This is a stable, 2-stage workflow designed to generate a base image using Qwen models and transform it into a clean 2D Anime style using SDXL (Animerge) without saturating your VRAM.
Why use this workflow?
Standard LoRA loading on Qwen models often causes immediate VRAM spikes and Out of Memory (OOM) crashes. This workflow bypasses the technical limitation by splitting the generation into two consecutive, isolated stages inside blue grouped boxes:
1. Stage 1 (Top Group): Generates the initial photorealistic image using Qwen, focusing on anatomy, pose, and background composition.
2. Stage 2 (Bottom Group): Automatically unloads Qwen from memory, takes the generated image, and passes it to an SDXL Anime model (Animerge) via VAE Encode to complete a flawless, high-quality style transfer using Ultimate SD Upscale.
Quick Setup:
- Set the denoise on the second stage sampler/upscaler between 0.45 and 0.60 depending on how much 2D anime style enforcement you want.
- Keep the CFG around 5.0 - 7.0 for optimal color saturation on the anime pass.



