v2 is out: realism LoRA stack + detail pass, 1536×1920. See the v2 tab.

Krea 2

Clean Krea 2 text-to-image with the sampler that actually works.

Krea 2 RAW with the turbo LoRA and one plain res_2s / beta pass. We tested the popular two-pass ClownsharK chain side by side: the refine changed nothing and this was sharper.

What you get

- Krea 2 text-to-image workflow

- Prompting notes: one moment, a camera line, no tag soup

What you need

- ComfyUI or the Gmanski RunPod template

- 24 GB+ VRAM

- VRAM: 24 GB+ · Level: Beginner

Models and packs

This workflow uses the krea2 pack from the Gmanski RunPod template, which installs the exact checkpoints, LoRAs and custom nodes it needs in one click: https://www.gmanski.com/runpod

Download

Free: the workflow JSON is attached to this listing. The full notes, tested settings and the model packs it needs are on its page:

https://www.gmanski.com/workflows/krea2

Help

Discord (the bot answers most questions in under a minute, a human follows up): https://discord.gg/6YSbrQamFF

---

Every face in the samples is a generated character. Gmanski makes workflows, a RunPod template, custom nodes and ready-made AI characters: https://www.gmanski.com