LTX 2.5 API

LTX 2.5 Image to video animates an opening frame you supply for 5 to 20 whole seconds at 720p, 1080p, 2K or 4K. image_url and prompt are required; last_image_url adds an optional closing frame, and duration_seconds, resolution and seed behave as they do on Text to video. There is no aspect_ratio field: the opening frame fixes subject and composition and sets the output shape. Clips come back with a generated soundtrack and tasks run asynchronously. Upload only frames you are authorised to use; you are responsible for how the output is used.

Choose a LTX 2.5 mode by starting material

The opening frame is followed closely

Identity, wardrobe, set and light come from image_url, and that frame is followed closely. It is the reliable route to a literal result: an uncensored opening frame comes back literal rather than softened. Because the image already carries the look, the prompt only has to carry what changes after it.

An optional frame for where the shot lands

last_image_url fixes the final composition and is valid only alongside image_url. It constrains the ending, not the route: the motion between the two frames is still generated, and across a long clip the middle can wander a long way. Two frames that disagree on light or wardrobe produce a visible negotiation between them, so prepare them as a continuous pair.

The same length, resolution and sound as the text endpoint

duration_seconds runs from 5 to 20 whole seconds, and longer clips wander further from a supplied closing frame. resolution offers 720p (default), 1080p, 2k (about 2560 px wide) and 4k. The output includes a generated audio track with no field to disable it, so plan to mute or replace it in the edit when you need your own mix.

A time-of-day transition between two approved frames

In the beach-to-bonfire example, image_url shows a couple on the beach at sunset and last_image_url shows the same two at night, wrapped in one blanket by a bonfire. The request runs 10 seconds at 1080p, and the prompt describes the passage rather than the destination: the sun going down and the firelight taking over. Pinning the ending lets the model invent the middle without drifting past a frame that is already approved.

Board-to-board shots for an edit

When a shot has to start on one storyboard frame and end on another, send both as image_url and last_image_url. The outgoing composition then matches the next cut, which is what makes a generated clip usable in a timeline instead of only as a standalone piece.

Motion added to a finished still

The candlelit-tub frame already holds the candles, the enamel and the wet emerald silk, so its prompt covers only the change: sinking into the water, hands dragged through soaked hair, eyes opening toward the camera. The deco-pool-backfall example is a physics beat instead, a backwards fall into water with rising bubbles and caustics crawling over the columns while the camera sinks along.

# Generate the idempotency key once; reuse it when retrying after a lost response.
export SPICY_IDEMPOTENCY_KEY="${SPICY_IDEMPOTENCY_KEY:-$(uuidgen)}"
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
  -H "Authorization: Bearer $SPICY_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $SPICY_IDEMPOTENCY_KEY" \
  -d '{
  "model": "lightricks/ltx-2.5/image-to-video",
  "input": {
    "duration_seconds": 5,
    "image_url": "https://example.com/reference.png",
    "prompt": "A slow aerial shot drifting over a misty pine forest at sunrise",
    "resolution": "720p"
  }
}'

# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
  -H "Authorization: Bearer $SPICY_API_KEY"

See the API reference for authentication, the complete schema, polling, and error handling.