# FLUX.1 [schnell] FP8

> FP8 quantized variant of Black Forest Labs' FLUX.1 [schnell] model, offering ~2x faster inference with reduced precision while maintaining high-quality image generation in 4 steps

## Quick Reference

- Model ID: flux.1-schnell-fp8
- Creator: Black Forest Labs
- Status: active
- Family: flux.1
- Base URL: https://api.lumenfall.ai/openai/v1

## Specifications

- Max Resolution: 1024x1024
- Input Modalities: text
- Output Modalities: image
- Supported Modes: Text to Image

## API Parameters

The compiled parameter schema for this model is available via the API: `GET /v1/models/flux.1-schnell-fp8?schema=true`.

### Core Parameters

- `prompt` (string) — REQUIRED: Text prompt for image generation
- `seed` (integer): Random seed for reproducibility

### Size & Layout

- `size` (string): Image dimensions as WxH pixels (e.g. "1024x1024") or aspect ratio (e.g. "16:9"). Values: 1563x670, 670x1564, 1365x768, 768x1365, 1254x836, 836x1254, 887x1182, 1024x1024, 1145x916, 916x1145, 1183x887
- `aspect_ratio` (string): Aspect ratio of the output image (e.g. "16:9", "1:1"). Values: 9:21, 9:16, 2:3, 3:4, 4:5, 1:1, 5:4, 4:3, 3:2, 16:9, 21:9
- `resolution` (string): Output resolution tier (e.g. "1K", "4K"). Values: 1K

### Output & Format

- `response_format` (string): How to return the image. Default: url. Values: url, b64_json
- `output_format` (string): Output image format. Values: png, jpeg, gif, webp, avif
- `output_compression` (integer): Compression level for lossy formats (JPEG, WebP, AVIF)
- `n` (integer): Number of images to generate. Default: 1

### Additional Parameters

- `cfg_scale` (number): Classifier-free guidance scale — higher values stick more closely to the prompt
- `num_inference_steps` (integer): Number of denoising steps for the image generation process.. Only available via fireworks


## Model Identifiers

- Primary Slug: flux.1-schnell-fp8

## Dates

- Released: October 2024

## Tags

image-generation, text-to-image, fast, open-weights, quantized

## Available Providers

### Fireworks AI

- Config Key: fireworks/flux.1-schnell-fp8
- Provider Model ID: accounts/fireworks/models/flux-1-schnell-fp8/text_to_image
- Regions: global
- Pricing: Free/image
  - Note: Free to try
  - Note: Normally priced at $0.00035 per inference step
  - Note: FLUX.1 [schnell] uses 4 steps by default, making the effective per-image cost $0.0014
  - Note: FP8 variant uses reduced precision for ~2x faster inference
  - Source: https://fireworks.ai/pricing


## Performance Metrics

Provider performance over the last 30 days.

### fireworks

- Median Generation Time (p50): 1239ms
- 95th Percentile Generation Time (p95): 4656ms
- Average Generation Time: 1614ms
- Success Rate: 87.6%
- Total Requests: 1031
- Time to First Byte (p50): 822ms
- Time to First Byte (p95): 1638ms


## Arena Benchmarks

### Chalkboard Menu

- Elo: 1148
- Record: 11W / 8L / 1T (20 battles)
- Rank: #12 of 69

### The Reversed Rodeo

- Elo: 1097
- Record: 1W / 7L / 4T (12 battles)
- Rank: #16 of 68

### Isometric Miniature Diorama Scenes

- Elo: 991
- Record: 0W / 4L / 0T (4 battles)
- Rank: #72 of 72

### The Halloween Invitation

- Elo: 987
- Record: 0W / 9L / 0T (9 battles)
- Rank: #70 of 70


## Use Cases & Category Performance

### Photorealism (Text-to-Image)

- Rank: #28 of 61
- Elo: 1174
- Record: 15W / 22L / 5T (42 battles)
- Win Rate: 35.7%

### Art (Text-to-Image)

- Rank: #13 of 24
- Elo: 1139
- Record: 1W / 16L / 4T (21 battles)
- Win Rate: 4.8%

### Creativity (Text-to-Image)

- Rank: #33 of 56
- Elo: 1179
- Record: 3W / 22L / 4T (29 battles)
- Win Rate: 10.3%

