Hi! I'm not an expert in workflows, but when I see the speed and quality that these Loras can achieve, I want to compare them with the TCD scheduler. Maybe it's a mess, but perhaps this will inspire or work for someone.<h2 id="heading-49">Notes:</h2><ul><li>To make this work, you'll need the loras with the number of steps you want to test:</li></ul>TCD Loras: <a target="_blank" rel="ugc" href="https://huggingface.co/h1t/TCD-SDXL-LoRA/tree/main">TCD LORAS: h1t/TCD-SDXL-LoRA at main (</a><a target="_blank" rel="ugc" href="http://huggingface.co">huggingface.co</a><a target="_blank" rel="ugc" href="https://huggingface.co/h1t/TCD-SDXL-LoRA/tree/main">)</a>HyperSD Loras: <a target="_blank" rel="ugc" href="https://huggingface.co/ByteDance/Hyper-SD/tree/main">HYPERSD LORAS: ByteDance/Hyper-SD at main (</a><a target="_blank" rel="ugc" href="http://huggingface.co">huggingface.co</a><a target="_blank" rel="ugc" href="https://huggingface.co/ByteDance/Hyper-SD/tree/main">)</a>Lighting Loras: <a target="_blank" rel="ugc" href="https://huggingface.co/ByteDance/SDXL-Lightning/tree/main">ByteDance/SDXL-Lightning at main (</a><a target="_blank" rel="ugc" href="http://huggingface.co">huggingface.co</a><a target="_blank" rel="ugc" href="https://huggingface.co/ByteDance/SDXL-Lightning/tree/main">)</a>LCM Loras: <a target="_blank" rel="ugc" href="https://huggingface.co/latent-consistency/lcm-lora-sdxl/tree/main">latent-consistency/lcm-lora-sdxl at main (</a><a target="_blank" rel="ugc" href="http://huggingface.co">huggingface.co</a><a target="_blank" rel="ugc" href="https://huggingface.co/latent-consistency/lcm-lora-sdxl/tree/main">)</a><ul><li>For the nodes, use the Comfyui Manager and "install missing custom nodes".</li><li>You can play with the strength of the Loras and the sigma (sigma 1 for less details, sigma 0 for a burned image. I think 0.3-0.6 is good).</li></ul><h2 id="heading-50">Conclusion</h2>In my test, I think TCD and "1step Hypersd" Lora were the best. For the "1step Hypersd" Lora, I used 5 or 8 steps. With one step, I only get a single color image.

Files with SHA256: a19ce1a2c57bafd9cc2a72e907de9f0de53b2c23b51fa23d02fee74fae4b397a