Z Image LoRA Trainer

Z Image LoRA Trainer

Playground

Z-Image-LoRA-Trainer – train custom image LoRA models from your own dataset, with zip uploads, auto-tuned defaults and fast iteration for brand, character or IP looks. Ready-to-use REST inference API, best performance, no coldstarts, affordable pricing.

Features

Z-Image LoRA Trainer is a high-performance custom model training service for the Z-Image-Turbo text-to-image generation models. It allows you to train lightweight LoRA (Low-Rank Adaptation) adapters for personalized styles, characters, and concepts, while preserving the fast and high-quality generation properties of Z-Image.

Training Architecture

The trainer is designed around Z-Image’s efficient diffusion architecture and its Turbo distilled variants, and produces one or more specialized LoRA adapters depending on your configuration:

• Base LoRA adapter Trains on the core Z-Image representation to capture your target style, character, or object, while keeping the base model frozen and stable.

• Turbo-aware fine-tuning (optional) When used with Z-Image-Turbo, the trainer applies Turbo-compatible optimization settings (step-aware learning rate, safe rank and scaling) to maintain high image quality even at low sampling steps.

This architecture ensures that your LoRA:

• Remains compact and easy to share
• Is plug-and-play with supported UIs and pipelines (e.g. ComfyUI / AI Toolkit)
• Preserves the speed and efficiency of Z-Image / Z-Image-Turbo

Training Process

1. Data Upload Prepare and upload a ZIP file containing your training images and, optionally, captions or prompts.

2. Automatic Preprocessing The trainer automatically:

• Validates and filters your dataset
• Resizes and normalizes images for Z-Image
• Aligns captions / prompts (or auto-generates basic captions if enabled)

3. LoRA Training The system runs a tailored LoRA optimization loop for Z-Image:

• Freezes the base model weights
• Trains only the low-rank adapter layers
• Applies Turbo-safe settings when targeting Z-Image-Turbo

4. Model Export After training completes, you receive:

• A LoRA adapter file compatible with Z-Image / Z-Image-Turbo
• Recommended loading settings (weight scale, steps, and sampling tips) for image generation

Price

• You pay $1.25 for every 1,000 training steps, billed proportionally to the total number of steps in your job.

Examples

Try more trainers

• Wan 2.2 Image LoRA Trainer - High-performance LoRA trainer for the Wan 2.2 image model, ideal for custom styles and characters that integrate smoothly into the Wan video/image ecosystem.

• Qwen Image LoRA Trainer - Built on the Qwen image model with strong multi-language prompt support and rich semantic understanding, great for text-sensitive image customization.

• Flux Dev LoRA Trainer - LoRA trainer tailored for the Flux Dev model, focusing on high-fidelity, creative visuals and experimentation with new artistic styles.

Guidance

• 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'
{
  "data": "https://github.com/mdn/interactive-examples/archive/refs/heads/main.zip",
  "steps": 1000,
  "learning_rate": 0.0001,
  "lora_rank": 16
}
JSON
)

# 1. Submit the prediction.
SUBMIT_RESPONSE=$(curl --silent --show-error --fail-with-body \
  -X POST "https://api.wavespeed.ai/api/v3/wavespeed-ai/z-image-lora-trainer" \
  -H "Authorization: Bearer ${WAVESPEED_API_KEY}" \
  -H "Content-Type: application/json" \
  -d "${REQUEST_BODY}")

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 "${PREDICTION_ID}" ]; then
  printf 'Submission response did not contain a prediction id
' >&2
  exit 1
fi
RESULT_URL=$(printf '%s' "${TASK}" | jq -r '.urls.get // empty')
if [ -z "${RESULT_URL}" ]; then RESULT_URL="https://api.wavespeed.ai/api/v3/predictions/${PREDICTION_ID}/result"; fi

# 2. Poll until the prediction finishes.
while true; do
  RESPONSE=$(curl --silent --show-error --fail-with-body \
    "${RESULT_URL}" \
    -H "Authorization: Bearer ${WAVESPEED_API_KEY}")
  RESULT=$(printf '%s' "${RESPONSE}" | jq 'if type == "object" and has("data") then .data else . end')
  STATUS=$(printf '%s' "${RESULT}" | jq -r '.status // empty')

  case "${STATUS}" in
    completed) printf '%s\n' "${RESULT}" | jq '.outputs'; break ;;
    failed|cancelled|timeout) printf '%s\n' "${RESULT}" | jq . >&2; exit 1 ;;
    created|processing) sleep 2 ;;
    *) printf 'Unexpected status: %s
' "${STATUS}" >&2; exit 1 ;;
  esac
done

Parameters

Task Submission Parameters

Request Parameters

Response Parameters

Result Request Parameters

Result Response Parameters