Wan 2.2 Image LoRA Trainer API Documentation

Playground

Train custom Wan 2.2 character/style LoRA models 10x faster. Style training, character training, object training. From concept to model in minutes, not hours. Upload a ZIP file containing images to start!

Features

Wan 2.2 LoRA Trainer is a high-performance custom model training service for the Wan 2.2 text-to-video generation model. Train personalized LoRA (Low-Rank Adaptation) models 10x faster than traditional methods, enabling custom styles, characters, and objects for video generation.

Training Architecture

The trainer leverages Wan 2.2’s innovative MoE (Mixture of Experts) architecture, producing two specialized LoRA models:

high_noise_lora: Optimized for high-noise denoising timesteps, handling initial structure and composition
low_noise_lora: Optimized for low-noise denoising timesteps, refining details and final output quality

This dual-model approach ensures superior training efficiency and generation quality across all denoising stages.

Training Process

Data Upload: Upload a ZIP file containing your training images
Automatic Processing: The system automatically processes and optimizes your dataset
Dual Model Training: Simultaneously trains both high_noise_lora and low_noise_lora models
Model Delivery: Receive two specialized LoRA models ready for video generation

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": 0.0002,
  "lora_rank": 32
}
JSON
)

# 1. Submit the prediction.
SUBMIT_RESPONSE=$(curl --silent --show-error --fail-with-body \
  -X POST "https://api.wavespeed.ai/api/v3/wavespeed-ai/wan-2.2-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

Parameter	Type	Required	Default	Range	Description
data	string	Yes	-	-	To train a WAN T2V LoRA, you need to upload a zip file containing at least 10 images. In addition to images the archive can contain text files with captions. Each text file should have the same name as the image file it corresponds to.
trigger_word	string	No	-	-	Optional trigger word. If a caption file exists, it is prepended when not already present. If no caption file exists and trigger_word is non-empty, a caption containing only the trigger word is created. Leave empty to enable no-caption training.
steps	integer	No	1000	1000 ~ 10000	Number of steps to train the LoRA on.
learning_rate	number	No	0.0002	0 ~ 1	-
lora_rank	integer	No	32	1 ~ 128	-

Response Parameters

Parameter	Type	Description
code	integer	HTTP status code (e.g., 200 for success)
message	string	Status message (e.g., “success”)
data.id	string	Unique identifier for the prediction, Task Id
data.model	string	Model ID used for the prediction
data.outputs	array	Output values, usually URL strings; some models return text strings or structured result objects (empty when status is not `completed`)
data.urls	object	Object containing related API endpoints
data.urls.get	string	URL to retrieve the prediction result
data.status	string	Status of the task: `created`, `processing`, `completed`, or `failed`
data.created_at	string	ISO timestamp of when the request was created (e.g., “2023-04-01T12:34:56.789Z”)
data.error	string	Error message (empty if no error occurred)
data.timings	object	Object containing timing details
data.timings.inference	integer	Inference time in milliseconds

Result Request Parameters

Parameter	Type	Required	Default	Description
id	string	Yes	-	Task ID

Result Response Parameters

Parameter	Type	Description
code	integer	HTTP status code (e.g., 200 for success)
message	string	Status message (e.g., “success”)
data	object	The prediction data object containing all details
data.id	string	Unique identifier for the prediction
data.model	string	Model ID used for the prediction
data.outputs	array<string \| object>	Array of generated outputs (empty when status is not completed). Items are usually URL strings, but may be text strings or structured result objects, depending on the model.
data.urls	object	Object containing related API endpoints
data.urls.get	string	URL to poll for the prediction result
data.status	string	Status: `created`, `processing`, `completed`, or `failed`
data.created_at	string	ISO timestamp of when the request was created
data.error	string	Error message (empty if no error occurred)
data.timings	object	Object containing timing details
data.timings.inference	integer	Inference time in milliseconds

Overview