Which base model should I use with these Krea 2 LoRAs?

Match the LoRA’s intended base (e.g., SD 1.5 vs. SDXL). Check the LoRA’s card or filename and load a compatible base checkpoint in the workflow. Mixing SDXL LoRAs with SD 1.5 (or vice versa) will produce poor results.

How do I add my own Krea 2 LoRA to this workflow?

Duplicate one of the existing style groups, swap the LoRA file path in that group’s loader, and keep the wiring to SaveImageAdvanced. Reuse the same prompt and seed to compare it fairly with the existing branches.

My images look over-stylized or washed out. What should I tweak?

Lower the LoRA strength, add targeted negatives (e.g., oversaturated, heavy artifacts, extra noise), and moderate CFG/steps. For subtle looks, keep style LoRAs below 0.8 and corrective LoRAs (like Nose Invisibles) around 0.2–0.4.

Where do my images go and what metadata is saved?

SaveImageAdvanced writes to the output folder you configured and embeds ComfyUI metadata (prompt, seed, and node parameters) into the file. This lets you reproduce or adjust settings later with confidence.

Krea 2 LoRAs Showcase - ComfyUI Workflow

This ComfyUI workflow is a hands-on showcase for six Krea 2 LoRAs, organized into clear groups: Nose Invisibles LoRA, VHS Analog LoRA, RetroAnime LoRA, Commodore 64 Style LoRA, NeonDrip LoRA, and PSX Low Poly LoRA. It lets you run the same prompt through multiple style branches so you can compare looks side-by-side. Outputs are written with SaveImageAdvanced, making it easy to review, sort, and document your best variants.

Technically, each group encapsulates a LoRA application path that adapts your base checkpoint at inference time, with controls for LoRA strength and prompt inputs. The workflow wires shared inputs to each branch so you can keep prompts consistent while varying LoRA weights. Custom nodes (IDs: 33681612-45f1-4901-b3fd-b83f0b48c9c0, 376d014e-b4fd-430e-8201-70629e00857f, 68a3ded7-ee8c-4dc0-8c96-128fdcdd4ef8, 307cc882-db9e-4f8c-93c5-130e09e2c305, fd7b28a7-8ec8-44ef-ad93-dff7afc6086b, 91c1ae42-2e94-4f8e-b10f-25415ac0b64e) bundle the LoRA logic per style group, while SaveImageAdvanced handles filenames, folders, and embedding ComfyUI metadata for reproducibility. This structure is ideal for quick A/B testing of style intensity, prompt phrasing, and seed changes without re-wiring your graph.

FAQ

Frequently Asked Questions