<h1 id="this-is-a-workflow-for-sdxl(noobai-xlillustrious-xl)-and-anima.">This is a workflow for SDXL(NoobAI XL/Illustrious-XL) and Anima.</h1><h3 id="last-update:-anima-turbo-lora-+-nag">Last update: Anima Turbo LoRA + NAG</h3><h2 id="anima-turbo-lora-+-nag">Anima Turbo LoRA + NAG</h2>By using <a target="_blank" rel="ugc" href="https://civarchive.com/models/2560840/">Anima Turbo LoRA</a>, you can use Negative Prompts with NAG while maintaining low steps and CFG 1.Since applying NAG reduces generation speed, when generating with the 8–12 steps recommended for Anima Turbo LoRA, it is recommended to set it to approximately half of the total steps, such as start_percent=0 and end_percent=0.5.<h2 id="anima-couple">Anima Couple</h2><ul><li>v1</li></ul>The mask is split to draw the characters separately. If a single LoRA is trained on multiple characters, it can distinguish between them, but it does not support combining multiple character LoRAs.Since Anima generation takes time, this uses a speed-up node (Accelerate subgraph). In exchange for faster generation, some degradation may occur. Adjust the parameters as needed.<ul><li>v2</li></ul>Support for applying multiple character LoRAs has been added.<h2 id="anima">Anima</h2>The processing order is: generation → Hires.fix → face-only Hires.fix → hands-only Hires.fix → upscale.<h2 id="anima-xy-plot-checkpoint">Anima XY Plot CheckPoint</h2>This is a workflow for comparing Anima-based checkpoints.<h2 id="require">Require</h2><ul><li>SageAttention</li></ul><a target="_blank" rel="ugc" href="https://github.com/woct0rdho/SageAttention/releases">here</a>Please download and install the appropriate version for your environment.<ul><li>triton-windows</li></ul>ex)<pre><code>pip install triton-windows</code></pre><ul><li>sam3</li></ul><a target="_blank" rel="ugc" href="https://huggingface.co/facebook/sam3">v3.0 here</a><a target="_blank" rel="ugc" href="https://huggingface.co/facebook/sam3.1">v3.1 here</a><ul><li>my custom nodes</li></ul><a target="_blank" rel="ugc" href="https://github.com/hybskgks28275/ComfyUI-hybs-nodes">https://github.com/hybskgks28275/ComfyUI-hybs-nodes</a><a target="_blank" rel="ugc" href="https://github.com/hybskgks28275/ComfyUI-Anima-NAG">https://github.com/hybskgks28275/ComfyUI-Anima-NAG</a><ul><li>Build Tools for Visual Studio</li></ul>If you enable torch.compile, please install triton-windows.Also, install Build Tools for Visual Studio and make sure that cl.exe is added to your system PATH.

Workflow for SDXL (NoobAI XL/Illustrious-XL) and Anima

Improved generation results when using multiple LoRAs.While generation speed may decrease, it allows you to limit the scope of application when using multiple LoRAs.

Files with SHA256: e19381e9e3488fb9d5aeae314099368ad4d924fd895b65bc3a6462fecf709cf4