Flux Kontext Dev LoRA API Documentation

Flux Kontext Dev LoRA API Documentation

Playground

Fast FLUX.1 Kontext [dev] endpoint with LoRA support for rapid image editing using pre-trained adapters for brand and style. Ready-to-use REST inference API, best performance, no coldstarts, affordable pricing.

Features

FLUX Kontext Dev LoRA is an instruction-based image-to-image editing model with built-in LoRA support. Provide a source image plus a natural-language edit request, and optionally attach up to 3 LoRAs to steer style, subject consistency, or domain-specific aesthetics during the edit.

Key capabilities

• Image-to-image editing from a source image + text instruction
• LoRA-enabled inference: apply up to 3 LoRAs via input parameters
• Works for both local edits (specific changes) and global transforms (overall look)
• Ideal for batch consistency: reuse the same LoRA set for a stable visual identity

Pricing

$0.03 per image.

Cost per run = num_images × $0.03 Example: num_images = 4 → $0.12

Inputs and outputs

Input:

• One source image (upload or public URL)
• One edit instruction (prompt)
• Optional: up to 3 LoRA items

Output:

• One or more edited images (controlled by num_images)

Parameters

Core:

• prompt: Edit instruction describing what to change and what to preserve
• image: Source image
• width / height: Output resolution
• num_inference_steps: More steps can improve fidelity but increases latency
• guidance_scale: Higher values follow the prompt more strongly; too high may over-edit
• num_images: Number of variations generated per run
• seed: Fixed value for reproducibility; -1 for random
• output_format: jpeg or png

LoRA (up to 3 items):

• loras: A list of LoRA entries (max 3)

• path: Either owner/model-name or a direct.safetensors URL from the Internet

• scale: LoRA strength (typically start around 0.6–1.0 and adjust)

Prompting guide

Use “preserve + edit + constraints” and let LoRAs drive the look:

Template: Keep [what must stay]. Change [what to edit]. Ensure [constraints]. Apply the LoRA style consistently without changing identity.

Example prompts

• Keep the person’s face, hairstyle, and pose unchanged. Replace the background with a clean studio gradient. Match lighting and shadows.
• Keep the product shape and label layout unchanged. Replace only the label text with “WaveSpeedAI”. Preserve typography perspective and print texture.
• Remove small blemishes and reduce glare on the forehead while keeping natural skin texture and pores.

Best practices

• Use fewer LoRAs when possible; add a second/third only if you need combined effects.
• If results drift or over-style, reduce LoRA scale or guidance_scale and strengthen the preserve clause.
• For consistent batches, keep the same LoRA set and scales, and fix seed when comparing prompt variants.

Reference

• How to train your LoRA?
• How to use your LoRA?

Authentication

For authentication details, please refer to the Authentication Guide.

API Endpoints

Submit Task & Query Result

set -euo pipefail

export WAVESPEED_API_KEY="your-api-key"

REQUEST_BODY=$(cat <<'JSON'
{
  "num_inference_steps": 28,
  "guidance_scale": 2.5,
  "num_images": 1,
  "output_format": "jpeg"
}
JSON
)

# 1. Submit the prediction once. Never retry this request automatically:
# a retry after an unclear failure can create a second, billed task.
CHECK='The task may already exist. Check History or the predictions list before submitting again.'
if ! SUBMIT_RESPONSE=$(curl --silent --show-error --fail-with-body \
  --connect-timeout 10 --max-time 60 \
  -X POST "https://api.wavespeed.ai/api/v3/wavespeed-ai/flux-kontext-dev-lora" \
  -H "Authorization: Bearer ${WAVESPEED_API_KEY}" \
  -H "Content-Type: application/json" \
  -d "${REQUEST_BODY}"); then
  printf 'Submit failed. %s\n' "${CHECK}" >&2
  exit 1
fi

TASK=$(printf '%s' "${SUBMIT_RESPONSE}" | jq 'if type == "object" and has("data") then .data else . end')
PREDICTION_ID=$(printf '%s' "${TASK}" | jq -r '.id // empty')
if [ -z "${PREDICTION_ID}" ]; then
  printf 'Submission response did not contain a prediction id. %s\n' "${CHECK}" >&2
  exit 1
fi
RESULT_URL="https://api.wavespeed.ai/api/v3/predictions/${PREDICTION_ID}/result"

# 2. Poll until the prediction reaches a final status.
BODY_FILE=$(mktemp)
trap 'rm -f "${BODY_FILE}"' EXIT
while true; do
  # Transient errors (429, 5xx) are retried, honoring Retry-After.
  HTTP_CODE=$(curl --silent --show-error --fail-with-body \
    --connect-timeout 10 --max-time 60 --retry 5 --retry-delay 2 \
    --output "${BODY_FILE}" --write-out '%{http_code}' \
    "${RESULT_URL}" \
    -H "Authorization: Bearer ${WAVESPEED_API_KEY}" || true)
  case "${HTTP_CODE}" in
    2??) ;;
    000|429|5??)  # dropped connection or temporary error: keep polling
      printf 'Task %s: HTTP %s, polling again in 10s\n' "${PREDICTION_ID}" "${HTTP_CODE}" >&2
      sleep 10; continue ;;
    *) printf 'Task %s: HTTP %s\n' "${PREDICTION_ID}" "${HTTP_CODE}" >&2
      cat "${BODY_FILE}" >&2; exit 1 ;;
  esac
  RESULT=$(jq 'if type == "object" and has("data") then .data else . end' "${BODY_FILE}")
  STATUS=$(printf '%s' "${RESULT}" | jq -r '.status // empty')

  case "${STATUS}" in
    completed) printf '%s\n' "${RESULT}" | jq '.outputs'; break ;;
    failed|cancelled|timeout|deleted)
      printf 'Task %s ended with status %s\n' "${PREDICTION_ID}" "${STATUS}" >&2
      printf '%s\n' "${RESULT}" | jq . >&2; exit 1 ;;
    *) sleep 2 ;;
  esac
done

Parameters

Task Submission Parameters

Request Parameters

Every model also accepts the global request options enable_sync_mode and enable_base64_output; see their pages for details.

Response Parameters

Result Request Parameters

Result Response Parameters