Wan 2.1 14b Vace

Wan 2.1 14b Vace

Playground

WAN 2.1 VACE is an all-in-one video model supporting Reference-to-Video (Image-to-Video), V2V, Masked V2V and Move/Swap/Animate capabilities. Ready-to-use REST inference API, best performance, no coldstarts, affordable pricing.

Features

Wan 2.1 14B VACE is a versatile, production-oriented video generation and editing model that supports multi-input workflows. You can provide a text prompt plus up to 5 reference images, and optionally add a source video, masks, or start/end frames to guide motion, structure, and edits. It also includes multiple task modes (e.g., depth) for more controlled video understanding and generation.

Key capabilities

• Prompt-driven video generation with multi-modal controls
• Up to 5 reference images to guide identity, style, wardrobe, or scene details
• Optional video input for video-to-video transformation workflows
• Mask support (mask_video / mask_image) for region-based edits
• First/last frame guidance (first_image / last_image) for better continuity
• Task modes (e.g., depth) for structured control and more predictable results

Use cases

• Reference-guided video generation (character/style consistency across shots)
• Video editing with masks (replace background, remove objects, localized changes)
• Start-to-end guided storytelling using first_image + last_image
• Video-to-video restyling (apply a new look while keeping motion)
• Controlled motion and composition using task settings (e.g., depth)

Pricing

Longer durations are billed in steps based on duration.

Inputs

• prompt (required): what should happen in the video
• images (optional): up to 5 reference images
• video (optional): source video for video-to-video workflows
• mask_video (optional): video mask for localized video edits
• mask_image (optional): image mask for localized edits
• first_image (optional): starting frame guidance
• last_image (optional): ending frame guidance
• negative_prompt (optional): what to avoid

Parameters

• task: control mode selector (e.g., depth)
• duration: video length (e.g., 5s)
• size: output resolution (e.g., 832×480, 1280×720)
• num_inference_steps: sampling steps
• guidance_scale: prompt adherence strength
• flow_shift: motion/flow behavior tuning
• context_scale: reference/context strength tuning
• seed: random seed (-1 for random; fixed for reproducibility)
• enable_fast_mode: speed-optimized mode (if available in your UI)

Prompting guide (multi-reference + optional masks)

A reliable structure:

1. Define the main subject and action
2. Specify environment and camera beats
3. Assign roles to references (identity/style/outfit/background)
4. If using masks, clearly state what changes inside vs. outside the mask
5. If using first/last frames, describe how the motion should transition between them

Template: Use image 1 for identity, image 2 for outfit, image 3 for style. Generate a 5-second clip where [action]. Keep identity consistent. If mask is provided, change only the masked region to [edit], keep everything else unchanged.

Example prompts

• An elegant lady carefully selects bags in a boutique. Soft natural lighting, shallow depth of field, subtle camera push-in, gentle hand movements, realistic fabric and leather textures.
• Use the reference images for the same character and outfit. Walk through a luxury store aisle, turn to examine a handbag, warm highlights on leather, calm cinematic pacing.
• If mask is provided: Replace only the masked background with a modern boutique interior, keep the subject unchanged, match lighting and shadows.

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'
{
  "prompt": "A cinematic ocean wave at sunrise, highly detailed",
  "task": "depth",
  "duration": 5,
  "size": "832*480",
  "num_inference_steps": 30,
  "guidance_scale": 5,
  "flow_shift": 16,
  "context_scale": 1,
  "seed": -1
}
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-2.1-14b-vace" \
  -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

Response Parameters

Result Request Parameters

Result Response Parameters