Pruna AI P Image Text To Image Trainer API Documentation

Pruna AI P Image Text To Image Trainer API Documentation

Playground

Pruna AI P-Image Text to Image Trainer is a fast AI model training workflow for customizing text-to-image generation models with user-provided data. Ready-to-use REST inference API for training custom styles, brand-specific visuals, character concepts, product image generation, marketing creatives, and personalized AI image workflows with simple integration, no coldstarts, and affordable pricing.

Features

Pruna AI P-Image Text-to-Image Trainer is a fast training workflow for creating custom LoRAs for the Pruna text-to-image ecosystem. Upload your training image data, optionally provide a default caption, choose the number of training steps, and generate a LoRA that can be used to steer future image generation toward your style, subject, or brand look.

Why Choose This?

• Fast custom LoRA training Train a text-to-image LoRA for specialized styles, subjects, or branded visuals.

• Simple training interface Provide your training image data, optional default caption, and training steps without a complex setup process.

• Optional caption guidance Use default_caption to provide consistent text conditioning across the training dataset.

• Flexible training depth Use steps to balance speed, cost, and how strongly the LoRA learns your dataset.

• Built for the Pruna image stack Trained outputs are intended for downstream use with Pruna text-to-image LoRA workflows.

• Production-ready API Suitable for custom style pipelines, branded asset generation, and repeatable image workflow customization.

Parameters

How to Use

1. Upload your training data — provide the image dataset you want to use for training.
2. Add a default caption (optional) — use a short caption if you want more consistent text conditioning across the dataset.
3. Set training steps — choose how many steps to run based on your desired balance of speed and training strength.
4. Submit — start the training job.
5. Use the trained LoRA — apply the resulting LoRA in downstream Pruna text-to-image LoRA workflows.

Example Workflow

Train a custom style LoRA from a curated image set, optionally using a shared default caption, then use the resulting weights in Pruna AI P-Image Text-to-Image LoRA for generation.

Pricing

Pricing is based on the selected steps value.

Billing Rules

• Pricing scales linearly with steps
• Cost is $1.80 per 1,000 steps
• Higher steps values increase total training cost proportionally
• default_caption does not affect pricing

Best Use Cases

• Custom style training — Create LoRAs for a distinct visual style or art direction.
• Brand consistency — Train reusable LoRAs for campaigns, products, or branded aesthetics.
• Subject-focused generation — Teach the model a recurring character, fashion look, or visual concept.
• Creative workflow personalization — Build specialized LoRAs for repeatable prompt-driven generation.
• Marketing asset pipelines — Create tailored generation tools for ongoing content production.

Pro Tips

• Use a clean, consistent training dataset for better LoRA quality.
• Add a default_caption when your dataset shares a common concept, subject, or style cue.
• Start with a moderate number of steps before pushing to larger training runs.
• Increase steps gradually if the initial LoRA is too weak or underfit.
• Keep your dataset focused on the style or subject you want the LoRA to learn.
• Test the trained LoRA in downstream generation workflows before scaling up training volume.

Notes

• image_data is required.
• default_caption is optional.
• steps defaults to 101.
• Pricing depends only on the selected steps value.
• LoRAs trained here are intended for Pruna text-to-image LoRA usage.

Related Models

• Pruna AI P-Image Text-to-Image LoRA — Generate images from prompts using a trained LoRA.
• Pruna AI P-Image Text-to-Image — Generate images directly from prompts without a LoRA.
• Pruna AI P-Image Edit Trainer — Train LoRAs for downstream image editing workflows.
• Pruna AI P-Image Edit LoRA — Edit images with LoRA-guided control.

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'
{
  "image_data": "https://github.com/mdn/interactive-examples/archive/refs/heads/main.zip",
  "steps": 101
}
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/pruna-ai/p-image/text-to-image-trainer" \
  -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