AI Video Generator: Complete Guide

The

on Melies lets you create cinematic video clips using 8 AI models from the world's leading providers. From quick concept clips to production-quality footage with native audio, this guide covers every feature available.

Available Models

Melies offers 8 video generation models, each with distinct strengths in quality, duration, resolution, and special features.

Model Comparison Table

Model	Provider	Speed	Quality	Credits	Resolution	Duration	Key Features
Veo 3.1	Google	Slow	Best	400	4K	4-8s	Native audio generation
Kling v3 Pro	Kuaishou	Medium	High	100	HD	3-15s	Multi-shot, lip sync, Voice ID
Kling O3 Standard	Kuaishou	Medium	High	80	HD	3-15s	Character elements
Kling v3 Standard	Kuaishou	Medium	Good	60	HD	3-15s	Character elements
Seedance v1 Pro	ByteDance	Medium	Good	80	1080p	2-12s	-
Hailuo 02	MiniMax	Medium	Good	70	512P/768P	6s or 10s	-
WAN v2.2	Alibaba	Medium	Good	60	480-720p	~5s	-
LTX 2 Pro	Lightricks	Fast	Good	50	HD	~4.8s	Camera movement presets (LoRA)

Choosing the Right Model

For the highest quality: Veo 3.1 produces the best visual quality at 4K resolution and includes native audio generation. At 400 credits per clip, it is best reserved for hero shots and final production footage.

For versatile filmmaking: Kling v3 Pro is the most feature-rich model with multi-shot sequences, lip sync, and Voice ID support. It offers the longest duration (up to 15 seconds) at a reasonable 100 credits.

For character-driven scenes: Kling O3 Standard and Kling v3 Standard both support character elements for maintaining visual consistency at 80 and 60 credits respectively.

For fast iteration: LTX 2 Pro is the fastest model at just 50 credits, with built-in camera movement presets. Great for previewing scenes and exploring ideas.

For budget-friendly clips: WAN v2.2 (60 credits) and LTX 2 Pro (50 credits) are the most affordable options for generating video concepts.

Text-to-Video

Text-to-video is the simplest way to generate a clip. Write a description of your scene and the model creates a video from scratch.

How to Generate Video from Text

1. Open the
   AI Video Generator
2. Select your preferred model
3. Write a detailed scene description in the prompt field
4. Choose duration and resolution settings (options vary by model)
5. Click generate

Tips for Better Text-to-Video Results

• Be specific about movement: Describe what is happening in the scene, not just what it looks like. "A woman walks slowly through a rainy street, looking over her shoulder" works better than "woman in rainy street."
• Include camera direction: Mention camera movements like "slow dolly forward" or "camera pans left" to guide the composition.
• Set the mood: Include lighting and atmosphere details. "Dimly lit with warm amber tones" gives the model clear visual direction.
• Keep it focused: Each clip should describe one clear action or moment. Avoid cramming multiple actions into a single prompt.

Image-to-Video

Image-to-video lets you animate a still image, turning it into a moving clip. This workflow gives you much more control over the visual result because you start with a known frame.

How to Generate Video from an Image

1. Generate an image using the
   AI Image Generator
   or upload your own
2. Open the
   AI Video Generator
3. Upload or select your starting image
4. Write a prompt describing what should happen in the video (the movement, action, camera motion)
5. Select your model and settings
6. Click generate

Why Use Image-to-Video

• Character consistency: Generate a character image first, then animate it - the video will closely match the starting frame
• Precise framing: Get the exact composition you want in the first frame before adding motion
• Storyboard to film: Turn storyboard frames into animated sequences
• Cost efficiency: Iterate on cheap images first (2 credits with Flux Schnell), then spend video credits only on frames you are happy with

Duration and Resolution Options

Each model has different duration and resolution capabilities.

