# zai-org/SCAIL

SCAIL: Towards Studio-Grade Character Animation via In-Context Learning of 3D-Consistent Pose Representations (CVPR 2026 Findings)

Repository: https://github.com/zai-org/SCAIL
Canonical: https://ross.abutalabs.com/products/scail
Language: Python
License: Apache-2.0
License Family: permissive
Topics: character-animation, video-generation, video2video
Last push: 2026-05-06T14:59:54+00:00

## Health v2 (maintenance only)
Score: 52/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 81, release rhythm 35, longevity 19
- inputs: {"age_days": 275, "days_push": 119, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1043, forks 59 (observed 2026-08-28T04:03:20.983138+00:00)

## What it is
SCAIL is the official inference implementation of a 14B diffusion transformer model that generates studio-grade character animation videos from a reference character image and a driving pose sequence. It uses in-context learning of 3D-consistent pose representations to handle complex motions like turning and flipping with coherent depth-aware motion.

## Use cases
- animate a character image from a motion capture or pose video
- generate studio-grade character animation from pose sequences
- drive a static character with complex motions like flips and turns
- video-to-video character reenactment with 3D-consistent poses
- produce production-quality animated character videos without a studio pipeline
- research controllable video generation with pose conditioning

## When to choose
- you need high-fidelity character animation driven by pose input
- your motions involve complex body rotation, flipping, or turning that simpler pose-guided models fail on
- you want to run inference with a pretrained 14B character animation model on GPU
- you are researching pose representation and conditioning for video diffusion models

## When to avoid
- you lack a high-VRAM GPU to run a 14B diffusion transformer
- you need training or fine-tuning code rather than inference
- you need real-time or low-latency animation
- you want lightweight 2D skeletal animation for games rather than generated video

## Facets
- artifact type: library
- maturity: active
- function: video-processing, machine-learning, deep-learning, llm-inference
- domain: artificial-intelligence, computer-vision, deep-learning
- platform: python
- tags: character-animation, video-generation, video2video, pose-driven-animation, diffusion-transformer, cvpr-2026, in-context-learning, 3d-pose-representation, image-to-video, research-model, video, gpu, linux

## Member repositories
- zai-org/SCAIL (main) score 52

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:20.983138+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T07:02:33.267517+00:00, confidence not recorded.
  - readme: https://github.com/zai-org/SCAIL (fetched 2026-08-28T04:03:20.983138+00:00, sha 89de47ab3b1a)
- Data as of 2026-08-30T08:39:29.467469+00:00.