This workflow enables image-to-video (I2V) and text-to-video (T2V) generation using Wan2.1 i2v-14B and t2v-1.3B with LoRA support. It is optimized with Torch Compile, Context Options, and TeaCache for efficiency.

Workflow Structure

1️⃣ Text & Image Embeddings: CLIP-based embeddings for text and image conditioning.
2️⃣ Model Loading & Compilation: Uses torch.compile() for speed optimization.
3️⃣ TeaCache Integration: Optimizes latent caching for faster generation.
4️⃣ Context Options: Ensures smooth motion and temporal consistency.
5️⃣ Video Sampling: Processes frames with unipc scheduler and blending techniques.
6️⃣ Output Encoding: Frames are merged into an H.264 MP4 video, with audio sync support.

Key Features

✅ Torch Compile for faster inference
✅ TeaCache for memory optimization
✅ Context-aware sampling for smooth motion
✅ LoRA fine-tuning for custom styles
✅ Built-in Wiki for parameter reference

Best Practices

• Increase context_frames & overlap for smoother motion.

• Reduce noise_aug_strength for cleaner images.

• Enable force_offload to optimize VRAM.

• Use Torch Compile (inductor) for speed improvements.

This workflow is optimized for high-quality, efficient video generation with temporal consistency and memory efficiency. 🚀