We Deliver Anywhere
/ THE PITCH
A small town keeps a post office that delivers impossible parcels: a lost shadow, a spare dream, a jar of laughter, rain for one flower. A young postwoman, a talking backpack and a small cloud help each address work out which miracle it actually needs.
Format
Animated series, 26 × 7′
Audience
Preschool
Stage
Look locked, first shots out
Origin
Take42 Original
/ OVERVIEW
Everything in the frame is made by hand and then lit like a tabletop: characters sewn from felt and knitted wool, the town cut from card and painted, pile and stitching and loose thread left visible. Morning sun, cream and mustard and dusty rose and sage. Nothing here is smooth, and that is the whole point of the show.
/ FINDING THE LOOK
The manner is not described in words, it is picked. One prompt fans out across the whole style bank in five passes, sorted by family: anime, cartoon, volume and material, print and graphics, photo and render. The seed is fixed at 42 for every branch, so the only thing that changes between two frames is the LoRA. The run comes back as contact sheets, two finalists go through the same shot, and one of them becomes the show.
/ ONE PARCEL PER EPISODE
Seven minutes hold one delivery and one feeling, so the shot list is short and the props do the work. A scene goes in as plain text and comes back as a set of frames: first the action, edited from the frame before it with the camera untouched, then the angle, run through a separate branch that only knows directions, heights and shot sizes. Splitting the two is what keeps a stack of near-identical frames from reading as one.
/ THE SHOTS
/ BEHIND THE LINE
The Pipeline
Style search / one prompt / the whole LoRA bank / five families / seed 42 Look / handmade 3D / felt, wool and card / tabletop scale Frame / Flux 2 Klein 9B / 1344 × 768 / 4 steps / cfg 1 Backup frame / Krea 2 Kroma / Flux 2 Dev Character / passport frames / identity edit on the same base Story / Claude Opus 5 / one action per frame / a fixed camera vocabulary Board / Qwen Image Edit 2511 / angles LoRA / 4 steps Manner return / second pass on the base model / denoise 0.35 Motion / MiniMax H3 / first and last frame / 24 fps / 5 s Backup motion / LTX 2.5 / first and last frame Upscale / SeedVR2 / stills and video in separate passes Voice and sound / ElevenLabs / mixed outside the graph Grade / DaVinci Resolve Runs on / ComfyUI / one RTX 3090 / 24 GB Team / five people hold the core / twelve roles run an episode
