Learning Timeline
Key Insights
Optimizing Generation Speed
Using the 'Lightning' LoRA allows you to generate images in just 4 steps instead of the usual 20. This speeds up the process by 5 times without drastically compromising quality.
Important CFG Note
When using the 'Lightning' workflow, ensure the CFG is set to 1.0. If the CFG value is too high, the resulting image may appear distorted or contain strange artifacts.
Prompts
Image Edit Command Example
Target:
Qwen Image Edit Prompt Node
turn this car blue
Step by Step
Folder and Model Setup
- Download the 'Quen imageedit diffusion model' (19 GB) and save it in the 'ComfyUI/models/diffusion_models' folder.
- Download the 'Quen image lightning four steps LoRA' (1.6 GB) and save it in the 'ComfyUI/models/loras' folder.
- Download the 'Text Encoder file' (9 GB) and save it in the 'ComfyUI/models/text_encoders' folder.
- Download the 'Quen image VAE' and save it in the 'ComfyUI/models/vae' folder.
- Ensure you have enough storage space as the total file size reaches nearly 30 GB.
Workflow Update and Import
- Open your ComfyUI interface.
- Click the 'Manager' button on the menu bar.
- Click the 'Update ComfyUI' button to ensure your system is running the latest version.
- Click 'Close' once the update is finished, then click 'Restart' on the ComfyUI interface.
- Obtain the official workflow JSON file (Right-click the workflow link and select 'Save link as').
- Drag and drop the downloaded JSON file directly into the ComfyUI interface.
Node Configuration and AI Parameters
- On the 'Load Diffusion Model' node, select the 'Qwen2.5-VL' model.
- On the 'VAE' node, select 'Quen image VAE'.
- Locate the purple LoRA node and press 'Ctrl + B' on your keyboard to 'unbypass' (activate) the node.
- In that LoRA node, select 'Quen image lightning four steps' from the dropdown menu.
- Go to the 'KSampler' node and change the 'Steps' value to '4'.
- Change the 'CFG' value to '1.0' (Mandatory for Lightning model usage).
- Set 'Seed' to 'randomize' if you want a different result for every run.
- Upload the image you want to edit into the image input node.
- Type your editing instructions in the text prompt (e.g., 'turn this car blue').
- Click the 'Queue Prompt' button to start the generation process.
