Learning Timeline
Key Insights
VRAM Requirements & Model Versions
The official model requires at least 24 GB of VRAM to run smoothly. However, you can use 'Quantized versions' which can run on PCs with specs as low as 8 GB of VRAM.
Image Quality Tip: Shift Value
If your images appear too dark, blurry, or of poor quality, try increasing the 'Shift' value. Conversely, decrease the 'Shift' value if you want to add more detail to the image.
Advantages of Open Source
Since Qwen Image is open-source, you can run it locally with unlimited usage and no subscription fees, provided your PC hardware is sufficient.
Step by Step
Installation Steps for Qwen Image Model Files
- Open the Hugging Face link provided in the description (Official Qwen Image repository).
- Click on the 'Split Files' section and enter the 'diffusion_models' folder.
- Download both diffusion model files found there and save them in the folder: 'ComfyUI/models/diffusion_models'.
- Download the 'Text Encoder' file (approximately 8.7 GB) from the provided link.
- Save the Text Encoder file in the folder: 'ComfyUI/models/text_encoders'.
- Download the 'VAE' file (approximately 242 MB) used for image encoding/decoding.
- Save the VAE file in the folder: 'ComfyUI/models/vae'.
- Open ComfyUI and click the 'Update' button in the interface menu. Wait for the process to complete and then click 'Close'.
- Click the 'Restart' button in ComfyUI to ensure all new models are detected by the system.
Workflow Setup and Image Generation
- Go to the official Qwen Image workflow page.
- Right-click the workflow download button and select 'Save link as' to save the JSON file to your PC.
- Open the ComfyUI interface in your browser, then drag and drop the downloaded workflow file directly into the ComfyUI workspace.
- On the model selection node, ensure you select the Qwen model you downloaded.
- On the VAE node, select the 'Quen image VAE' file from the dropdown menu.
- Set the 'Width' and 'Height' according to your desired image orientation (e.g., Horizontal).
- Set the 'Batch size' to 1 for your first run.
- In the 'KSampler' section, set the 'Seed' to 'Random' for image variety, or use a fixed number if you want identical results.
- Adjust the 'Steps' (typically 20-30). Use 10 steps if you want faster processing with a slight trade-off in quality.
- Set the 'CFG' value (Higher = follows the prompt more strictly, Lower = more creativity/variation).
- Select a 'Sampler Name' (e.g., Euler or SA Solver) based on your preference.
- Click the 'Queue Prompt' button (or Run) to start the image generation process.
