GGUF | 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

FP16: 10GB VRAM + 64GB RAM + Diffusion Model Loader KJ + triton

It'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

More info | Example Prompts

• 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.

• Cool Features: Turns static pics into action simulations; keeps edits consistent with gravity, motion, etc. Handles square or landscape/portrait sizes up to 1024x1024.

• 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.

• 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.

• Basic Specs: 14 billion parts, Diffusers format, open license for business. input pic + text, output edited pic, no extra hassle needed.