SparseCtrl Animation Comfyui Workflow Guide

This AnimateDiff LCM Animation Workflow is designed to make subtle animations easy and compatible with low VRAM setups. This workflow integrates a range of models, control nets, and custom nodes to provide refined, high-quality animations using Stable Diffusion.

Getting Started: Initial Setup and Downloads

You’ll need several model files and custom nodes to replicate this setup. Here’s a list to get started:

1. Recommended Checkpoints: AniVerse - V5.0, IMP, PerfectDeliberate and PhotonLCM high-quality Stable Diffusion checkpoints.

2. AnimateDiff Motion Model & LoRA: AnimateLCM by wangfuyun – Includes both the motion model and LoRA for smooth frame-by-frame animations.

3. Advanced Control Nets: Advanced ControlNet by Kosinkadink – Adds scheduling, masking capabilities, and sliding context support for refined animation control.

4. SparseCtrl Control Net: SparseCtrl – Provides sparse control for precise motion guidance.

5. QRcode Monster Control Net: QRcode Monster – this is responsible for the morphing animations.

6. Clip Vision Model: Clip Vision is essential for enhancing image conditioning.

7. Upscaler Model: 4x NMKD Siax Upscaler – Used for the final upscale pass to ensure your animation looks sharp and high-quality.

Tip: Make sure to place these models in their appropriate directories within ComfyUI (e.g., ./models/checkpoints, ./models/controlnets, ./models/upscalers).

Input Steps for Creating Animations

Configure the Initial Setup:

• Checkpoint: Select your Stable Diffusion checkpoint, such as AniVerse V5.0.

• Required LoRA: AnimateLCM_sd15_t2v_lora @ strength of 0.5 - 1.0

• Optional LoRAs: Include any LoRA models you’d like to enhance specific features.

• VAE, Seed, Prompt: Set your VAE for image quality, choose a seed for consistency

• Models for Additional Control: Select the CLIP Vision and SAM models for added conditioning and segmentation used by the detailer nodes.Craft Prompt:

• Simple is better. You can use prompt travel to change the prompt on specific animation frames.

• Prepare the Input Image:

  • Resize and Adjust: Upload an input image that serves as a base for the animation. Resize it to ensure compatibility with IPAdapter and SparseCtrl Control Net.

  • Quality Check: Optimizing the image dimensions helps IPAdapter and SparseCtrl process it more effectively, creating smoother animations.

    (tip: use right-click "queue group output nodes (rgthree)" to view your adjustments to the image resize and crop before queuing up the entire workflow.

• AnimateDiff Configuration:

  • AnimateDiff Model: Load the AnimateDiff LCM model, specifically designed for seamless motion.

  • Set the Scheduler and Looping: Use the beta scheduler to control frame variation and enable looping for continuous animation. (models are trained on a context_length of 16 and a context_overlap of 4. You can adjust it, but your results may vary). Select if you want a loop or not with closed_loop.

• Control Net Configuration:

  • Adjust Control Net Strength: The SparseCtrl and QRcode Monster control nets should work well with default settings, but adjusting the strength can fine-tune the animation’s motion.

  • Optional Customization: To influence the animation’s style more, experiment with the control net parameters to match your desired effect.

---

Processing Steps for Refining Animation Quality

1. Initial Rough Sample:

   • This step creates a rough animation pass to establish the movement, though it may appear choppy initially. It’s a foundational pass intended for further refinement.

2. Latent Upscale and Resampling (Highres Fix):

   • Apply a high-resolution fix to enhance image quality. This latent upscale reduces artifacts and improves clarity, particularly useful for animations generated on lower VRAM setups.

3. Face and Eye Detailing:

   • Use the face and eye detailer to reduce common artifacts for close-up animations. This optional step can refine facial details for smoother, more natural animations.

   • If you experience VRAM limitations, consider bypassing this step to optimize processing speed.

4. Final Upscale and Interpolation to 30 FPS:

   • The final upscale pass uses the 4x NMKD Siax Upscaler model to enhance the resolution.

   • 30 FPS Interpolation: This final step smooths motion by interpolating frames, delivering a polished animation with fluid motion at 30 frames per second.

---

Tips and Recommendations

• Experiment with Settings: Adjusting parameters like checkpoint model, image size, control net strength, and IPAdapter configuration can significantly impact animation quality.

• Handling VRAM Constraints: This workflow can work well even on lower VRAM systems. If you encounter memory issues, try simplifying by bypassing the face detailer step or reducing image resolution.

• Custom Nodes: This setup uses several custom nodes in ComfyUI’s manager, including MTB, Efficiency, AnimateDiff, RIFE, and Face Detailer. Ensure all custom nodes are installed and up-to-date.

Final Notes

This SparseCtrl Animation Workflow offers a straightforward way to create expressive animations, combining powerful movement, quality, and detail tools. If you’re new to Stable Diffusion animations, this guide provides a solid starting point.

Enjoy experimenting, and don’t hesitate to share feedback or troubleshooting questions.

Posting images to this page is appreciated for the buzz ⚡⚡⚡?

描述:

V1 includes - IPAdapter, SparseNet control net, AnimateDiff, and face detailer

训练词语: