Framepack

Framepack

Playground

Framepack is an efficient autoregressive Image-to-Video model that generates smooth, temporally consistent videos from a single image. Ready-to-use REST inference API, best performance, no coldstarts, affordable pricing.

Features

FramePack — wavespeed-ai/framepack

FramePack is an image-to-video model designed for smooth, cinematic animation from a single input image. Upload a reference image to anchor composition and subject identity, then use a director-style prompt to control motion, pacing, and camera language (push-in, pull-back, reveals, etc.). FramePack exposes frame-level control via num_frames, making it convenient for generating clips at different lengths while keeping output stable and consistent.

Key capabilities

• Image-to-video generation anchored to a reference image
• Strong at cinematic camera moves (push-in, pull-back, reveal, orbit, tilt)
• Frame-level length control via num_frames for flexible clip duration
• Supports negative_prompt to reduce jitter, blur, distortion, and artifacts
• Resolution and aspect_ratio controls for common output formats

Use cases

• “Living poster” animations: bring key art to life with subtle motion
• Cinematic reveals: close-up → pull-back to establish scene context
• Mood shots and b-roll from a single still (rain, neon, dust motes, fog, wind)
• Trailer-style beats for marketing and social content
• Rapid iteration by keeping the same image and varying prompt/seed/frames

Pricing

Pricing scales with the number of frames generated.

Inputs

• image (required): reference image (subject/composition anchor)
• prompt (required): motion + camera direction
• negative_prompt (optional): what to avoid (blur, jitter, distortion, etc.)

Parameters

• image: input image (upload or URL)
• prompt: director-style motion description
• negative_prompt: optional “avoid list”
• aspect_ratio: output aspect ratio (e.g., 16:9)
• resolution: output resolution (e.g., 720p)
• num_inference_steps: sampling steps
• num_frames: total frames to generate (controls clip length)
• guidance_scale: prompt adherence strength
• seed: random seed (set for reproducible results)

Prompting guide (I2V)

Write prompts like a shot list:

• Start framing: close-up / medium / wide
• Camera move: push-in / pull-back / pan / orbit
• Motion: hair, cloth, rain, particles, light flicker, subtle facial change
• Mood/lighting: neon, rim light, fog, bokeh, cinematic contrast
• Constraints: keep the subject identity and composition consistent

Example prompts

• Animate with a deliberate, unfolding sense of drama. Start with a tight close-up on the eyes, then slowly pull back to reveal the full figure on a rain-slick balcony, neon city lights shimmering in the background, subtle wind and drifting rain, cinematic lighting, smooth camera motion.
• Slow push-in on the subject, soft fog rolls through the scene, gentle light flicker, filmic contrast, no jitter, stable face and hands.

# Submit the task
curl --location --request POST "https://api.wavespeed.ai/api/v3/wavespeed-ai/framepack" \
--header "Content-Type: application/json" \
--header "Authorization: Bearer ${WAVESPEED_API_KEY}" \
--data-raw '{
    "aspect_ratio": "16:9",
    "resolution": "720p",
    "seed": 0,
    "num_inference_steps": 25,
    "num_frames": 180,
    "guidance_scale": 10
}'

# Get the result
curl --location --request GET "https://api.wavespeed.ai/api/v3/predictions/${requestId}/result" \
--header "Authorization: Bearer ${WAVESPEED_API_KEY}"