Give it the shot list (one metaphor per shot, llm-pick-a-visual-metaphor) and the
aspects the piece ships in; get the style line and the video phrase that make every
still and clip read as one handmade clay world. The process around it (style frames,
the locked still, the stills on edit, one motion test, then the batch) is
guide-pick-a-stylized-look; the prompt grammar is nano-banana-prompt-format and
h3-prompt-format.
Use clay when a handmade, tactile treatment supports the approved concept and brand. It animated cleanly on H3 Max in the worked toy-world ads; that is a production advantage once the treatment is chosen, not a reason to choose it for every brand.
Inputs
- The scene of each shot, with the camera and which way every figure faces.
- The hero, if one recurs: a toy-like figure with a simple friendly face, described in
one line, and its sheet (
nano-banana-character-sheet) when it appears in more than two shots. - The aspects: plan every 16:9 frame with the action in the middle third when a 1:1 or
9:16 ships too (
guide-pace-a-short-for-social).
Run it
The style line, identical in every still prompt of the job:
warm, stylized 3D miniature, like a handmade toy-model world photographed with a
tilt-shift macro lens; soft matte clay and painted wood; rounded shapes; shallow depth
of field; toy-like figures with simple friendly faces. Not photoreal.
- The first still on nano banana 2.1, 2K, the deliverable's aspect, two takes; the pick locks the look and the hero.
- Every other still on nano banana 2.1 edit with the locked still as Image 1: "use it ONLY for the art style and the character. Do not copy its room." End each prompt "No text, no logos." A prop that recurs gets its earlier still as a reference too ("use it ONLY for the robot").
- Video: H3 Max image-to-video (
h3-max-image-to-video) with "3D CG claymation" as the style, 768P, a fixed seed, mouths closed and no speaking verbs. Four-second shots worked for the five-beat ad; the current script determines each shot's duration.
What you get
Stills that share one clay world, and clips at 768P (1344×768 for 16:9) that scale to 1080p for delivery with lanczos and hold up: soft matte clay has little fine detail to lose.
Tips and traps
- 768P, not 1080P: half the price, and clay holds up scaled.
- Sizes by body parts, not objects. "A robot the size of a teacup" came back as a literal teacup; "as tall as her knee" works.
- Laptops and phones grow real manufacturers' logos despite the prompt: keep devices out of the scene, or pick the take without one rather than editing it out.
- Native 9:16 redraws of each still, not crops (
guide-pace-a-short-for-social).
Check it
- A sheet of every still (
python-contact-sheet,images): one material, one light, the hero on model, no lettering, no logos on devices. - One clip's motion approved before the batch (
guide-pick-a-stylized-look, step 5).
