ComfyUI: build a text-to-image workflow with LoRAs

This beginner tutorial builds a ComfyUI text-to-image workflow from a blank canvas. The presenter uses Juggernaut XL, an SDXL checkpoint downloaded from Civitai, and shows where to place the file in ComfyUI's models/checkpoints folder. He reports running the demonstration with 8 GB VRAM; this is his setup, rather than a stated minimum requirement.

The walkthrough connects the checkpoint to positive and negative CLIP text prompt nodes, then adds KSampler and an Empty Latent Image node for dimensions and batch size. VAE Decode converts the sampled latent output into an image, which goes to a preview or save node. An apple-on-a-table prompt provides the first test.

Later examples cover seeds, prompt changes and image dimensions. The presenter selects DPM++ 2M with the Karras scheduler and compares the results with earlier generations. Larger dimensions take longer in his demonstration, and he notes that wide images can produce awkward compositions with the SDXL model he uses.

The LoRA section covers downloading files, placing them in the models folder, refreshing node choices and connecting loaders. A final detour uses the Pixelcut Uncrop website to extend a square image into a wider format. That step takes place outside the local AI workflow. ControlNet and a full upscaling workflow are reserved for later tutorials.