Workflow Introduction

First, use the RMBG cutout node and the mask cropping node to preprocess the main image and text image. Utilize the COMPOSITOR V3 node to freely adjust the size and position of the main subject on the canvas by scrolling the mouse wheel and dragging the mouse for visual composition. Finally, redraw the image structure using FLUX CONTROLNET. This workflow is highly suitable for poster layout design, similar to the combination of KRITA, PS, and COMFYUI for texture redrawing.

Due to the absence of the COMPOSITOR V3 plugin on LIBLIB, users in need can first download the workflow locally for use. The installation method for the COMPOSITOR V3 plugin is as follows:

1. https://github.com/erosDiffusion/ComfyUI enricos nodes.git Install via the COMFYUI Node Manager GIT URL, then restart COMFYUI;

2. Search for erosDiffusion in the COMFYUI Node Manager, click install, and restart COMFYUI;

Regarding Model Selection and Parameter Adjustment

1. Large model selection: FLUX1 DEV FP8;

2. CONTROLNET selection: Choose FLUX's union, with the recommended types being depth or canny/lineart/anime lineart/mlsd (currently, the two types with the best tested results). CONTROLNET application parameters: intensity is recommended at 0.6–0.7, and the end time is recommended at 0.6 (extensive testing has shown that these parameters within the recommended range maintain higher similarity to the original image).

3. LORA models: Choose xlabs flux/FLUX photo ora V6 to enhance detail realism.safetensors;

4. Prompt words should be simple descriptions based on the original image, including information about the background image and the main subject image;

Tips for Using the COMPOSITOR V3 Node

1. image1 is the bottom layer, and we usually assign the background image here. image2 is the second layer, and so on. When adjusted, the layer jumps to the top layer and automatically returns after adjustment.

2. Duplicate multiple subjects. For example, connect Product A sequentially to image2, 3, and 4 in the nodes, and there will be three instances of Product A on the canvas.

3. You can use the text image node to add text to the canvas, but its recognition of Chinese is not very high, so English is recommended. Text font and color can be explored; the default is white.

4. The original image must go through cutout processing, then mask cropping, and then be cut out again. This ensures that the final subject on the canvas fits the border perfectly without extra parts; otherwise, it will be difficult to adjust multiple layers on the canvas.

5. After loading the image, click the execute queue button once. Once the image appears on the canvas, make adjustments. After adjustments, click the continue button at the bottom of the canvas to execute subsequent redrawing. For further canvas adjustments or parameter tweaks, use the continue button instead of the execute queue button. Of course, clicking the execute queue button is also an option, but it requires clicking twice.