CivArchive
    ← All articles
    Published March 23, 2026by FloyoAI

    [GUIDE] Turn Lineart into Styled Illustrations with Qwen Image Edit 2511 in ComfyUI

    301 views0 reactions0 comments on CivitAI1 collected
    comfyuiqwenqwen 2511loraaiimg2imglineartstyle transferworkflowworkflowsimageeditting

    You drew the pose. The composition is right. Now you need colours, lighting, and a finished look without redrawing anything.

    This workflow feeds your sketch into Qwen Image Edit 2511 with dx8152's Style Transfer LoRA attached. Show it your lineart, a reference image in the style you want, and one prompt. It does the rest.
    Run this ready-made workflow now on Floyo without any installation!

    What you need

    Models (pre-loaded in the workflow):

    • qwen_image_edit_2511_bf16.safetensors

    • style-transfer-1_20.safetensors (the LoRA)

    • Qwen-Image-Edit-2511-Lightning-8steps-V1.0-fp32.safetensors

    VRAM: 16GB minimum. 24GB recommended.

    Custom nodes: ComfyUI_LayerStyle

    The 4 inputs that matter

    1. Your lineart → LoadImage node (Figure 1) Clean outlines work best. Heavy hatching or rough gestures will confuse the structure preservation. If your sketch is messy, clean it up first.

    2. Your style reference → LoadImage node (Figure 2) Any artwork whose colors, brushwork, or rendering feel you want to copy. Your own past piece, a painting, a screenshot. The model reads both images together.

    3. Your prompt The trigger phrase is style transfer. Use this exact format:

    style transfer. Change the style of Figure 1 to the style of Figure 2.

    Add style descriptors after if you want to push a direction: cel shading, vibrant colors or painterly, loose brushwork, warm light.

    4. LoRA strength → LoraLoaderModelOnly node Default: 1.0. Drop to 0.8 if the reference is overwhelming your sketch structure. Go to 1.2 if colors feel weak.

    Sampler settings (don't change these)

    The Lightning LoRA is calibrated for exactly 8 steps.

    • Steps: 8

    • CFG: 1.0

    • Sampler: euler

    Going higher doesn't add detail. It overcooked results.

    Two things to check in the output

    Is the pose matching your sketch? Qwen has good structure preservation, but complex occlusion (crossed arms, heavy foreshortening) can simplify. Simplify the lineart before input for tricky poses.

    Is the style matching your reference? If not, first check your prompt Figure 1 must match the slot holding your lineart, Figure 2 the style reference. Swapping the labels in the prompt fixes most style mismatches instantly.

    Best use

    Rapid iteration. Sketch 3–5 compositions roughly, run each through this workflow with the same style reference. You get 5 styled versions without rebuilding anything. Compare and pick.

    Attachments