| --- |
| |
| tags: |
| - lora |
| - wan-video-2.2-i2v-a14b |
| - text-to-image |
| - diffusers |
| - template:diffusion-lora |
| - action |
| - civitai |
| - imported |
| library_name: diffusers |
|
|
| --- |
| |
| # Forced BJ |
|
|
| <Gallery /> |
|
|
| ## Model description |
|
|
| Trained on https://yorespot.com \u2014 where creators train faster, cheaper, and with minimal censorship (within legal limits). If you\u2019re serious about building models, that\u2019s where you should be. |
|
|
| For the hottest working Loras for wan2.2 that CivitAI has removed visit the site above we got them before they were deleted. |
|
|
| WAN 2.2 I2V LoRA |
| ======================================================== |
| Trigger Word |
| ysp_forcbj24 |
| |
| Overview |
| This is a WAN 2.2 I2V LoRA trained for high-quality portrait video generation. It uses a two-phase training approach to properly separate low-noise detail learning and high-noise motion structure \u2014 resulting in cleaner motion, stronger consistency, and better frame coherence than single-pass LoRAs. |
| |
| Training Details |
| Base Model:\u2022 WAN 2.2 I2V 14B\u2022 (LOW + HIGH LoRA files) |
| Method:\u2022 Musubi Tuner video LoRA workflow\u2022 Two-phase training (LOW timesteps \u2192 HIGH timesteps) |
| Dataset:\u2022 16 portrait video clips (512\u00d7512)\u2022 Hand-captioned (.txt per clip)\u2022 Full frame extraction (~30 FPS)\u2022 Up to 33 frames per clip\u2022 4 repeats per clip |
| Training Config:\u2022 LoRA rank: 32\u2022 Alpha: 32\u2022 Learning rate: 2e-4\u2022 Optimizer: AdamW (8-bit)\u2022 Scheduler: Cosine\u2022 Precision: bf16 + gradient checkpointing |
| Training Process:\u2022 50 epochs (LOW noise model)\u2022 50 epochs (HIGH noise model)\u2022 Total: 100 effective passes across both experts |
| Training Time:\u2022 ~20 hours total runtime |
| |
| Recommended Usage (IMPORTANT) |
| This is a dual-LoRA system. You MUST use both files together. |
| HIGH noise LoRA:\u2022 Strength: 0.55 |
| LOW noise LoRA:\u2022 Strength: 0.90 |
| Best results come from using both simultaneously in a WAN 2.2 I2V workflow. |
| Using only one will significantly reduce quality. |
| |
| Why This Matters |
| Most LoRAs fail at motion consistency or overfit to static detail. This setup avoids that by splitting the learning problem: |
| \u2022 LOW model \u2192 structure, identity, fine detail\u2022 HIGH model \u2192 motion, transitions, temporal coherence |
| The result is smoother, more stable, and more realistic video output. |
| |
| If you want to train models li... |
| |
| ## Model Details |
| |
| - **Type:** lora |
| - **Base Model:** Wan Video 2.2 I2V-A14B |
| - **Version:** LowN |
| - **Upvotes:** 85.0 |
| - **Downloads:** 2927 |
| - **Size:** 585.1 MB |
| |
| |
| |
| ## Credits |
| |
| Original model by: YoReSpot |
| Civitai: [Link](https://civitai.com/models/2649114) |
| |
