Learning Timeline
Key Insights
Benefits of Character Consistency
Flux 2 Klein is highly effective at maintaining character consistency, keeping the subject's original features intact even after changing their clothing or background.
Image Resolution Tips
By default, the workflow may perform an upscale. Use 'Ctrl + B' on the upscale node if you want the output size to match the original dimensions of the uploaded image.
VRAM Performance
The generation process is very fast; it takes about 15 seconds to load the model into memory and only 4-7 seconds to produce the final image on a 16GB VRAM GPU.
Prompts
Single Image Modification
Target:
Flux 2 Klein
turn her hanfu to red
Single Image Color Change
Target:
Flux 2 Klein
make her hanfu black
Merging Two Subjects
Target:
Flux 2 Klein
the person in image one and the person in image two are having dinner at a fancy restaurant
Step by Step
How to Edit a Single Image
- Open the ComfyUI interface and load the downloaded Flux 2 Klein workflow file.
- Ensure 'Option 1' (Single Input) is activated and bypass 'Option 2' if necessary.
- Click the 'Upload Image' node and select the image you wish to modify from your computer.
- Enter your modification instructions in the text prompt area (e.g., 'turn her hanfu to red').
- Select the 'Flux 2 Klein 4B' model from the model selection dropdown menu.
- Select 'Quen 3 text encoder' from the text encoder dropdown menu.
- Select 'Flux 2 VAE' from the VAE dropdown menu.
- To keep the original input image resolution without upscaling, click the megapixels/upscale node and press 'Ctrl + B' to bypass it.
- Click the 'Run' or 'Queue Prompt' button to start the image generation process.
- Check the final result in the 'Save Image' node or look inside the ComfyUI 'output' folder.
How to Merge Two Reference Images
- Activate 'Option 2' in the workflow by selecting the relevant nodes and disabling their bypass status.
- Upload the first reference image into the 'Image 1' slot.
- Upload the second reference image into the 'Image 2' slot.
- Write a prompt that combines both subjects (e.g., 'the person in image one and the person in image two are having dinner').
- Choose the FP4 version of the text encoder model if you want to save VRAM or speed up the process.
- To set the image size manually, click 'Expand' on the relevant subgraph.
- Delete the existing connections for the width and height inputs.
- Manually enter your desired 'Width' and 'Height' values (e.g., 1280x720).
- Adjust 'Sampler' settings, such as the 'CFG' value, to determine how literally the AI should follow the prompt.
- Click the 'Run' button to generate the merged image.
