Learning Timeline
Key Insights
Turbo vs. Full Model Tips
The Turbo model is significantly faster, requiring only 8 steps to produce high-quality images, whereas the Full model requires a much higher step count.
Dynamic Allocation Feature
ComfyUI features 'dynamic memory allocation', which allows you to run large models like Ernie Image (16GB) even if you don't have very high VRAM.
Creativity Control (CFG Scale)
The CFG value determines how literally the AI follows your prompt. Higher values make the AI more literal, while lower values allow for more creativity. For Ernie Image, a value of 1.0 is optimal.
Step by Step
ComfyUI Update and Initial Setup
- Open your main ComfyUI installation folder.
- Locate and open the folder named 'update'.
- Double-click the 'update_comfy.bat' file to begin updating to the latest version.
- Once finished, return to the main folder and double-click the 'run.bat' (or 'run_nvidia_gpu.bat') file to launch ComfyUI.
- Within the ComfyUI interface, click the 'Load' button and select the Ernie Image or Ernie Image Turbo workflow.
How to Download and Install Models
- If you see red errors (missing models), click on the left sidebar and select the 'Download' button for the detected missing models to download them automatically.
- To download manually: Download the 'Ernie Image Turbo' model and place it in the 'ComfyUI/models/diffusion_models' folder.
- Download 'Mistral 3B Text Encoder' and place it in the 'ComfyUI/models/clip' folder (or the specific model folder designated in the workflow).
- Download 'Flux 2 VAE' and place it in the 'ComfyUI/models/vae' folder.
- Once all files have finished downloading, press 'R' on your keyboard to Refresh the model list in ComfyUI.
Workflow Configuration and Image Generation
- In the loader node, select 'Ernie Image Turbo' (or the full version) in the model section.
- In the CLIP name section, select the 'Mistral 3B' model.
- In the VAE section, select 'Flux 2 VAE'.
- Type your image description in the 'Prompt' box.
- Set the 'Width' and 'Height' values according to your desired image size.
- Expand the 'K Sampler' node for advanced settings: Set 'Steps' to '8' for the Turbo model, and set 'CFG' to '1.0'.
- Click the 'Queue Prompt' button to start generating. The finished image will be automatically saved in the 'ComfyUI/output' folder.