EP04 - Finetune of wan2.1 - Beautiful Blondes available on Ko-Fi https://ko-fi.com/s/ae70e26cdc .This update is to provide EP02 which is not a fine tune.
EP04 Finetune brings major improvements across the board:
✨ Drastically improved physics
🎥 Better handling of long-form video generations
🎯 Increased stability across a wide range of seeds
Whether you're pushing complex scenes or just looking for better consistency in your outputs, this version should give you way more control and realism. Stay tuned 👀 This update includes EP02 which is not a finetune V1.1 WIP. EP02 Version 1.1 Wan Video:
Additional training as I tinker with Wan. Introduced more jiggling and bouncing breasts; the effect is subtle in some cases, exaggerated in others. V1.1. Version 1.0 Wan Video WIP: Test of the Wan Video Lora WIP. Give it a shot and share your posts. Thank you. Trained on 14B T2V however the post used I2V have not made any T2V runs as of yet. Version 2.0 Enhanced + Cinematics: 🚨 Early Access Alert! 🚨
🚨 Version 2.0 is only available through https://ko-fi.com/singularunity Until March 27, 2025. Civitai Download on this page will provide you with Version 1.5 Enhanced until then. Thank you🚨
This LoRA model is now available for temporary purchase as part of an exclusive early access period through Ko-Fi: https://ko-fi.com/singularunity. Monthly subscribers can download new releases for free.
Example prompt style: A beautiful blonde woman with large breasts stands nude in a softly lit room, her fair skin glowing. She has long, straight hair cascading over her shoulders, and her blue eyes gaze directly at the camera. Her breasts are full and natural, with pink nipples. She has a slender waist and a flat stomach. Her pose is relaxed, with her arms hanging loosely by her sides. The background features a white wall and a wooden dresser with a green plant on top. The camera angle is eye-level, capturing her from the waist up.
This Lora version includes additional training on a curated, high-resolution dataset (images + videos) to push the boundaries of realism. The model now delivers:
✅ Improved physics/dynamic motion (e.g., walking, dancing, bouncing)
✅ Improved long form videos 97+ frames. Best at 129 frames.
✅ Cinematic elements and good coherence at denoise of 0.9.
✅ Enhanced detail in anatomy, particularly for more natural and lifelike results
✅ improved clothing designs.
✅ Sharper outputs thanks to the high-res training data
✅ Improved Camera Dynamics (Pans, Zoom, Orbits) The model has only been tested using my workflow. Super Simple HV2V - Now with UPSCALER and INTERPOLATION Version 1.5 Enhanced Details: Additional training on a curated small dataset (Hence the high Epochs) of high-resolution images and videos to enhance the quality of outputs and physics of bouncing jiggling breasts. I haven't done too much testing on it but so far it seems to generate more realistic bouncing breasts.
Prompting in general is the same; however, just make sure you include more jiggling and bouncing. Refer to the trigger words attached to this version and the info below. I haven't done many high-frame gens with this, just testing the ability to fine-tune or continue training existing lora's