<h1 id="kiko9-comfyui-wan-2.1-native-workflow-pem0fqldv">🧠 Kiko9 ComfyUI WAN 2.1 Native Workflow</h1>ComfyUI image-to-video (I2V) pipeline built around WAN 2.1 using native ComfyUI and Torch compilation (<code>torch.compile</code>) for performance gains. The design includes 2-pass generation, frame interpolation, upscaling, and slow motion — tailored for high-fidelity AI-enhanced video generation. <a rel="ugc" href="https://civarchive.com/models/1397307">Link</a> to <a rel="ugc" href="https://civarchive.com/models/1397307">workflow</a> I use for start <a rel="ugc" href="https://civarchive.com/models/1397307">image</a>: <hr /><h2 id="workflow-overview-10tvq216i">📦 Workflow Overview</h2><hr /><h2 id="project-breakdown-s2ss06hf6">🛠️ Project Breakdown</h2><h3 id="project-settings-tw365x4d1">🔧 Project Settings</h3><ul><li>Project File Path Generator: Allows saving outputs with a defined base path. Set this to your local output folder.<ul><li>✅ User Action: Update <code>root_path</code> to your preferred save location.</li></ul></li></ul><hr /><h3 id="aspect-ratio-logic-(don't-touch)-9rvptlk4t">🧮 Aspect Ratio Logic (Don't Touch)</h3><ul><li>Calculates <code>width</code> and <code>height</code> from image size using a float-to-int conversion for maintaining aspect ratio.<ul><li>⚠️ Do not modify unless you understand aspect ratio propagation.</li></ul></li></ul><hr /><h3 id="image-generation-for-video-(optimized-resolution)-3qhiw4cvt">📸 Image Generation for Video (Optimized Resolution)</h3><ul><li>When creating video frames using image generation tools like FLUX / SDXL, it's important to generate at the right resolution to maintain sharpness and consistency.</li></ul><h3 id="target-video-resolution-vc9fii3kl">🎯 Target Video Resolution</h3><ul><li>Target Size: <code>480x832</code></li><li>Aspect Ratio: <code>480 ÷ 832 ≈ 0.577</code></li></ul><h3 id="ideal-generation-resolution-44u2xd88h">✅ Ideal Generation Resolution</h3>To preserve details and allow for high-quality downscaling, generate at 2x or higher resolution. A perfect match in aspect ratio ensures you avoid cropping or distortion.Gen ResolutionAspect RatioNotes<code>960x1664960 ÷ 1664 ≈ 0.577</code>✅ Perfect aspect ratio match<code>1024x15361024 ÷ 1536 ≈ 0.6667</code>🔶 Slight crop or padding needed<h3 id="workflow-i1kmjvtb3">🔄 Workflow</h3><ol><li>Generate High-Res Images Use <code>960x1664</code> or larger with the same aspect ratio. Using FLUX, SDXL, etc.</li></ol><h3 id="why-this-works-yifvzr9do">🧮 Why This Works</h3><ul><li>High-res generation reduces artifacts and increases fidelity.</li><li>Downscaling averages pixels, smoothing jagged edges and noise.</li><li>Maintaining the same aspect ratio avoids warping or unnecessary padding.</li></ul><hr /><h3 id="loaders-8lgklpsh4">📥 Loaders</h3><ul><li>Load Checkpoint (WAN2.1): Load the WAN 2.1 native (ComfyUI) model checkpoint.</li><li>VAE &amp; CLIP Loader: Loads required VAE and CLIP encoders.</li><li>Power LoRA Loader (optional): For Power LoRa.</li><li>Tile Cache, Enhance, and CLIP Vision: Load auxiliary models.<ul><li>✅ User Action:<ul><li>Set <code>ckpt_name</code>, <code>vae_name</code>, and <code>clip_name</code> according to local model files.</li><li>Ensure files are in your configured ComfyUI model folders.</li></ul></li></ul></li></ul><hr /><h3 id="image-resize-016p38ye1">🖼️ Image / Resize</h3><ul><li>Load Image / Resize: Loads the input image or first frame from a video clip, resizes it to model-appropriate dimensions.</li></ul><hr /><h3 id="global-settings-0rfhc7tlz">🌍 Global Settings</h3><ul><li>CLIP Text Encode (Prompt &amp; Negative): Prompts for conditioning the model.<ul><li>✅ User Action: Customize these prompts per your subject/style.</li></ul></li><li>Seed Generator / Upscale Factor: Controls random seed and image scale-up.<ul><li>✅ User Action: Set <code>seed</code> for reproducibility or leave -1 for random.</li></ul></li></ul><hr /><h3 id="1st-pass-(initial-generation)-r432q552l">🔁 1st Pass (Initial Generation)</h3><ul><li>KSampler: Runs the initial inference.</li><li>VAE Decode &amp; Video Combine: Decodes latent space to image, combines with source.</li><li>Slow Motion / PlaySound: Optional audio sync and slow-mo settings.</li><li>Select last frame for 2nd pass start frame. (Pop Up window)</li></ul><hr /><h3 id="2nd-pass-(refine-and-extend)-jx881s2wv">🔁 2nd Pass (Refine &amp; Extend)</h3><ul><li>Similar to 1st Pass but optimized for longer inference or higher quality.