Language

Krea2 text-to-image workflow

Imported from ModelTrends

Krea2 is a very fast, very capable new foundational model. Use natural prompting without negatives. A low-step workflow: 8 steps, guidance 1. Unlike the first Krea model it is not based on Flux 2 or Qwen, though it uses the Qwen encoders. A versatile, high-quality, safe-for-work model for commercial use.

Compatible / Available in Graydient-powered Apps

Sample renders

Made with the models this workflow runs. Prompts are credited on BunchaPrompts; the model opens on ModelTrends. Pick a thumbnail to see the render large with the prompt that made it.

Sample render 1

Prompt by Cinnadust on BunchaPrompts · Krea2 - Melancholy

Prompt

zidiusArt, anime-style tiny fox, made of sparkling white and black fire, sleeping, curled up, cradled in a nest of multicolored painterly styled butterflies. Forest floor at late evening, red and yellow golden sunset lighting. Autumn leaves.

Prompts from BunchaPrompts. Workflows and models on ModelTrends.

Options

OptionMaps toDefault
(prompt)CLIP Text Encode → text—
/guidanceKSampler → cfg1
/imagesEmpty Latent Image → batch_size1
/seedKSampler → seedrandom
/sizeEmpty Latent Image → width, height1024 × 1024
/stepsKSampler → steps8

Models used

Krea2 (v7) Model imported by shortcode /run:krea2 from ModelTrends See it on ModelTrends

For Developers (optional)

Example API snippet. You can build your own app with this workflow for personal use. Contact us for high availability servers when you're ready to deploy to the world. See the full rendering API documentation.

curl https://cloud.graydient.ai/api/v3/render/ \
  -X POST \
  -H "Content-Type: application/vnd.api+json" \
  -H "Accept: application/vnd.api+json" \
  -H "Authorization: Bearer $GRAYDIENT_TOKEN" \
  -d '{
    "prompt": "/workflow /run:krea2 dog eating pizza",
    "callback_url": "https://example.com/webhook/render"
  }'

This is how it works behind the scenes. This is already configured for you. The parameters in the table above plug into these slots, allowing you to change how it works and which models get loaded. You don't have to understand or modify this - it just works, already set up.

Load Diffusion Model MODEL to EasyLoraStack; Load CLIP CLIP to EasyLoraStack; EasyLoraStack MODEL to KSampler; EasyLoraStack CLIP to CLIP Text Encode (Prompt); CLIP Text Encode (Prompt) CONDITIONING to KSampler; Empty Latent Image LATENT to KSampler; KSampler LATENT to VAE Decode; Load VAE VAE to VAE Decode; VAE Decode IMAGE to Save Image
Load Diffusion Model#1
MODEL
unet_nameKrea2 base model (dyn…
Load CLIP#13
CLIP
clip_nameqwen3vl_4b_fp8_scaled
EasyLoraStack#17
modelMODEL
clipCLIP
lora_1Krea2 LoRA (dynamic)
lora_2Krea2 LoRA (dynamic)
lora_3Krea2 LoRA (dynamic)
lora_4Krea2 LoRA (dynamic)
lora_5Krea2 LoRA (dynamic)
CLIP Text Encode (Prompt)#6
clipCONDITIONING
textdog eating pizza(prompt)
Empty Latent Image#5
LATENT
width1024/size
height1024/size
batch_size1/images
KSampler#3
modelLATENT
positive
latent_image
seedrandom/seed
steps8/steps
cfg1/guidance
noteno negative prompt
Load VAE#4
VAE
vae_nameqwen_image_vae
VAE Decode#8
samplesIMAGE
vae
Save Image#9
images
filename_prefixGraydient

Read-only view. The highlighted pills are the options a call can set. Natural prompting, no negatives.

All the workflows: Powerful Workflows

See plans and pricing