What types of text prompts work best with Wan v2.6 text-to-video?

Wan v2.6 text-to-video responds well to detailed scene descriptions specifying subject, action, environment, lighting, and camera style. Prompts with clear motion descriptors such as speed, direction, and character behavior yield the most coherent results. Abstract or highly symbolic prompts may produce less predictable outputs compared to concrete visual descriptions.

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

Wan v2.6 text-to-video is available on the eachlabs platform under the model ID wan-v2.6-text-to-video. Submit a text prompt via the eachlabs unified API and receive a generated video clip. Access Alibaba's full Wan v2.6 model family alongside 150+ other AI models on eachlabs with pay-as-you-go pricing.

Example inputhover

prompt: "Humorous but Premium Mini-Trailer Concept: A tiny porcelain robot curator controls reality like a film set. Visual style (global): Extreme photoreal 4K, cinematic lighting, shallow depth of field, subtle film grain, smooth stabilized camera, premium VFX realism. Shot 1 — [0–3s] Scene: Macro close-up of a small porcelain robot curator. It gently taps a miniature tuning fork engraved with “eachlabs”. Soft reverb in the air. Dialogue (natural, calm): “Begin.” Shot 2 — [3–6s] Hard cut: A vast Arctic ice plain under pale blue sky. Wind blows snow across the ground. Wide cinematic shot. The robot stands in the foreground, tiny against the scale. It slightly turns its head. Dialogue: “More space.” (The horizon stretches wider, camera pulls back.) Shot 3 — [6–10s] Hard cut: A glowing underwater coral canyon. Sun rays penetrate the water, particles floating. The robot calmly walks along the seabed, unaffected by water. Camera slowly tracks forward between corals and fish. Dialogue (soft, curious): “Let it breathe.” Shot 4 — [10–15s] Hard cut: A silent lunar surface at dawn. Earth rising in the background. Slow orbital camera move around the robot as it looks at the horizon. It nods once. Dialogue: “Ready for the next reality.”"
aspect_ratio: "16:9"
resolution: "1080p"
duration: "15"
negative_prompt: "low resolution, error, worst quality, low quality, defects"
enable_prompt_expansion: true
multi_shots: true
enable_safety_checker: true

Wan v2.6 · Text to Video

Video·wan-v2.6·by Alibaba

Wan 2.6 is a text-to-video model that generates high-quality videos with smooth motion and cinematic detail.

Try it now →

API reference

Runtime (p50): 5m
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-text-to-video",
    "version": "0.0.1",
    "input": {
        "prompt": "Humorous but Premium Mini-Trailer\n\nConcept: A tiny porcelain robot curator controls reality like a film set.\n\nVisual style (global):\nExtreme photoreal 4K, cinematic lighting, shallow depth of field, subtle film grain, smooth stabilized camera, premium VFX realism.\n\nShot 1 — [0–3s]\n\nScene:\nMacro close-up of a small porcelain robot curator.\nIt gently taps a miniature tuning fork engraved with “eachlabs”.\nSoft reverb in the air.\n\nDialogue (natural, calm):\n\n“Begin.”\n\nShot 2 — [3–6s]\n\nHard cut:\nA vast Arctic ice plain under pale blue sky. Wind blows snow across the ground.\nWide cinematic shot.\nThe robot stands in the foreground, tiny against the scale.\n\nIt slightly turns its head.\n\nDialogue:\n\n“More space.”\n\n(The horizon stretches wider, camera pulls back.)\n\nShot 3 — [6–10s]\n\nHard cut:\nA glowing underwater coral canyon. Sun rays penetrate the water, particles floating.\nThe robot calmly walks along the seabed, unaffected by water.\nCamera slowly tracks forward between corals and fish.\n\nDialogue (soft, curious):\n\n“Let it breathe.”\n\nShot 4 — [10–15s]\n\nHard cut:\nA silent lunar surface at dawn. Earth rising in the background.\nSlow orbital camera move around the robot as it looks at the horizon.\n\nIt nods once.\n\nDialogue:\n\n“Ready for the next reality.”",
        "aspect_ratio": "16:9",
        "resolution": "1080p",
        "duration": "15",
        "negative_prompt": "low resolution, error, worst quality, low quality, defects",
        "enable_prompt_expansion": true,
        "multi_shots": true,
        "enable_safety_checker": true
    },
    "webhook_url": ""
}' \
  https://api.eachlabs.ai/v1/prediction/

Documentation8 sections

Overview
wan-v2.6-text-to-video — Text to Video AI Model

Developed by Alibaba as part of the wan-v2.6 family, wan-v2.6-text-to-video is a cutting-edge text-to-video AI model that transforms text prompts into cinematic multi-shot videos up to 15 seconds long with synchronized audio. This Alibaba text-to-video solution excels in generating coherent narratives with smooth transitions, character stability, and professional camera control, solving the challenge of creating high-quality short-form video content without extensive editing. Ideal for developers seeking a text-to-video AI model with multi-shot capabilities, it supports 720p and 1080p resolutions at 30 fps in MP4 format, delivering polished outputs for commercial use.
Capabilities
- Generates smooth, high-quality 1080p videos with cinematic detail, reduced jitter, and graceful depth/perspective transitions
- Native audio integration with phoneme-level lip-sync, including emotional micro-gestures for realistic talking animations
- Strong prompt adherence for complex instructions, multi-character scenes, and action sequences
- Video-to-video motion transfer for stable character consistency and multi-shot storytelling
- Multilingual support for text prompts and audio generation, enabling localized content
- Efficient rendering for batch production of short-form videos like social media or educational clips
- Versatile inputs: text, images, reference videos; aspect ratios for various formats
Use cases
Use Cases for wan-v2.6-text-to-video

Content creators producing social media reels can input a prompt like "A bustling city street at dusk transitioning to a cozy cafe interior with soft jazz audio syncing to barista movements" to generate a 10-second multi-shot video with seamless camera pans and ambient sound, ready for platforms like TikTok or Instagram.