</li><li>Take last frame from 1st pass as 2nd pass starting image.</li><li>Get Mask Range From Clip: Extracts mask regions for attention.</li><li>Image Batch Multi: Processes multiple frames simultaneously.</li></ul><hr /><h3 id="upscaling-and-frame-interpolation-suxdj29m2">📈 Upscaling &amp; Frame Interpolation</h3><ul><li>Image Sharpen / Restore Faces: Post-processing enhancements.</li><li>Upscale Image (Real-ESRGAN or similar).</li><li>Frame Interpolation (RIFE / FILM): Smooth transitions for higher FPS.</li><li>Slow Motion: Optional, adds frames and blends for cinematic slow-mo.</li></ul><hr /><h3 id="experimental-(optional-long-runtime)-tzgj6ycwj">🧪 Experimental (Optional, Long Runtime)</h3><ul><li>Advanced enhancement or second-stage denoising/refinement.</li><li>Useful for batch rendering with very high quality needs.<ul><li>⏱️ Warning: These steps significantly increase processing time.</li></ul></li></ul><hr /><h2 id="torch-compile-setup-(very-important)-6kue8tcxd">⚡ Torch Compile Setup (VERY IMPORTANT)</h2>To unlock native acceleration via <code>torch.compile</code>, ensure you meet these requirements:<h3 id="requirements-rht72ucuo">✅ Requirements</h3><ul><li>PyTorch 2.1+ with CUDA</li><li>NVIDIA GPU with Ampere or later architecture (RTX 30XX, 40XX)</li><li>Use latest nightly ComfyUI or manually apply <code>torch.compile()</code> patching.</li></ul><hr /><h2 id="saving-outputs-4qj2mmb98">💾 Saving Outputs</h2><ul><li>Controlled via Project Path Generator and Video Combine nodes.</li><li>Output format (e.g. <code>.mp4</code>, <code>.png</code>, <code>.webm</code>) should be explicitly set in <code>Video Combine</code>.</li></ul><hr /><h2 id="notes-q81vezyrx">📋 Notes</h2><ul><li>⚠️ First run of torch.compile will be slow due to graph tracing.</li><li>🧠 Prompt tuning is crucial for WAN 2.1 — try detailed descriptions.</li><li>⚠️ Not optimized for older machines.</li></ul><hr /><h2 id="faq-a4yxupugr">🙋 FAQ</h2>Q: My output is laggy or missing frames.<ul><li>Check interpolation settings and slow motion settings — disable one if not needed.</li></ul>Q: Workflow crashes during torch compile.<ul><li>Ensure you're using PyTorch 2.1+, and your GPU is Ampere or newer.</li></ul>Q: Can I use this with other models like SDXL?<ul><li>You can, but WAN 2.1 is optimized for this specific setup. Results may vary.</li></ul><hr /><h2 id="credits-mu8eqt5sg">📎 Credits</h2><ul><li>Workflow design by Kiko9</li><li>WAN 2.1</li><li>ComfyUI team for the powerful modular engine</li></ul><hr /><h2 id="folder-structure-example-7va7nnbcg">📂 Folder Structure Example</h2>ComfyUI/ ├── models/ │ ├── checkpoints/ │ ├── vae/ │ ├── clip/ ├── output/ │ └── generated/ ├── custom_nodes/ │<hr /><h3 id="end-to-end-wan-2.1-generation-summary-8gns2acjt">📊 End-to-End WAN 2.1 Generation Summary</h3>StepDescriptionTime / Count. ResolutionPrompt StartInitial prompt execution begins 92.95 secModel LoadLoaded WAN21 model weights ~15,952 msFirst Comfy-VFI PassGenerated frames with TeaCache initialized ~6 min 13sec 480x832Frames Generated (1st pass)Comfy-VFI output 231 frames 480x832Second Comfy-VFI PassRepeats generation with same steps ~6 min 28 sec 480x832Frames Generated (2nd pass)Comfy-VFI output(Implied 480x832WanVAE Load (1st)Loaded latent space model ~1220 ms —WanVAE Load (2nd)Loaded again for reuse ~1304 ms —Face Restoration (GFPGAN)GFPGANv1.4 restored images 152 frames 512x512Comfy-VFI Run (3rd)Generated additional frames ~unknown 960x1664 Frames Generated(3rd pass)Comfy-VFI output 456 frames 960x1664Comfy-VFI Run (4th)Final batch of generation~unknown 960x1664 Frames Generated(4th pass)Comfy-VFI output304 frames960x1664Prompt EndFinal step of pipeline 1050.60 sec—ℹ️ Notes:<ul><li>"TeaCache skipped" 12 conditional + 12 unconditional steps per 30 = ~20% optimization.</li><li>Face restoration step was applied to a subset (152 frames).</li><li>The 960x1664 resolution used in the last two passes matches the 480x832 aspect ratio perfectly, ideal for downscaling or 2x video output.</li></ul><h2 id="feedback-and-contributions-yg0kd5qaj">🗨️ Feedback &amp; Contributions</h2>Feel free to submit issues if you encounter bugs or want to contribute improvements.<hr />🔥 Happy rendering!

Files with SHA256: 74717423d5d4a8a91b4c157cf6b7a19bc10ea6d2ceaba3db865d9efd7f166d77