post
Generative AI as an internal pipeline tool at a stop-motion/VFX studio
Stop-motion production map
Definition
At Tippett Studio -- Phil Tippett's practical-effects house, still shooting its own stop-motion features -- generative AI is not a replacement animation method but an internally built acceleration layer inside an existing VFX pipeline: trained on the studio's own material rather than a licensed third-party model, and scoped to specific production tasks such as background generation and animation assistance rather than to generating finished shots unsupervised.
Overview
Speaking on a panel at Italy's VIEW Conference (October 2025) -- CEO Gary Mundell, principal software engineer Remko Noteboom, art director Mark Dubeau, animation supervisor Tom Gibbons and VFX supervisor Marc Morissette, with Phil Tippett joining remotely -- Tippett Studio described why it built its own generative-AI tools rather than adopting an outside product: a closed, internally trained system keeps client material and creative decisions inside the studio, rather than exposed to a third-party vendor. The stated purpose is narrow and task-level -- "accelerate the process" on specific jobs like background generation and animation assistance -- not to hand a shot to a model end to end.
That internal training set is itself a craft decision. Noteboom said the studio trains its models on photographs of real physical objects specifically to capture tactile qualities that are difficult to get out of pure CGI -- an approach that keeps the model's raw material close to what a creature shop or a stop-motion stage already produces. He was equally direct about where it breaks down: "AI is really bad at creatures that are non-human," which is a significant caveat for a studio whose reputation rests on exactly that kind of creature work (Star Wars' tauntauns and AT-ATs, RoboCop's ED-209, Jurassic Park, and Tippett's own stop-motion feature Mad God, 2021).
The panel's caution was as pointed as its enthusiasm. Gibbons called the tool "a new kind of hammer" and sketch pad, but said flatly that "the emotion -- or the emoting that you get out of AI performances -- is seriously lacking." Dubeau described the arc of adopting it on a real job: after the initial excitement, "the bloom comes off the rose" once directionality and stylistic fit are tested against what a shot actually needs. Tippett's own framing set the tone for the panel: AI as "a creative partner, not as a thing in and of itself," with a dog-training analogy for how much back-and-forth that partnership takes -- "the dog trains you just as much as you train it."
Mad God is included here as background on the studio's practical, hand-crafted stop-motion identity, not as a confirmed credit for this AI pipeline -- the panel described the studio's general approach, not a specific shot or production. That distinction matters for a page like this one: the account so far is a conference panel, not a named production credit tying a specific generative-AI-assisted background or animation pass to a specific released shot.