Marketers crafting product demos use wan-v2.6-text-to-video for text-to-video AI generation of explainers, such as turning "Slow-motion reveal of a smartphone on a rotating pedestal with sparkling reflections and upbeat music sync" into a 1080p clip that highlights features with realistic physics and lighting, bypassing costly shoots.

Developers building apps with Alibaba text-to-video integration leverage its API for automated video assets, feeding prompts with optional audio to create personalized user content like "Avatar character walking through a futuristic city, narrating in a calm voice with matching lip sync," ensuring high consistency for interactive experiences.

Filmmakers prototyping scenes input detailed storyboards to produce 15-second test footage with professional rhythm and transitions, accelerating pre-production for narrative shorts or ads.
Tips & tricks
How to Use wan-v2.6-text-to-video on Eachlabs

Access wan-v2.6-text-to-video seamlessly on Eachlabs via the Playground for instant testing, API for production-scale wan-v2.6-text-to-video API calls, or SDK for custom apps. Provide a text prompt, optional audio file, duration (2-15s), and resolution (720p/1080p); it outputs MP4 videos at 30 fps with multi-shot narratives and sync. Eachlabs delivers fast, high-fidelity results optimized for your workflows.
---
Technical spec
What Sets wan-v2.6-text-to-video Apart

wan-v2.6-text-to-video stands out in the text-to-video landscape through its rebuilt narrative engine, enabling precise interpretation of storyboard-style prompts for multi-shot sequences with natural camera movements and rhythm control—unlike single-clip generators. This allows users to produce full cinematic stories from a single text description, streamlining workflows for promotional clips and explainers.

It supports integer durations from 2 to 15 seconds in 720p or 1080p at 30 fps, with optional audio input for lip-sync and ambient sound synchronization, maintaining temporal stability over extended lengths. Developers integrating the wan-v2.6-text-to-video API benefit from fast inference and high subject fidelity, reducing post-production needs.
- Multi-shot narrative engine: Handles complex scene sequences and transitions for professional-grade storytelling.
- Audio-video sync: Generates or syncs audio to match lip movements and scene context, perfect for talking-head or dynamic videos.
- Extended 15s HD support: Delivers 1080p videos with consistent lighting, motion, and character identity across shots.
Things to be aware of
- Users report dramatic improvements in audio sync and motion smoothness over Wan 2.5, with fewer artifacts and more human-like gestures
- Early adopters highlight faster processing and accessibility, ideal for iterative workflows
- Benchmarks show efficiency gains with sparse attention, reducing generation time significantly
- Resource needs scale with model size; cloud-optimized but larger 14B variant demands more for fidelity
- Community notes strong character consistency across shots and stable video-to-video pipelines
- Positive feedback on prompt accuracy for precise executions, rivaling higher-end models in specific categories
- Some discussions mention optimization for 5-15s clips, with chaining for longer content
Key considerations
- Use detailed, procedural prompts for best literal accuracy in multi-character scenes or complex actions to leverage the model's strength in precise execution
- Optimal for short clips (5-15s); chain multiple generations for longer narratives to maintain consistency
- Balance model size: 5B for speed, 14B for higher fidelity in demanding scenes
- Prioritize reference videos or images for video-to-video mode to enhance motion transfer and character stability
- Avoid overly abstract or highly interpretive prompts, as the model favors cinematic clarity over loose creativity
- Test lip-sync with clear audio inputs for natural emotional cues like gestures and expressions
Limitations
- Limited to short durations (5-15s per generation), requiring chaining for extended videos which may introduce minor inconsistencies
- Best for structured prompts; struggles with highly abstract or overly interpretive cinematic styles compared to specialized models
- Higher resolutions and longer clips increase render times, though mitigated by optimizations like sparse attention

Related models

4 models

Seedance V1.5 Pro · Text to VideoBytedance

Kling o3 4K · Text to Video AI model preview

Kling o3 4K · Text to VideoKling

Veo 3.1 Lite · Text to VideoGoogle

Pixverse v5.6 · Text to VideoPixverse

* FAQ

About Wan v2.6 · Text to Video

01 / 03

What is Wan v2.6 text-to-video and how does it generate video from text?

Wan v2.6 text-to-video is Alibaba's latest generation text-to-video model that generates high-quality video clips directly from natural language descriptions. It uses an advanced diffusion-based architecture with improved motion modeling to produce temporally coherent, visually detailed videos across diverse scenes, styles, and subject matters.

Wan v2.6 · Text to Video