muapi Logomuapi Logo
muapi Logomuapi Logo

Flux 1 Dev Style LoRA Inference: Lora Support

Generate images with a custom-trained FLUX.1 [dev] style LoRA applied. Pass the LoRA from flux-1-dev-style-lora-trainer straight into inference. Try free — pay per generation.

Interactive model controls

Run FLUX.1 [dev] text-to-image inference with your trained LoRA applied. Pass the `air` identifier returned by the flux-1-dev-style-lora-trainer job to apply your trained style or character.

📝

Overview

About this model

FLUX.1 [dev] Style LoRA Inference runs text-to-image generation with a trained style LoRA applied on top of the FLUX.1 [dev] base model. Pass the lora_url returned by a completed flux-1-dev-style-lora-trainer job (or any compatible .safetensors URL) to reproduce your trained style or character in new generations.

1Style-consistent generation: Generate new images that carry a trained style or character from your own dataset.
2Brand asset production: Reuse a trained brand-style LoRA across unlimited new prompts.
3Character illustration: Generate new scenes featuring a character LoRA trained on reference art.
4A/B style testing: Compare `lora_weight` values to dial in how strongly the trained style shows through.
💰

Pricing & Value

Cost analysis

muapiapp$0.025 per generation

Flat per-generation price, independent of which trained LoRA is applied.

Fal.ai$0.03 per generation

Comparable FLUX.1 [dev] LoRA inference is available but priced slightly higher.

ReplicateVaries

Usage-based pricing that depends on the selected hardware tier.

** Competitor pricing is estimated based on similar model architectures and usage tiers.

⚙️

Technical Details

Configuration schema

Promptstring

Text prompt describing the image.

Default ValueA portrait in the trained style, studio lighting.
LoRA URLstring

The LoRA file URL returned in `outputs` from a completed flux-1-dev-style-lora-trainer job.

Default Valuehttps://cdn.muapi.ai/outputs/generated/example.safetensors
LoRA Weightfloat

LoRA weight multiplier.

Default Value1
Aspect RatioEnum (7 options)

The aspect ratio of the generated image.

Default Value1:1
📖

Implementation Guide

Developer documentation

How to Use FLUX.1 [dev] Style LoRA Inference

  1. Train a Style LoRA (if you haven't already)

  2. Write Your Prompt

    • Describe the image you want, referencing your trigger word if the LoRA was trained with one.

  3. Apply the LoRA

    • Set lora_url to the trained .safetensors file, and lora_weight (0–4, default 1) to control how strongly it's applied.

  4. Set the Aspect Ratio

    • Choose aspect_ratio for the output image.

  5. Submit via the API

curl -X POST https://api.muapi.ai/api/v1/flux-1-dev-style-lora-inference \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A portrait in the trained style, studio lighting.",
    "lora_url": "https://cdn.muapi.ai/outputs/generated/example.safetensors",
    "lora_weight": 1.0,
    "aspect_ratio": "1:1"
  }'
  1. Poll and Download
    • Poll GET /api/v1/predictions/{request_id}/result until status is completed, then retrieve the image URL from output.
❓

Common Questions

Frequently asked

What does this endpoint do?

It generates FLUX.1 [dev] images with a trained style LoRA applied, so your output carries the style or character captured during training.

Where does the `lora_url` come from?

The `outputs` of a completed flux-1-dev-style-lora-trainer job, or any other publicly reachable `.safetensors` file compatible with FLUX.1 [dev].

What does `lora_weight` control?

How strongly the LoRA influences the output, from 0 (no effect) to 4 (maximum). The default of 1.0 works well for most trained styles.

Do I need a trigger word in my prompt?

If your LoRA was trained with a trigger word, include it in the prompt to reliably activate the trained concept.

What aspect ratios are supported?

16:9, 9:16, 1:1, 3:4, 4:3, 21:9, and 9:21, set via `aspect_ratio`.