ComfyUI: Generate and upscale images using Zee Image and PiD

Step-by-step

1. Drag and drop the workflow file onto the ComfyUI interface
2. Download the Zee Image Turbo model and place it in 'models/diffusion_models'
3. Download the Qwen 2 1.5B or 7B clip model and place it in 'models/text_encoders'
4. Press 'R' to refresh the model list and select the downloaded models in the dropdowns
5. Select Gemma 2 for the text encoder and the PiD Flux.1 version for the diffusion model in the upscaler section
6. Enter your prompt in the green text box
7. Set the longest side of the resolution to 1024 to match the PiD upscaler requirements
8. Configure the K-Sampler settings and click 'Run'

Options

• Choose between BF16 (12GB) or FP4 (compressed) Zee Image Turbo model variants
• Select between full, FP8, or FP4 versions of the Qwen text encoder
• Adjust the batch size to generate multiple images at once
• Modify the seed, steps, CFG, and sampler algorithm in the K-Sampler

Watch out for

• The FP4 model variant requires a newer CUDA GPU
• The PiD upscaler requires the original image's longest side to be exactly 1024 to output 4K correctly

Tips

• Use 7 to 9 steps for Zee Image Turbo as it is the 'sweet spot'
• Keep the CFG at 1 for Zee Image generations
• Leave the additional micro settings in the upscaler step at their default values