bytedance/lynx
Input
Hint: Drag and drop image files from your computer, images from web pages, paste from clipboard (Ctrl/Cmd+V), or provide a URL. Accepted file types: jpg, jpeg, png, webp, gif, avif
Type # to reference inputs.
Customize your input with more control.
Logs
Lynx | [image-to-video]
ByteDance's Lynx delivers subject-consistent video generation at $0.60 per second, trading inference speed for precise identity preservation across motion. Built on ByteDance's image-to-video architecture, it maintains character consistency through dedicated identity preservation scaling, critical for product demos, character animation, and branded content where visual continuity matters more than generation time.
Use Cases: Product Demonstrations | Character Animation | Marketing Content
Performance
At $0.60 per second of generated video, Lynx positions in the mid-range for image-to-video generation on fal, prioritizing identity preservation control over raw speed.
| Metric | Result | Context |
|---|---|---|
| Resolution | Up to 720p | 16:9, 9:16, or 1:1 aspect ratios |
| Video Length | Up to 81 frames | 5-30 FPS configurable output |
| Cost per Second | $0.60 | Based on default 81-frame generation |
| Identity Preservation | 0-2.0 scale | Dedicated `ip_scale` parameter for subject consistency control |
| Related Endpoints | Lynx Examples | Gallery and use case demonstrations |
Precision Identity Control for Subject-Consistent Animation
Lynx separates identity preservation (`ip_scale`) from reference image influence (`strength` and `guidance_scale_2`) in its parameter architecture, giving you independent control over how closely the subject's appearance matches versus how much the composition follows the reference.
What this means for you:
-
Independent identity scaling: Adjust subject consistency (0-2.0 range) separately from compositional guidance, preventing the "locked-in-place" effect, common when high reference adherence restricts motion
-
Multi-resolution output: Generate at 480p, 580p, or 720p with configurable aspect ratios (16:9, 9:16, 1:1) for platform-specific delivery without post-processing
-
Extended inference control: 50 inference steps by default with 1-75 range lets you balance quality against generation time based on content complexity
-
Frame-level customization: Configure 9-81 frames at 5-30 FPS for precise output duration control, critical for platform-specific video requirements (Instagram Reels, YouTube Shorts, product loops)
Technical Specifications
| Spec | Details |
|---|---|
| Architecture | Lynx |
| Input Formats | Single reference image (JPG, PNG, WebP, GIF, AVIF) + text prompt |
| Output Formats | MP4 video |
| Resolution Options | 480p, 580p, 720p with 16:9, 9:16, or 1:1 aspect ratios |
| License | Commercial use permitted |
API Documentation | Quickstart Guide | Enterprise Pricing
How It Stacks Up
Fabric 1.0 Image to Video – Lynx prioritizes independent identity preservation controls through dedicated scaling parameters for subject consistency workflows. Fabric 1.0 focuses on different architectural approaches for image-to-video generation.
MuseTalk Image to Video – Lynx offers configurable resolution and aspect ratio options (480p-720p, three aspect ratios) for platform-specific output. MuseTalk specializes in different image-to-video capabilities.
Kling Video v2.6 Pro Image to Video – Lynx provides granular control over identity preservation versus compositional guidance through separate parameter scaling. Kling Video v2.6 Pro targets different quality-speed tradeoffs in the image-to-video space.