ComfyUI: Anima Base, Microsoft Lens and Pause Image setup

This ComfyUI tutorial covers Anima Base anime generation, Microsoft Lens setup and Pixaroma node updates. The presenter starts by updating ComfyUI and the custom nodes, then suggests clearing the browser cache if node buttons or previews look wrong.

Anima Base uses a Qwen text encoder and Qwen VAE alongside the diffusion model. The prompting section covers lowercase keyword tags, positive and negative prompts, and artist tags. A Gemma 4 workflow turns short natural-language requests into formatted prompts. Combining it with Anima requires more VRAM, according to the presenter. Changing the text-generation seed produces a new prompt; repeat runs can reuse the existing one.

Pause Image stops execution for review before upscaling or editing. Continue runs the downstream steps, while regenerating a different image requires a random seed. The tutorial also demonstrates Flash Attention installation and before-and-after upscale comparisons, with hardware support varying by computer.

Lens uses a BF16 diffusion model, GPT-OSS text encoder and Flux 2 VAE. The presenter says it needs more VRAM than Anima but gives no minimum. He cites support up to 1440 pixels, warns of possible duplication at larger sizes and notes that some workflows may not work with Flash Attention. The final examples use pause nodes to crop and resize without rerunning generation.