How does Wan v2.6 image-to-video differ from the Flash variant?

Wan v2.6 image-to-video standard offers superior motion quality, richer visual detail, and more complex scene animation compared to the Flash variant. Flash prioritizes speed and cost for high-throughput needs. Choose the standard Wan v2.6 when maximum output quality and motion coherence are required for final production or premium content delivery.

How do I use Wan v2.6 image-to-video through the eachlabs API?

Wan v2.6 image-to-video is accessible on the eachlabs platform under the model ID wan-v2.6-image-to-video. Submit an input image and receive a generated video clip via the eachlabs unified API. Access is pay-as-you-go with no Alibaba Cloud account required, and the model is available alongside the full Wan model family.

Example inputhover

prompt: "A comedic yet premium cinematic sequence where a printed object transforms reality. The scene begins exactly from the input image: the creator grips the stack of printed papers labeled “eachlabs” on the desk. [0–4s] The creator slowly pulls the stack apart. The paper resists like heavy fabric, emitting a deep mechanical sound. The creator murmurs, slightly amused: “This feels… expensive.” [4–8s] Hard match cut: the paper stretches outward and becomes a vast snow-covered mountain landscape surrounding the desk, wind moving snow and clouds as if the scene was folded inside the paper. The desk still exists at the center. The creator looks around, surprised: “That was not in the margins.” [8–12s] Smash cut: the paper sharply folds again and snaps open into a neon-lit futuristic city at night, rain reflecting colorful lights, cinematic depth and motion. The creator laughs softly: “Okay. That’s on me.” [12–15s] Hard cut back to the original studio. The paper stack settles back onto the table, perfectly intact. The “eachlabs” text is visible again. The creator looks directly into camera and says calmly: “One input. Multiple realities.” Photoreal 4K, cinematic lighting, strong match cuts, smooth camera motion, coherent main character, natural dialogue, no subtitles, no UI, no watermark."
image_url
resolution: "1080p"
duration: "15"
negative_prompt: "low resolution, error, worst quality, low quality, defects"
enable_prompt_expansion: true
enable_safety_checker: true

Wan v2.6 · Image to Video

Video·wan-v2.6·by Alibaba

Wan 2.6 is an image-to-video model that transforms images into high-quality videos with smooth motion and visual consistency.

Try it now →

API reference

Runtime (p50): 1m
Estimated price: From $0.1

Call the API

prediction.sh

curl -X POST \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{
    "model": "wan-v2-6-image-to-video",
    "version": "0.0.1",
    "input": {
        "prompt": "A comedic yet premium cinematic sequence where a printed object transforms reality.\n\nThe scene begins exactly from the input image: the creator grips the stack of printed papers labeled “eachlabs” on the desk.\n\n[0–4s] The creator slowly pulls the stack apart. The paper resists like heavy fabric, emitting a deep mechanical sound. The creator murmurs, slightly amused: “This feels… expensive.”\n\n[4–8s] Hard match cut: the paper stretches outward and becomes a vast snow-covered mountain landscape surrounding the desk, wind moving snow and clouds as if the scene was folded inside the paper. The desk still exists at the center. The creator looks around, surprised: “That was not in the margins.”\n\n[8–12s] Smash cut: the paper sharply folds again and snaps open into a neon-lit futuristic city at night, rain reflecting colorful lights, cinematic depth and motion. The creator laughs softly: “Okay. That’s on me.”\n\n[12–15s] Hard cut back to the original studio. The paper stack settles back onto the table, perfectly intact. The “eachlabs” text is visible again. The creator looks directly into camera and says calmly: “One input. Multiple realities.”\n\nPhotoreal 4K, cinematic lighting, strong match cuts, smooth camera motion, coherent main character, natural dialogue, no subtitles, no UI, no watermark.",
        "image_url": "https://storage.googleapis.com/magicpoint/inputs/wan-v2-6-image-to-video-input.png",
        "resolution": "1080p",
        "duration": "15",
        "negative_prompt": "low resolution, error, worst quality, low quality, defects",
        "enable_prompt_expansion": true,
        "enable_safety_checker": true
    },
    "webhook_url": ""
}' \
  https://api.eachlabs.ai/v1/prediction/

Documentation8 sections

Overview
wan-v2.6-image-to-video — Image-to-Video AI Model

Developed by Alibaba as part of the wan-v2.6 family, wan-v2.6-image-to-video transforms static images into cinematic 1080p videos up to 15 seconds long, with native audio synchronization and multi-shot narrative consistency that outperforms typical image-to-video AI models.

This lightweight flash variant excels in rapid inference for production workflows, preserving subject structure, lighting, and framing while generating smooth, realistic motion from a single input image and text prompt—ideal for creators seeking Alibaba image-to-video solutions without chaotic movements or identity drift.

