Character training API

How it works

Training runs as a job like everything else: POST the photos, poll the character until its status is ready, then use it. Behind the scenes it fits a LoRA against one of the SDXL checkpoint families and stores it server-side.

You never handle a LoRA file. The API does not accept or return weights — the character is referenced by id, and the server resolves the file and its trigger token at generation time.

• Choose the family at training time: general-image (photoreal), anime, or cartoon
• A character is usable on any model in the family it was trained against
• Photos are kept so a character can be retrained for free if needed
• Delete a character at any time and the weights go with it

Train a character

Send public https URLs rather than base64 — the request body has to stay under about 4 MB.

curl https://api.fastgencloud.com/v1/characters \
  -H "Authorization: Bearer $FASTGEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
      "name": "Ada",
      "class": "woman",
      "base_model": "general-image",
      "images": [
        "https://example.com/1.jpg",
        "https://example.com/2.jpg",
        "https://example.com/3.jpg",
        "https://example.com/4.jpg"
      ]
    }'

Generate the trained character

Once the character reports status "ready", reference it by id on a normal generation call.

curl https://api.fastgencloud.com/v1/images/generations \
  -H "Authorization: Bearer $FASTGEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
      "model": "general-image",
      "prompt": "a woman in a linen shirt reading at a cafe table, soft afternoon light, shallow depth of field",
      "character": {
        "id": "chr_0123456789ab"
      },
      "size": "832x1216"
    }'

Choosing training photos

Photo quality determines likeness far more than step count does.

• One person per photo, face clearly visible
• Vary angle, distance, lighting and expression — 10–20 varied photos beat 30 near-identical ones
• Avoid heavy filters, sunglasses, motion blur and group shots
• The default 1000 steps suits most sets; more steps tighten likeness but can overfit a small set

Pricing and slots

Each training run is billed once from your wallet. Storing a trained character occupies a slot: every account gets two free, and monthly packages add more.

If a package lapses, characters enter a grace period before being archived — and the original photos are retained so you can retrain without re-uploading.