This workflow is designed specifically for generating high-quality anime-style images, with a focus on Illustrious-based checkpoints.
📦 Required Custom Nodes 📦
Make sure you have these installed before loading the workflow:
Impact-Pack (core)
Impact-Subpack (additional SEGS tools)
comfyui_ultimatesdupscale (for Ultimate SD Upscale)
comfyui-ppm (for advanced prompt handling)
nd-super-nodes (for easy LoRA loading)
I recommend using ComfyUI Manager to install them easily.
⚠️⚠️ VRAM & Performance – Important! ⚠️⚠️
This workflow was optimised for 12 GB VRAM and more (tested on RTX 3060 12 GB).
If you're running on 8 GB VRAM, you'll need to tweak a few settings to avoid OOM errors – but the quality will stay the same.
Ultimate SD Upscale - tile_width/tile_height: 512 (instead of default)
batch_size: 1
Detailer SEGS - guide_size: 512 (instead of 1024)
max_size: 1024
noise_mask_feather: 8–10
⏱️ Generation Times (estimated) ⏱️
12 GB VRAM (default settings) – about 5–6 minutes per image
8 GB VRAM (with above tweaks) – slightly longer, but still within 7–8 minutes (your mileage may vary depending on the GPU).
Feel free to experiment – these are just starting points. You can push tile sizes higher if you have more memory, or lower them further if you're on a tight budget.
Happy generating! :)
Description
First upload, waiting for comments!
Comments (3)
How can I adjust the image size to get the composition I want?
Sup! You need the "Generation" group, it's marked in blue. At the very bottom of this block is the "Empty Latent Image" node where you can change the size. I don't recommend setting large values cuz this can slow down generation. Btw, below this node I added a note with pre-defined optimal sizes that you can change to suit your needs. I hope this helps :)
@mestari I saw it, I'll try again. Thank you. :)