### Text Rendering (Text-to-Image)

- Rank: #37 of 58
- Elo: 1166
- Record: 13W / 25L / 2T (40 battles)
- Win Rate: 32.5%

### Product, Branding & Commercial (Text-to-Image)

- Rank: #38 of 56
- Elo: 1167
- Record: 13W / 21L / 2T (36 battles)
- Win Rate: 36.1%

### Prompt Adherence (Text-to-Image)

- Rank: #43 of 60
- Elo: 1188
- Record: 15W / 27L / 6T (48 battles)
- Win Rate: 31.3%

### Aesthetics (Text-to-Image)

- Rank: #46 of 62
- Elo: 1172
- Record: 5W / 28L / 5T (38 battles)
- Win Rate: 13.2%

### 3D Imaging & Modeling (Text-to-Image)

- Rank: #29 of 29
- Elo: 1049
- Record: 1W / 5L / 0T (6 battles)
- Win Rate: 16.7%


## Image Gallery

5 images available for this model. Browse all at https://lumenfall.ai/models/black-forest-labs/flux.1-schnell-fp8/gallery

### Curated Examples

- [Cinematic wide shot of a high-end, minimalist boutique storefront at dusk. The shop's large glass...](https://assets.lumenfall.ai/p3PGE8NU2DZnO1QpCD80_9LTFaUptwlWfG3urTbB70k/rs:fit:1500:1500/plain/gs://lumenfall-prod-assets/s6kjof7uf4giml4wmd19sf4duujc@jpeg)

### Arena Competition Results

- [Chalkboard Menu](https://assets.lumenfall.ai/7mFyHbXgy4xvCJcQFTQI5QzXRaBNTcx5d3ui1tup6YI/rs:fit:1500:1500/plain/gs://lumenfall-prod-assets/kblfpmmrzuojgc924vzkjaazjfog@jpeg): #12 of 69 (Elo 1148)
- [The Reversed Rodeo](https://assets.lumenfall.ai/T8yz3YnxoQqd0r4TAYQnHykLv-eeu7geztWLIOL1ZTQ/rs:fit:1500:1500/plain/gs://lumenfall-prod-assets/tzqz9texg5akmn8wavi4ram6q4iq@jpeg): #16 of 68 (Elo 1097)
- [Isometric Miniature Diorama Scenes](https://assets.lumenfall.ai/ul0e1uFpR-jCW5C67wuXZz92-wj97tbTQGiOIFWdhOk/rs:fit:1500:1500/plain/gs://lumenfall-prod-assets/1ciliwu4w2a0j463zvdrban64gqd@jpeg): #72 of 72 (Elo 991)
- [The Halloween Invitation](https://assets.lumenfall.ai/mU-yCTbrfOv-xfcBH_Hq3k4YiGBV-9fgCnvScYWI8Yo/rs:fit:1500:1500/plain/gs://lumenfall-prod-assets/lb40wcjtregyh2bt4f7a52r4s0ps@jpeg): #70 of 70 (Elo 987)

## Example Prompt

The following prompt was used to generate an example image in our playground:

A cozy street-side flower shop with a large chalkboard sign that reads "FRESH BLOOMS & WILD SUNFLOWERS" in elegant cursive. A golden retriever sits by the door, while a small capybara rests quietly behind a bucket of tulips in the background.

## Code Examples

### Text to Image (/v1/images/generations)

#### cURL

curl -X POST \
  https://api.lumenfall.ai/openai/v1/images/generations \
  -H "Authorization: Bearer $LUMENFALL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "flux.1-schnell-fp8",
    "prompt": "",
    "size": "1024x1024"
  }'

# Response:
# { "created": 1234567890, "data": [{ "url": "https://...", "revised_prompt": "..." }] }

#### JavaScript

import OpenAI from 'openai';

const client = new OpenAI({
  apiKey: 'YOUR_API_KEY',
  baseURL: 'https://api.lumenfall.ai/openai/v1'
});

const response = await client.images.generate({
  model: 'flux.1-schnell-fp8',
  prompt: '',
  size: '1024x1024'
});

// { created: 1234567890, data: [{ url: "https://...", revised_prompt: "..." }] }
console.log(response.data[0].url);

#### Python

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://api.lumenfall.ai/openai/v1"
)

response = client.images.generate(
    model="flux.1-schnell-fp8",
    prompt="",
    size="1024x1024"
)

# { created: 1234567890, data: [{ url: "https://...", revised_prompt: "..." }] }
print(response.data[0].url)


## About

## Overview
FLUX.1 [schnell] FP8 is a quantized version of Black Forest Labs’ distilled text-to-image model, optimized for maximum inference speed. By utilizing 8-bit floating-point precision, this variant achieves significantly lower latency and reduced memory overhead compared to the standard model. It is specifically designed for high-throughput applications where generating competitive imagery in a handful of steps is the primary requirement.

