InvokeAI and ComfyUI: local image workflows and editing

This tutorial introduces InvokeAI and ComfyUI as ways to run models locally for image generation and editing on a PC or Mac. The speaker compares Invoke's layer-based workspace to Photoshop and explains its text prompts, reference images, and style-reference options.

The InvokeAI section also covers downloading Krea 2. According to the speaker, this gated model requires a Hugging Face account, a user access token, and an access request. That requirement does not apply to every model discussed. The speaker describes the models as free and open source, but the tutorial does not establish their individual licensing terms.

ComfyUI uses templates and a node workflow. The walkthrough focuses on image editing, though the speaker also shows template categories for video, audio, 3D, and LLMs. For GPUs with less VRAM, he recommends the 4B image-editing and text-to-image options; he gives no minimum hardware specification.

The editing example starts with a workflow that accepts up to two reference images. The speaker removes the second image input and asks for long blonde hair and a Viking outfit. After resolving missing image inputs, he reports a completed result, while noting that the face does not look exactly like him. He prefers ComfyUI because he finds it more stable than Invoke, a judgment based on his own use.