omerbt/TokenFlow

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1708 stars · 143 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

TokenFlow is the official PyTorch implementation of an ICLR 2024 paper for text-driven, temporally consistent video editing using a pre-trained text-to-image diffusion model. It propagates diffusion features across frames via inter-frame correspondences, requiring no training or fine-tuning.

Use cases

• edit a video with a text prompt while preserving motion
• apply text-to-image editing methods consistently across video frames
• change the appearance of objects in a video without retraining a model
• research on diffusion feature consistency for video
• reproduce TokenFlow paper results

When to choose

• you need training-free, temporally consistent video editing with Stable Diffusion
• you want a research-grade reference implementation of the TokenFlow method
• you already have a text-to-image editing method and want to extend it to video

When to avoid

• you need real-time or production video editing at scale
• you lack a GPU or cannot run diffusion models locally
• you want to generate videos from scratch rather than edit existing footage

Facets

library · maturity stable

video-processing machine-learning image-processing artificial-intelligence computer-vision deep-learning python stable-diffusion diffusion-models text-driven-editing video-editing research-code iclr-2024 video gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="omerbt/TokenFlow")