<a rel="ugc" href="https://huggingface.co/QuantStack/ChronoEdit-14B-GGUF/tree/main">GGUF</a> | Wan2.1 T2V LoRA:s are compatible (NSFW)For now, you need to change ComfyUI to the nightly version, as some nodes are not available in the stable version yet. It may be updated by next week. Workflow in the zip file<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/38bb6161-be20-4e25-853e-50304b94d38d/width=525/38bb6161-be20-4e25-853e-50304b94d38d.jpeg" />FP16: 10GB VRAM + 64GB RAM + Diffusion Model Loader KJ + tritonIt's an image editor. it takes a starting photo and text instructions, then spits out a changed version of that photo. But it borrows smarts from video models (like Wan2.1) to handle stuff like motion or physics better, treating the edit like a super-short "video" in its brain for more realistic results. It shares the same text encoder and other bits from Wan video models since it's built on top of them<a target="_blank" rel="ugc" href="https://huggingface.co/nvidia/ChronoEdit-14B-Diffusers">More info</a> | <a target="_blank" rel="ugc" href="https://research.nvidia.com/labs/toronto-ai/chronoedit/">Example Prompts</a><ul><li>Smarter Setup: Builds on a big video model for time-based editing, adding physics smarts to simulate actions like robot moves or object grabs, beats basic image editors by handling real-world dynamics.</li><li>Cool Features: Turns static pics into action simulations; keeps edits consistent with gravity, motion, etc. Handles square or landscape/portrait sizes up to 1024x1024.</li><li>What It Excels At: Best for PhysicalAI tasks like robot planning or interactive scenes; trained on fake world data, so shines there but might slip on everyday pics.</li><li>Easy Tips: Pair an image with short text instructions (under 300 words), like "make the robot pick up the ball realistically." Run on NVIDIA GPUs for speed; add safety checks for real use.</li><li>Basic Specs: 14 billion parts, Diffusers format, open license for business. input pic + text, output edited pic, no extra hassle needed.</li></ul>

ChronoEdit

chronoedit_fp16.safetensors

chronoedit_fp16_trainingData.zip

FP16

FP8-e4m3fn

Join LUXED AI, the best AI community: <a target="_blank" rel="ugc" href="https://discord.gg/HxfP9TnctJ">https://discord.gg/HxfP9TnctJ</a> <img src="https://resouces.modelscope.cn/proxy-image/eab28bf5995ed66879ec21a7af64f18f5a5c91fdc7e5aed28b5d76402fd120d0.png" width="400" />💚 <a target="_blank" rel="ugc" href="https://research.nvidia.com/labs/toronto-ai/chronoedit/">ChronoEdit</a>    ｜    🖥️ <a target="_blank" rel="ugc" href="https://github.com/nv-tlabs/ChronoEdit">GitHub</a>    |   🤗 <a target="_blank" rel="ugc" href="https://huggingface.co/collections/nvidia/chronoedit">Hugging Face</a>   |   🤖 <a target="_blank" rel="ugc" href="https://huggingface.co/spaces/nvidia/ChronoEdit">Gradio Demo</a>   |    📑 <a target="_blank" rel="ugc" href="https://arxiv.org/abs/2510.04290">Paper</a> ChronoEdit: Towards Temporal Reasoning for Image Editing and World Simulation ChronoEdit-14B enables physics-aware image editing and action-conditioned world simulation through temporal reasoning. It distills priors from a 14B-parameter pretrained video generative model and separates inference into (i) a video reasoning stage for latent trajectory denoising, and (ii) an in-context editing stage for pruning trajectory tokens. ChronoEdit-14B was developed by NVIDIA as part of the ChronoEdit family of multimodal foundation models. This model is ready for commercial use.<img src="https://research.nvidia.com/labs/toronto-ai/chronoedit/assets/method.png" alt="ChronoEdit Method Overview" />Overview of the ChronoEdit pipeline. From right to left, the denoising process begins in the temporal reasoning stage, where the model imagines and denoises a short trajectory of intermediate frames. These intermediate frames act as reasoning tokens, guiding how the edit should unfold in a physically consistent manner. For efficiency, the reasoning tokens are discarded in the subsequent editing frame generation stage, where the target frame is further refined into the final edited image.

Files with SHA256: 3294a5795e6cb5d16c8f3f34b8a22fe0a600e35cb49624c52f2100cd320a5988