## Strengths
*   **Generation Speed:** Produces usable 1024x1024 images in just 1 to 4 sampling steps, making it one of the fastest high-resolution open-weight models available.
*   **Standardized Resource Efficiency:** The FP8 quantization reduces the VRAM footprint and computational load, allowing for roughly 2x faster inference times compared to the full-precision version without a proportional loss in visual quality.
*   **Prompt Adherence:** Despite the lowered precision and distillation, the model retains the architectural ability to follow complex descriptive prompts and render legible, coherent text within images.
*   **Output Consistency:** It maintains the structural integrity and composition characteristic of the FLUX.1 family, even at extremely low step counts.

## Limitations
*   **Artistic Nuance:** Due to the distillation and quantization, it offers less stylistic flexibility and fine-grained detail compared to the [dev] or [pro] iterations of FLUX.1.
*   **Precision Loss:** FP8 quantization can occasionally lead to minor artifacts or less smooth gradients in complex lighting scenarios that would be better handled by 16-bit or 32-bit models.
*   **Step Sensitivity:** The model is strictly tuned for low-step counts; increasing the sampling steps beyond the recommended range usually yields diminishing returns or visual regressions.

## Technical Background
FLUX.1 [schnell] is a latent diffusion model based on a flow-based transformer architecture. This specific FP8 variant applies post-training quantization to the model weights, mapping them to 8-bit precision to optimize throughput on modern hardware. The "schnell" version itself is the result of a performance-oriented distillation process, allowing the model to reach a converged image state in a fraction of the time required by standard diffusion processes.

## Best For
This model is ideal for real-time applications, rapid prototyping, and high-volume image generation workflows where operational cost and latency are critical. It is a strong choice for "generate-as-you-type" interfaces or large-scale content pipelines that require decent photorealism at minimal compute expense. FLUX.1 [schnell] FP8 is available for testing and integration through Lumenfall's unified API and interactive playground.

## Frequently Asked Questions

### How much does FLUX.1 [schnell] FP8 cost?

FLUX.1 [schnell] FP8 is free to use through Lumenfall's unified API.

### How do I use FLUX.1 [schnell] FP8 via API?

You can use FLUX.1 [schnell] FP8 through Lumenfall's OpenAI-compatible API. Send requests to the unified endpoint with model ID "flux.1-schnell-fp8". Code examples are available in Python, JavaScript, and cURL.

### Which providers offer FLUX.1 [schnell] FP8?

FLUX.1 [schnell] FP8 is available through Fireworks AI on Lumenfall. Lumenfall automatically routes requests to the best available provider.

### What is the maximum resolution for FLUX.1 [schnell] FP8?

FLUX.1 [schnell] FP8 supports images up to 1024x1024 resolution.

## Links

- Model Page: https://lumenfall.ai/models/black-forest-labs/flux.1-schnell-fp8
- About: https://lumenfall.ai/models/black-forest-labs/flux.1-schnell-fp8/about
- Providers, Pricing & Performance: https://lumenfall.ai/models/black-forest-labs/flux.1-schnell-fp8/providers
- API Reference: https://lumenfall.ai/models/black-forest-labs/flux.1-schnell-fp8/api
- Benchmarks: https://lumenfall.ai/models/black-forest-labs/flux.1-schnell-fp8/benchmarks
- Use Cases: https://lumenfall.ai/models/black-forest-labs/flux.1-schnell-fp8/use-cases
- Gallery: https://lumenfall.ai/models/black-forest-labs/flux.1-schnell-fp8/gallery
- Playground: https://lumenfall.ai/models/black-forest-labs/flux.1-schnell-fp8/playground
- API Documentation: https://docs.lumenfall.ai