<h2 id="inference">Inference</h2>NOTE: Use it only on your high noise samplerIMPORTANT: This LoRA is very sensitive to the prompt structure. There are certain phrases that I strongly recommend including in your prompt. Like:<ul><li>"Sequence of quick scene transitions of"</li><li>"The sequence of changes includes the transitions"</li></ul>You can find prompt examples in the attached videos.For inference I used ComfyUI.Trigger words: Banostasis concept, (and the phrases above)Strength: 0.8-1.0<h2 id="trainig-details">Trainig details</h2>Trained only on videos.<h3 id="high-noise-lora">HIGH noise LoRA</h3><ul><li>dataset: 33 videos 480x270 25,33,65,81 frame videos</li><li>steps: 15147</li><li>LR: 5e-5</li><li>optimizer: AdamW Optimi</li><li>rank: 32</li><li>batch size: 1</li><li>gradient accumulation steps: 1</li><li>min_t = 0.875</li><li>max_t = 1</li></ul>For training I used the diffusion-pipe repo.

Files with SHA256: 3b260c673225da6a0dcdb55fa7bd300019126e035c60dad9acdd2b5f25827516