EchoShot: Multi-Shot Portrait Video Generation
EchoShot is a multi-shot portrait video generation model built on Wan2.1. Generates consistent portraits from multiple reference images (NeurIPS 2025).
EchoShot
Wan2.1PortraitMulti-ShotNeurIPS 2025Multi-shot portrait video generation framework built on Wan2.1. Generates identity-consistent portrait videos across multiple shots with flexible content controllability. Accepted at NeurIPS 2025.
| Developer | D2I-ai |
| Release Date | 2025-06 |
| Architecture | Wan2.1-based with identity LoRA |
| Capabilities | Text-to-multi-shot portrait video generation with identity consistency |
| Venue | NeurIPS 2025 |
Overview
EchoShot is a native and scalable multi-shot framework for portrait video customization built upon Wan2.1. Unlike conventional video diffusion models that are limited to single-shot generation, EchoShot enables users to generate multiple video shots showing the same person across different scenes and prompts while maintaining identity consistency.
The model uses identity LoRA embeddings to preserve facial features across shots, allowing flexible content controllability through customized text prompts. This makes EchoShot well-suited for storytelling, virtual avatars, and cinematic portrait video production where consistent character appearance across varied scenes is essential.
ComfyUI Integration
EchoShot is available in ComfyUI via the Kijai WanVideoWrapper custom nodes. The fp16 and fp8 scaled model variants are hosted on Hugging Face under Kijai/WanVideo_comfy.
Resources
| Resource | Link |
|---|---|
| Paper | arXiv:2506.15838 |
| GitHub | github.com/D2I-ai/EchoShot |
| HuggingFace (Original) | JonneyWang/EchoShot |
| HuggingFace (Kijai) | Kijai/WanVideo_comfy |
| Project Page | echoshot.org |
Guides and workflows related to this model series.