Skip to content
FlixMLFlixML Studioopen-source generative AI image and video tool

Krea 2 Turbo

Krea 2 Turbo makes an image from a prompt in 8 steps, 1K to 2K. Krea 2 Community License: https://www.krea.ai/krea-2-licensing

Workflow id: krea2_turbo · Text → Image

Calling it

POST /api/image/generate
{
  "workflow": "krea2_turbo",
  "prompt": "..."
}

Params

Every input this workflow takes. Your install is the source of truth:GET /api/workflows/krea2_turbo returns these as they are on your machine.

  • prompt · str · required
    Natural language, not a tag list: Krea recommends natural-language prompts and says long, detailed ones give the best results. Name the medium (a photograph, a digital painting) along with the subject, framing, setting and light. Put any text to render in quotes.
  • width · int · default 1024
    Krea 2 Turbo generates from 1K to 2K. Multiples of 16.
  • height · int · default 1024
    Multiples of 16.
  • seed · int · optional
    Optional. Leave it out and every run is a new random take. Send a seed only to remake an image exactly: the seed a job used is in its record.
  • steps · int · default 8
    Turbo is distilled for 8 steps, the number Krea and ComfyUI both run.
  • sampler · str · default euler
  • scheduler · str · default simple
  • lora_name · str · optional
    Optional LoRA file in the node's loras folder. Leave empty and the slot is removed from the graph entirely. Use a LoRA made for Krea 2: its authors train on Krea 2 Raw and run on Turbo.
  • lora_strength · float · default 1
    Weight of the first LoRA. The right range is the LoRA author's: some Krea 2 LoRAs want 0.01, sliders run negative, others ask for 3 to 5. Ignored when lora_name is empty.
  • lora_name_2 · str · optional
    Second LoRA, chained after the first, for stacking one LoRA on another. Same rules as lora_name.
  • lora_strength_2 · float · default 1
    Weight of the second LoRA, from its author. Ignored when lora_name_2 is empty.
  • unet · str · default krea2_turbo_fp8_scaled.safetensors
    Diffusion model file on the node. ComfyUI recommends the fp8 build for most users.
  • clip · str · default qwen3vl_4b_fp8_scaled.safetensors
  • vae · str · default qwen_image_vae.safetensors