# SDXL Lightning 4-step - Core.Today AI API

> ByteDance's 4-step SDXL Lightning — the most-run model on Replicate (1B+ runs). Near-instant 1024px image generation at one of the lowest prices in the catalog.

- **Provider**: ByteDance
- **Model ID**: bytedance/sdxl-lightning-4step
- **Category**: Image Generation
- **Credits**: 4 per image (num_outputs billed per image)
- **Speed**: Fast
- **Quality**: Standard

## Features
- 1B+ runs — the most-run model on Replicate
- 4-step generation: near-instant results
- Up to 1280px output, up to 4 images per request
- Negative prompt and scheduler control
- One of the lowest prices in the catalog (5 credits/image)

## Use Cases
- Rapid concept iteration
- High-volume image pipelines
- Real-time preview generation
- Placeholder and mockup art
- Prompt experimentation before premium models

## API Endpoint

Base URL: https://api.core.today/v1
Create Prediction: POST /predictions
Get Status: GET /predictions/{job_id}

## Authentication
Header: X-API-Key: YOUR_API_KEY

## Input Parameters

### Optional
- **prompt**: string (default: self-portrait of a woman, lightning in the background) - Input prompt
- **negative_prompt**: string (default: worst quality, low quality) - What to avoid in the image
- **width**: integer (default: 1024) - Output width (256-1280). 1024 or 1280 recommended
- **height**: integer (default: 1024) - Output height (256-1280). 1024 or 1280 recommended
- **num_outputs**: integer (default: 1) - Number of images (1-4). Billed per image
- **scheduler**: string (default: K_EULER) - scheduler
  Options: DDIM, DPMSolverMultistep, HeunDiscrete, KarrasDPM, K_EULER_ANCESTRAL, K_EULER, PNDM, DPM++2MSDE
- **num_inference_steps**: integer (default: 4) - Denoising steps (1-10). 4 gives the best results
- **guidance_scale**: number (default: 0) - Classifier-free guidance scale. Keep 0 for Lightning
- **seed**: integer (default: 0) - Random seed. Leave blank to randomize
- **disable_safety_checker**: boolean (default: false) - Disable safety checker for generated images

## Examples

### Instant Concept Art
Generate a 1024px image in under a second of compute
```json
{
  "model": "bytedance/sdxl-lightning-4step",
  "input": {
    "prompt": "self-portrait of a woman, lightning in the background",
    "width": 1024,
    "height": 1024,
    "num_inference_steps": 4
  }
}
```

## Response Format
```json
{
  "job_id": "abc123",
  "status": "pending | processing | completed | failed",
  "result": "URL or data (when completed)"
}
```

## Usage Flow
1. POST /predictions with model and input -> receive job_id
2. GET /predictions/{job_id} -> poll until status is completed or failed
3. Result contains output URL(s)

## Tips
- Keep num_inference_steps at 4 and guidance_scale at 0 — that's what Lightning is tuned for
- Use it to iterate prompts cheaply, then re-run winners on FLUX or Seedream for final quality
- num_outputs up to 4 for quick variations; billed per image
- 1024x1024 or 1280x1280 give the best quality

## Documentation
https://replicate.com/bytedance/sdxl-lightning-4step