This workflow is a streamlined Video2Image2Video to create live portraits. You can cartoonify yourself or use videos to animate your favorite characters! I'll include a .json file of the workflow and an example copyright-free video for you to follow.Here’s a breakdown of each step:<h1 id="step-1:-load-the-video-and-get-the-first-frame-c0k3b2877">Step 1: Load the Video and Get the First Frame</h1>To start, you'll need to grab the first frame from your video, which will be the image to which the nodes map all the facial points for the animation.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/d9bfa6bd-eed7-4b3c-9c15-4faf9df8b3b8/width=525/d9bfa6bd-eed7-4b3c-9c15-4faf9df8b3b8.jpeg" /><ul><li>Bypass all other groups in the workflow</li><li>Load the video and queue up your prompt to start.</li><li>Copy the first frame to your clipspace once it's loaded. <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/4fa3fdc5-024b-43cf-adef-ebaba135c073/width=525/4fa3fdc5-024b-43cf-adef-ebaba135c073.jpeg" /></li><li>If your video requires audio, connect the audio output to the corresponding node.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/f6161dd6-3a91-4b3c-8684-1eaafdba72e2/width=525/f6161dd6-3a91-4b3c-8684-1eaafdba72e2.jpeg" /></li><li>After copying the frame, bypass this node and move to step 2.</li></ul><h1 id="step-2:-generate-the-character-or-face-using-image-to-image-(i2i)-d5yo8vuk4">Step 2: Generate the Character or Face Using Image-to-Image (I2I)</h1>Now, it's time to create the character or face you want to animate.<ul><li>Bypass all other groups except for this one.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/1073ff6e-278b-4bba-a3f3-5fba7f2b8783/width=525/1073ff6e-278b-4bba-a3f3-5fba7f2b8783.jpeg" /></li><li>Paste the first frame from your clipboard into the load image node. (you can stretch a pipe with the image to the upscale image node if you prefer.)<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3cf0ffe8-7dbb-4dcc-833b-bd6948eac040/width=525/3cf0ffe8-7dbb-4dcc-833b-bd6948eac040.jpeg" /></li><li>Choose a ControlNet model that ensures the face's position stays consistent throughout the video (OpenPose and LineArt work well here). Adjust the ControlNet strength and end_percent and the scheduler denoise to achieve the image you like. The basic guidance is turning the denoise down, and the ControlNet strength and end_percent up makes the input image influence the final result more. <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/52937563-7b72-4370-89e6-e7a8b842f729/width=525/52937563-7b72-4370-89e6-e7a8b842f729.jpeg" /></li><li>Run image-to-image multiple times to adjust and fine-tune the facial features and animation style until you’re happy with it. I like using the style selector from the Easy-use custom nodes pack to capture styles easily. (I highly recommend this checkpoint by 7whitefire7 for cartoon styles. It is amazing! <a target="_blank" rel="ugc" href="https://civarchive.com/models/125907/realcartoon-xl">RealCartoon-XL - V7 | Stable Diffusion XL Checkpoint | Civitai)</a><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/35eee3cd-cd22-42d1-8753-57abcbf85f6b/width=525/35eee3cd-cd22-42d1-8753-57abcbf85f6b.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/4c533273-5789-4042-a412-5d69736cd9b8/width=525/4c533273-5789-4042-a412-5d69736cd9b8.jpeg" /></li><li>If needed, you can copy the output image, paste it back into the load image node, and run it again to refine details further while keeping the size and framing intact.</li><li><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/4fb5a1ef-601e-471c-8370-ad82589d2c34/width=525/4fb5a1ef-601e-471c-8370-ad82589d2c34.jpeg" />Once you're satisfied, copy the I2I output to clip space and move on to step 3. (again, you can connect the pipe from the VAE decode node to the step 3 upscale image node if you prefer.)<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3e2731f6-8136-4e7d-9f3e-e963c293fa4e/width=525/3e2731f6-8136-4e7d-9f3e-e963c293fa4e.jpeg" /></li></ul>Pro Tip: A 1:1 ratio works best with live portrait nodes, so keep that in mind while generating.<h1 id="step-3:-match-the-first-frame-for-best-face-mapping-y84n8fy2m">Step 3: Match the First Frame for Best Face Mapping</h1>This step ensures that your animation starts in sync with the first frame of the video.<ul><li>Bypass groups 2 and 4 .<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/f195d6e4-712e-450f-85b0-b701dc239c67/width=525/f195d6e4-712e-450f-85b0-b701dc239c67.jpeg" /></li><li>Paste the I2I output image from step 2 into the load image node.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/e1eb2dbe-8a40-4ece-9e21-0a2a97f048e2/width=525/e1eb2dbe-8a40-4ece-9e21-0a2a97f048e2.jpeg" /></li><li>Queue the prompt to match the first frame's expression and head position. This is where the tracking points will be mapped. In this example, I did not get a good map. I will turn down the sample_ratio input so that sample frame 1 has less influence. <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/168dc2b7-b001-4644-a889-ff31df06db7f/width=525/168dc2b7-b001-4644-a889-ff31df06db7f.jpeg" /></li><li>Adjust expressions if needed (e.g., increase "aaa" for an open mouth or "smile" if the character is smiling). In this example, I used rotate_pitch to turn her head down a little, adjusted me and woo to change her mouth shape and turned the sample_ratio down. <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/c54533dc-a23b-4dc0-83e5-6971b1983d38/width=525/c54533dc-a23b-4dc0-83e5-6971b1983d38.jpeg" /></li><li>Ensure auto-queuing is enabled or queue up the prompt manually to see the expression changes in the expression editor node.</li><li>Adjust the crop factor input to zoom in 2.0 is no crop, and 1.0 is a max crop. Match this setting in the advanced live portrait node in step 4.</li><li>Once the face mapping is set, move on to the final step.</li></ul><h1 id="step-4:-generate-the-final-video-interpolate-and-save-kpwh5k86x">Step 4: Generate the Final Video, Interpolate, and Save</h1>This is where it all comes together!<ul><li>Bypass step 2, and make sure steps 1, 3, and 4 are active.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/2765451a-053f-4ee8-ac72-abbd0b886671/width=525/2765451a-053f-4ee8-ac72-abbd0b886671.jpeg" /></li><li>Check if you're including the audio, ensuring it's connected properly.</li><li>Queue up your prompt, and the workflow will handle your face mapping and interpolation.</li><li>Once everything looks good, save your final video!<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/2b0d1db2-d1d8-4c0d-af97-2feccf1dca30/width=525/2b0d1db2-d1d8-4c0d-af97-2feccf1dca30.jpeg" /></li></ul><h3 id="wrapping-up-2v1ewmp6r">Wrapping Up</h3>That’s it! This simple four-step workflow makes using a video to guide an animated portrait easily.If you run into any issues or have questions, comment, and I’ll be happy to help!Workflow Author: OpalSky Resources:<ul><li><a target="_new" rel="ugc" href="https://github.com/PowerHouseMan/ComfyUI-AdvancedLivePortrait">ComfyUI Advanced LivePortrait</a></li><li><a target="_new" rel="ugc" href="https://github.com/KwaiVGI/LivePortrait">LivePortrait</a></li><li><a target="_new" rel="ugc" href="https://github.com/kijai/ComfyUI-LivePortraitKJ">ComfyUI LivePortraitKJ</a></li></ul>Enjoy creating your live portraits, and I can't wait to see what you come up with!

Files with SHA256: 1c5ca0c22df99cd2444432b3584a5c10561f331e028c2318d077328d896cbd63