Krea 2 Emoji LoRA — custom Inference Endpoint
Custom-handler repo that serves linoyts/Krea2-emoji-LoRA
(a 3D-emoji style LoRA) on top of krea/Krea-2-Turbo
via Hugging Face Inference Endpoints.
The LoRA weights (pytorch_lora_weights.safetensors) are bundled here; the base model is pulled
from the Hub at startup. Inference uses the Turbo recipe: 8 steps, guidance_scale=0.0.
Request
import requests, base64
API_URL = "<your-endpoint-url>"
headers = {"Authorization": "Bearer <hf-token>", "Accept": "image/png"}
resp = requests.post(API_URL, headers=headers, json={
"inputs": "a wise wizard with a long white beard and a tall blue pointed hat, 3d emoji",
"parameters": {"num_inference_steps": 8, "guidance_scale": 0.0, "seed": 0},
})
with open("emoji.png", "wb") as f:
f.write(resp.content)
End prompts with 3d emoji to trigger the style.
Parameters
| name | default | notes |
|---|---|---|
num_inference_steps |
8 | Turbo is distilled for ~8 steps (max 30) |
guidance_scale |
0.0 | no CFG on Turbo |
height / width |
1024 | max 1536 |
seed |
random | int for reproducibility |