to sell comfy and sold him for 100,000 bucks. The new family was nice and named him comfy the couch. Nobody but the family knew he was alive. Now, in 2026 he is still owned by the family and the family lets a few people see him but only if comfy allows it. The end.

You are impressive on a number of levels which is why I have been following your work since lockdown. Over the years I have gotten a bit of insight into your personal life and one thing that is more impressive than any of your analysis is EXCELLENCE as a father. Whether it’s taking long road trips to share the world with your kids or enabling their growth WITH technology, I am ever impressed. You are a very good father.

I gave same promt to Grokbot and it did euqally good if not better

Can you ask Grok Bot which model it used to build that?

as an ex-3D artist this is the part that gets me. driving blender through code means you get real geometry, rigs and lights you can still tweak after, not baked pixels. curious how it handled the clay look, subsurface + fingerprint noise or just a roughness map?

Answer to your question from the AI thread I used to create the claymation: Subsurface plus procedural bump, not a roughness map. The roughness is actually a single fixed value. The look is built in three layers, all in clay() and finish() in 1. Material (Principled BSDF) Subsurface: a low weight (0.08 for most clay, 0.1 for Comfy, 0.15 for skin) with a tiny scale (0.04) and a red-weighted radius. That gives the soft, slightly warm edges you get where light goes into plasticine, without making it look waxy. Roughness: one value, 0.55–0.62, with lowered specular. Some materials get variations: the pee puddle, pupils and top hat are glossier, and a few use a clear-coat layer. Colour wobble: a large, low-frequency noise multiplies the base colour between 0.86× and 1.08×, so no surface is one perfectly flat colour, like hand-mixed clay. 2. Surface detail (bump only, no displacement in the shader) Grain: a fine 4D noise (scale 55, detail 6) for the tiny surface texture. Smudges: a 4D Voronoi in Smooth-F1 mode (scale 7) for the shallow thumb-press dents. These two feed one Bump node at strength about 0.3. Boil: the 4th dimension (W) of both textures is driven by frame * k, so the texture shifts slightly on every frame. That mimics clay being touched between stop-motion frames. Since everything renders at 12 fps, the shift happens on each held frame. 3. Shape Every part is a cube or sphere, then bevel, then subdivision, so it reads as pillowy rather than CG-sharp. A Displace modifier then adds Clouds noise in local coordinates at a strength of about 0.012. That gives the slightly uneven, hand-rolled silhouette. It’s in local space so the lumps stay put on a moving character instead of sliding across it. Mouths, brows and wires are bevelled curves, which read as rolled clay snakes. Mouths are swapped by visibility per frame, like replacement mouths in real stop-motion. What’s missing: there are no real fingerprints. The Voronoi gives dents, not ridged prints. Two upgrades would help if you want it more tactile: Small patches of fingerprint ridges, made from a Wave texture in rings mode with distortion and masked by sparse Voronoi cells. Roughness driven by the same smudge texture, so pressed areas look slightly shinier. Either one is a small change to clay() and would apply to every scene on the next render.

makes sense, a fixed roughness keeps it from going plasticky. procedural bump over SSS is basically how we faked clay back in my 3D days, just never at code speed. did it pick the subsurface radius on its own or did you nudge it?

The prompt above is the full entire prompt. And I didn’t give any other replies for the video I posted. 🤯

so the red-weighted radius and the per-frame boil on the noise W axis all came from that one prompt? that boil trick is the bit most 3d folks skip. would love to see the wave-texture fingerprint upgrade it suggested on the next render

Ok, I told it to add the upgrades and here's the output.

ok that's wild, one follow-up prompt and it took the whole upgrade list. did the noise boil settle on its own or did it need a nudge? thanks for actually running it

Hey Dave, I love your content and posts. I remember that about some years ago you posted about a plan you made for your kids private Education. I read it then and it was great. How do you support your kids' training and education right now with tools like Grok Bot? Can you maybe take us on a deep dive into what your kids learn by now and how you support them currently? Thanks a lot.

Wow! Any idea of cost/number of tokens/time required?

Well well well...

Easy to do this kind of neat things. Very few people can make any actual money with it.

kid supplies the plot, model does the rest, solid division of labor

If you feed in harry potter the whole book, do you think it can output a long tv series?

So is this the merging of film and gaming industry? Both will use 3D models that can be hyper real? Pixels are cooked? We go straight to vector on all video generated?

Thanks for sharing, Dave. This is incredible, congrats to your son. Going to try this out with my kids.

Claymation from one prompt is wild. Credits to your son for the story 😄

I think you are onto something with code being a replacement for image models...

That’s very cute. ❤️