Tasks and inputs
SCAIL-2 is listed for Image-to-video, Video-to-video. It accepts image, video and produces video. Use the publisher’s input format and preprocessing for your chosen deployment.
SCAIL-2 is Zhipu's open character-animation model that drives a reference character from any video, with no pose skeletons or masks. MIT licence.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: No API pricing stated; open weights for self-hosting
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Pose skeletons were always a workaround. They break under fast motion, they cannot describe a cat, and they make swapping one person for another in an existing clip awkward. SCAIL-2 drops them entirely. You give it a reference character and a driving video, and it animates the character end to end.
Zhipu (Z.ai) researchers open-sourced the model and inference code on 9 June 2026, with ComfyUI support the same day.
The team synthesised 60K motion pairs using several existing animators, then trained through a single Unified Motion Transfer Interface with dedicated masking channels and a custom RoPE design. A reverse-driving recipe let the student surpass its teachers. Three abilities emerged that the teachers lacked:
| Item | Detail |
|---|---|
| Resolutions | 512p and 704p end to end; pose-driven and replacement work better at 704p |
| Size rule | Height and width both divisible by 32 (for example 704×1280) |
| Bundled | Wan VAE and umT5-XXL text encoder inside the checkpoint |
| Licence | MIT |
Multi-reference inference arrived on 13 June and reached ComfyUI on 17 June. A relighting LoRA for replacement mode followed on 15 July. Training code landed on 6 August on a separate branch, with a community VRAM-friendly fork. A Bias-Aware DPO LoRA, which treats the biases of pose-driven generators as preferences to correct, sharpens fine detail.
Zhipu's hosted video models, such as CogVideoX-3, generate from prompts. SCAIL-2 is a control tool: the motion comes from footage you already have. For other ways to animate stills, see the image-to-video directory.
Record three driving clips: a slow walk, a fast dance, and a dog running. Animate one illustrated character and one photo of a real person with each, at 704p. Look at hands, the feet's contact with the ground and identity at the last frame. Then try a replacement on an existing clip with and without the relighting LoRA. That covers most of what you would ship.
Compare the model on the work you need to complete.
SCAIL-2 is listed for Image-to-video, Video-to-video. It accepts image, video and produces video. Use the publisher’s input format and preprocessing for your chosen deployment.
Choose duration, resolution and frame rate together. Compare the cost of an accepted clip, including repeat generations.
The weights are downloadable under MIT. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
tbc
TBC mischief 1 is an image-to-video model built on Wan 2.2, sold on fal and OpenRouter at about half a cent per second for 3-5 s clips at up to 720p.
NVIDIA
Cosmos3-H-Dreams is NVIDIA's open real-time surgical world model: it streams Versius robot video from one endoscope frame and two-arm commands.
NVIDIA
Cosmos3-H-Embodied-Nano is NVIDIA's open 16B surgical world model that does forward dynamics, inverse dynamics and policy from one checkpoint.
NVIDIA
Cosmos3-Nano-Sim-Bimanual is NVIDIA's open 16B world model that simulates video of a two-armed AgiBot robot as it moves its camera and manipulates.
noiz-ai
Helix World (preview) is an open real-time world model from Noiz AI and HKUST that streams explorable 768x512 video with stereo sound that follows the camera.
Alibaba
Wan-Animate-2 is Alibaba Wan's open 14B character-animation model that drives a character from video directly, with text-controlled camera angle.