An AI influencer training API should produce something reusable, not force every application to rebuild a GPU pipeline. MakeInfluencer accepts one to twenty account-owned reference images, prepares a twenty-image character dataset when necessary, trains the identity asynchronously, and returns two useful outputs:

- An `influencerId` for character-consistent MakeInfluencer generations
- A `loraUrl` for advanced workflows that need the trained LoRA file directly

The API uses the same credits as the MakeInfluencer dashboard and MCP integration. There is no second developer wallet, and the client cannot select training steps or submit its own price.

## When to Train an AI Influencer

Training is useful when the same identity must appear across many unrelated scenes, outfits, campaigns, or content formats. A reference-image generation endpoint can help with one request, but a trained character is a better fit when your application needs:

- A recurring virtual creator for social content
- Character-consistent product advertisements
- A stable spokesperson across images and talking videos
- A reusable identity for automated content pipelines
- Direct access to a LoRA for compatible external inference systems

For small changes to an already approved image, editing may still be faster. The [character-consistent image API guide](/guides/character-consistent-image-api) explains when to choose references, editing, or training.

## The Training Workflow

### 1. Upload the reference images

Create each asset with the [asset upload endpoint](/developers/assets). The training endpoint only accepts JPEG, PNG, and WebP URLs in the authenticated account's MakeInfluencer upload namespace. This prevents one customer from training with another customer's private uploads.

One clear portrait is enough to start. When fewer than twenty images are provided, MakeInfluencer creates the missing character-sheet variations before training. Supplying twenty good references skips that preparation stage.

### 2. Create an idempotent training job

```bash
curl --request POST https://www.makeinfluencer.ai/api/v1/training/jobs \
  --header "Authorization: Bearer $MAKEINFLUENCER_API_KEY" \
  --header "Idempotency-Key: train-avery-001" \
  --header "Content-Type: application/json" \
  --data '{
    "name": "Avery",
    "style": "realistic",
    "gender": "female",
    "assetUrls": ["YOUR_OWNED_ASSET_URL"],
    "consentAcknowledged": true
  }'
```

The rights confirmation is required because custom model training is materially different from generating a new fictional image. Only submit photographs you own or have documented permission to use.

The accepted response contains the IDs and exact server-defined credit charge:

```json
{
  "trainingId": "TRAINING_ID",
  "influencerId": "INFLUENCER_ID",
  "status": "queued",
  "ready": false,
  "terminal": false,
  "refunded": false,
  "creditsCost": 60000,
  "loraUrl": null
}
```

An `Idempotency-Key` is mandatory. Retrying the same logical request with the same body returns the original job instead of creating another influencer or spending credits twice.

### 3. Poll the owned training resource

```bash
curl https://www.makeinfluencer.ai/api/v1/training/jobs/TRAINING_ID \
  --header "Authorization: Bearer $MAKEINFLUENCER_API_KEY"
```

Stop polling when `terminal` becomes true. Jobs move through `queued`, `processing`, and then `succeeded` or `failed`. Training IDs are ownership-scoped, so an API key cannot inspect another account's job.

### 4. Use the influencer or LoRA

A successful response includes the trained file:

```json
{
  "trainingId": "TRAINING_ID",
  "influencerId": "INFLUENCER_ID",
  "status": "succeeded",
  "ready": true,
  "terminal": true,
  "refunded": false,
  "creditsCost": 60000,
  "loraUrl": "https://assets.makeinfluencer.ai/models/model.safetensors"
}
```

Use `influencerId` with supported MakeInfluencer image models when you want the hosted workflow. Use `loraUrl` when your application needs the weights for another compatible inference stack.

## Billing, Failure, and Retry Behavior

Training cost is calculated by the server from the fixed MakeInfluencer recipe. Callers do not control the number of steps, provider, dataset target, or charged amount.

The API rejects insufficient balances and foreign assets before creating a training job. If character-sheet preparation or LoRA training fails after credits are spent, the existing training failure path restores the complete training charge and the terminal status returns `refunded: true`.

Do not create a new job while the original is queued or processing. Poll the existing `trainingId`, and reuse the original idempotency key when recovering from a network failure around job creation.

The complete request contract, Axios and Python examples, status fields, and upload fallback are available in the [AI influencer training API documentation](/developers/training).

## AI Influencer Training API FAQ

### How many images are required to train an AI influencer?

The API accepts one to twenty images. MakeInfluencer pads smaller submissions to a twenty-image character dataset. Clear, varied and well-lit references still improve the result.

### Does the API return the LoRA file?

Yes. After successful training, the owner-authenticated status response includes `loraUrl` as well as the reusable MakeInfluencer `influencerId`.

### Can the client choose training steps?

No. Training settings stay server-controlled so credit charging and model quality remain predictable. The response reports the actual MakeInfluencer credit charge.

### What happens when training fails?

Failed character-sheet or LoRA jobs become terminal and restore the training credits. The status response reports `refunded: true`.

### Can I pass an external image URL?

No. Upload images through [POST /api/v1/assets/presign](/developers/assets) or the browser Library first. Training accepts only image assets owned by the authenticated account.

### Can the trained influencer generate videos?

The trained identity is primarily used to create consistent source images. Approved images can then be animated with an [image-to-video API](/guides/ai-image-to-video-api-for-influencers) or turned into a speaking avatar with the [InfiniteTalk API](/developers/models/infinitetalk-api).


Use the MakeInfluencer AI influencer training API to create a reusable consistent character from 1–20 images, poll the job, and retrieve both an influencer ID and LoRA URL.

AI Influencer Training API: Train a Character and Get the LoRA

How many images are required to train an AI influencer?

Can the trained influencer generate videos?

AI Influencer Training API: Train a Character and Get the LoRA

Create Your AI Influencer

Continue Reading

Seedance 2.0: AI Video Guide

How to Create AI UGC Video Ads: Practical Workflow

AI Character Consistency Guide

Ready to try it yourself?