README.md

metadata

title: Krea 2 Style Reference
emoji: 🎨
colorFrom: indigo
colorTo: blue
sdk: gradio
sdk_version: 6.20.0
app_file: app.py
pinned: false
hardware: zero-a10g
short_description: Style-guided image generation with Krea 2 Turbo
python_version: '3.10'
startup_duration_timeout: 30m
models:
  - krea/Krea-2-Turbo
  - ostris/krea2_turbo_style_reference

Krea 2 Style Reference

Generate an image from a text prompt, guided by a style reference image. This Space uses:

• Base model: krea/Krea-2-Turbo (8-step distilled Krea 2)
• LoRA: ostris/krea2_turbo_style_reference (style reference conditioning)
• Pipeline: ostris/Krea2OstrisEdit (custom community pipeline for reference-image conditioning)

How it works

Upload a style reference image and type a prompt describing what you want to generate. The LoRA conditions the Krea 2 Turbo model on both:

1. Text encoder — the reference image is embedded into the prompt via Qwen3-VL vision placeholders
2. Latent space — clean VAE latents of the reference are appended to the transformer sequence at flow time t=0

This produces an image that follows your prompt while adopting the visual style of the reference image.

Inference recipe

• Steps: 8 (Turbo distilled)
• Guidance: 0.0 (disabled, per Krea 2 Turbo convention)
• LoRA scale: 1.0 (adjustable)
• Resolution: 1024×1024 (adjustable, must be multiples of 16)