Trigger: She is wearing a [insert length here], [insert color here] dress featuring double side slits.
This model seems to be very adaptive in my early testing. You don't have to word the trigger phrase in those exact words. As long as you are describing a dress with side slits, it should give you what you are after.
Example Prompt: The setting is a lavish, dimly lit gala ballroom, bathed in the warm, golden glow of crystal chandeliers. The air hums with the soft murmur of sophisticated conversation and the distant strains of a live orchestra. It's late evening, the atmosphere one of exclusive allure. An elegant woman with sleek, dark hair styled in an intricate updo moves gracefully across the polished marble floor. She wears a floor-length velvet gown in a deep sapphire blue, its fabric clinging to every curve, featuring double side slits that ascend provocatively high on her thighs, revealing glimpses of her toned legs with every step. The camera begins with a slow, sensual tracking shot, gliding across the ballroom, following her subtle sway. As she pauses to accept a glass of champagne, she subtly shifts her weight, causing one of the high slits to part dramatically, revealing a long, exquisite stretch of her leg as she meets the camera's gaze with a confident, alluring smile. The camera performs a deliberate, unhurried push-in, emphasizing the luxurious fabric, the provocative slits, and the powerful, seductive confidence in her eyes, all set against the backdrop of glittering exclusivity.
Description
FAQ
Comments (2)
playtime_ai have you figured out how to run LoRAs through the text encoders with 2.2 (assuming this is how trigger words get activated)? In my current experimental setup, I have a bunch of high and low LoRAs in their own sequences right after their respective models. I'm using the LoraLoaderModelOnly because I can't join the high and low LoRA clip output back together to run into the CLIP Text Encodes node. I thought of duplicating the encodes (one set for high and one set for low) but I think I would have to also duplicate WanImageToVideo node and I'm not sure that would work with the latent / KSampler.
I'm using a modified version of the default 2.2 14B I2V in comfy (basically just added the LoRA chains and sage attention).
stability ai seems to win against flux, latent modifers, and koya hifix, are nuts, but thats the problem, flux works out of the box, stability ai doesnt play nice at all, but the end results are beyond comprehension.
Details
Files
Wan2.2 - T2V - Side Slit Dress - LOW 14B.safetensors
Mirrors
Wan2.2 T2V Fixed.zip
Mirrors
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip
Wan2.2 T2V Fixed.zip