A minimal, fast Krea 2 Turbo workflow for ComfyUI, set up the way I use it for AI character work.
What it does
Generates clean character portraits and scenes from a text prompt using Krea 2 Turbo's distilled model — built for speed: a full generation in 8 steps, so you can iterate on a character's look, outfit, and setting quickly before committing to slower, heavier runs.
Settings (already dialed in)
Steps: 8
CFG: 1
Sampler: euler
Model format: fp8 build for reasonable VRAM use
What you need to download
Krea 2 Turbo model (fp8, ~13 GB)
Matching text encoder (~5 GB)
VAE
Total is roughly 18.4 GB. Note: this needs a real GPU — it won't run usefully on CPU. If you're on Comfy Cloud, the Krea 2 models are already available there. Krea 2 is released under the Krea 2 Community License (commercial use under $1M revenue).
Tips
Write prompts like a photo description: subject, outfit, location, lighting. Short and concrete beats long and poetic.
This pairs with the style-reference version of the workflow for steering a character's look from a reference image.
This free workflow is the text-to-image half of the character pipeline I use. The full consistency setup — Flux 2 Dev workflow with two-reference-image identity anchoring, both Krea 2 workflows, my prompt library, and a quick-start guide — is in my AI Character Consistency Pack on Whop: https://whop.com/same-face-lab/same-face-every-photo-ai-character-consistency-pack/