Users upload JPG, PNG, or WebP images (up to 50MB) alongside prompts describing motion, enabling quick generation of short-form content like promotional clips or concept visuals via the wan-v2.6-image-to-video API.
Capabilities
- Generates high-fidelity 1080p videos from images with fluid motion and lighting consistency
- Native audio generation with precise lip-sync, dialogue, sound effects, and background music
- Multi-shot storytelling with coherent character consistency and smooth match cuts/transitions
- Supports aspect ratios like 16:9, 9:16, 1:1 for versatile framing
- Photorealistic outputs with strong temporal coherence and detail retention
- Motion transfer from reference videos or images, including camera logic and pacing control
- Multilingual prompt understanding (Chinese, English, others) for global use
- Versatile for text-to-video, image-to-video, reference-to-video modes
Use cases
Use Cases for wan-v2.6-image-to-video

Content creators turn product photos into engaging promo videos: upload a static image of a gadget and prompt "smooth pan around the device on a modern desk with soft lighting and subtle activation sounds," yielding a 1080p clip with synced audio for TikTok or Instagram Reels.

Marketers building e-commerce visuals use multi-shot capabilities to animate lifestyle scenes, inputting a character image with "multi-shot sequence: person walks into kitchen, pours coffee, smiles at camera with morning ambiance audio," maintaining consistency for compelling ads without studio shoots.

Developers seeking Alibaba image-to-video API integrate it for app prototypes, feeding user-uploaded images and prompts to generate personalized video previews, leveraging fast inference and lip-sync for interactive demos or virtual try-ons.

Filmmakers experiment with concept art: start with a storyboard frame prompting "cinematic zoom into fantasy landscape with wind rustling leaves and distant echoes," producing 15-second tests with natural motion and effects to refine pitches efficiently.
Tips & tricks
How to Use wan-v2.6-image-to-video on Eachlabs

Access wan-v2.6-image-to-video seamlessly on Eachlabs via the Playground for instant testing—upload an image (JPG/PNG up to 50MB), add a motion prompt, select duration (2-15s), resolution (720p/1080p), and optional audio— or integrate through the API/SDK for production apps, receiving high-quality 30 fps MP4 outputs with audio sync in minutes.
---
Technical spec
What Sets wan-v2.6-image-to-video Apart

wan-v2.6-image-to-video distinguishes itself in the image-to-video AI model landscape through its distilled flash architecture, delivering 720p or 1080p MP4 outputs at 30 fps in 2-15 seconds with average run times around 150 seconds—optimized for fast, scalable inference.
- Native audio-visual sync with lip-sync and ambient effects: Generates synchronized sound matched to scene context and lip movements from image prompts alone, enabling realistic dialogue or effects without post-production. This empowers users to create complete audiovisual clips instantly, perfect for social media reels.
- Multi-shot narrative consistency: Maintains subject fidelity across multiple shots with coherent transitions, a wan-v2.6 exclusive for storytelling sequences from a single starting image. Developers integrating image-to-video AI models gain tools for dynamic, professional-grade narratives without stitching clips manually.
- Restrained, cinematic motion control: Produces stable animations with natural camera movements and high frame rates, reducing common AI jitter for photorealistic or stylized outputs up to 1080p. This supports versatile short-form content like ads or previews with minimal iteration.
Input formats include images and optional audio (MP3, WAV), outputting H.264-encoded videos ready for professional use.
Things to be aware of
- Experimental multi-shot chaining achieves longer narratives but may vary in transition smoothness
- Known quirks: Better with clear input images; complex scenes can show minor motion jitter
- Performance: 14B variant offers higher fidelity but slower than 5B; cloud-optimized, no local GPU needed
- Resource requirements: Higher for 1080p/15s (e.g., increased latency/cost scaling with duration)
- Consistency strong across shots/characters, improved over Wan 2.5 per user benchmarks
- Positive feedback: Praised for integrated audio sync, speed, and production-ready quality
- Common concerns: Limited to 15s per clip; occasional need for prompt tweaks to avoid artifacts
Key considerations
- Use clear subjects with good lighting in input images for best animation results
- Enable prompt_expansion for short prompts to generate detailed internal scripts
- Set seed to a fixed integer for reproducible results or -1 for random variation
- Balance resolution and duration trade-offs: higher resolutions like 1080p increase processing time and cost
- Employ negative prompts to avoid artifacts like watermarks, text, distortion, or extra limbs
- For optimal motion, describe specific camera moves, story beats, and styles in prompts
- Limit to short clips (5-15s) per generation; chain multi-shots for longer narratives
- Test CFG scale at 1 for image-to-video to maintain stability
Limitations
- Restricted to short durations (max 15s per generation), requiring chaining for longer videos
- Optimal for 480p-1080p; no native 4K support currently
- May exhibit minor inconsistencies in highly complex motions or low-quality input images

Related models

4 models

Alibaba HappyHorse 1.0 · Image to Video AI model preview

Alibaba HappyHorse 1.0 · Image to VideoAlibaba

PixVerse C1 TransitionPixverse

Skyreels v4 · Image to Video AI model preview

Skyreels v4 · Image to VideoSkywork AI

PixVerse Multi Transition AI model preview

PixVerse Multi TransitionPixverse

* FAQ

About Wan v2.6 · Image to Video

01 / 03

What is Wan v2.6 image-to-video and what video quality does it produce?

Wan v2.6 image-to-video is Alibaba's latest image-to-video generation model that creates high-quality, motion-consistent video clips from static input images. It delivers improved temporal coherence, smoother motion trajectories, and better scene understanding compared to earlier Wan versions, supporting a range of video lengths and styles for commercial and creative applications.

Wan v2.6 · Image to Video