Duration by Model

Model	Minimum	Maximum	Notes
Veo 3.1	4s	8s	Includes native audio
Kling v3 Pro	3s	15s	Longest single-clip duration
Kling O3 Standard	3s	15s	-
Kling v3 Standard	3s	15s	-
Seedance v1 Pro	2s	12s	-
Hailuo 02	6s	10s	Fixed at 6s or 10s
WAN v2.2	~5s	~5s	Fixed duration
LTX 2 Pro	~4.8s	~4.8s	Fixed duration

Resolution by Model

Model	Resolution Options
Veo 3.1	Up to 4K
Kling v3 Pro	HD
Kling O3 Standard	HD
Kling v3 Standard	HD
Seedance v1 Pro	1080p
Hailuo 02	512P or 768P
WAN v2.2	480p to 720p
LTX 2 Pro	HD

Camera Movement Presets

LTX 2 Pro includes built-in camera movement presets powered by LoRA (Low-Rank Adaptation) technology. Instead of describing camera motion in your prompt, you can select a preset for precise, consistent results.

Available Presets

Preset	Description
Dolly	Camera moves forward or backward along a track, creating depth
Jib	Camera moves vertically up or down, simulating a crane shot
Static	Camera remains fixed, focusing entirely on subject movement

These presets are exclusive to LTX 2 Pro and are applied automatically during generation. They produce smoother, more predictable camera motion than prompt-based directions.

Multi-Shot Sequences

Kling v3 Pro supports multi-shot sequence generation, a powerful feature for creating connected scenes.

What Are Multi-Shot Sequences

Instead of generating individual clips and editing them together, multi-shot mode lets you generate several connected shots in a single generation. The model maintains visual consistency between shots, making transitions feel natural.

When to Use Multi-Shot

• Dialogue scenes: Generate shot and reverse-shot sequences
• Action sequences: Create connected beats in an action scene
• Scene transitions: Move from one framing to another smoothly
• Short narratives: Tell a brief visual story in one generation

Lip Sync and Voice ID

Kling v3 Pro includes lip sync and Voice ID capabilities for creating dialogue-driven scenes.

Lip Sync

Provide dialogue text and the model generates a character speaking those words with synchronized lip movements. This is valuable for:

• Creating talking-head scenes without live actors
• Animating AI-generated characters with speech
• Prototyping dialogue scenes before production

Voice ID

Voice ID lets you maintain a consistent voice across multiple generated clips. Once you establish a voice for a character, you can reuse it in new generations, keeping your characters sounding the same throughout your project.

Extending Videos

Melies lets you extend generated videos beyond their initial duration. After generating a clip, you can extend it to continue the action or scene.

How Video Extension Works

1. Generate a video clip using any model
2. Select the extend option on the generated clip
3. The model generates additional frames that continue seamlessly from where the original clip ended
4. Repeat to build longer sequences

This is useful for creating clips longer than a single model's maximum duration, or for extending a particularly good generation that you want to continue.

Audio Generation

Native Audio with Veo 3.1

Veo 3.1 is the only model that generates synchronized audio alongside video. This includes:

• Ambient sounds matching the scene (rain, wind, crowds)
• Sound effects tied to on-screen actions
• Environmental audio that matches the visual setting

This makes Veo 3.1 uniquely suited for producing complete audiovisual clips without a separate audio step.

Adding Audio During Export

For all other models, you can add audio during the export process.

• Music: Add background music tracks with volume control
• Voice: Layer voiceover or dialogue
• SFX: Add sound effects

Character Consistency with Reference Images

Maintaining consistent characters across multiple video clips is one of the biggest challenges in AI filmmaking. Melies offers several tools to help.

Workflow for Consistent Characters

1. Create your character using the
   AI Image Generator
   with a detailed description
2. Extract the subject to isolate the character from the background
3. Use as reference when generating new images and videos
4. Select character element models like Kling O3 Standard or Kling v3 Standard for built-in character consistency

Character Elements

Kling O3 Standard and Kling v3 Standard support character elements, which allow you to define persistent character traits that the model maintains across generations. This is more reliable than prompt-based consistency alone.

Export Formats and Settings

When your clips are ready, Melies offers flexible export options.

Video Formats

Format	Best For
MP4	Universal compatibility, social media, most players
WebM	Web playback, smaller file sizes

Export Settings

Setting	Options
Resolution	480p, 720p, 1080p
Quality	Low (CRF 28), Medium (CRF 23), High (CRF 18)
Frame Rate	1-60 FPS (default 25)
Aspect Ratio	16:9, 3:2, 1:1, 2:3, 9:16

Additional Export Options

• Zoom mode: Apply zoom effects to your final video
• Poster frame: Set a specific frame as the video thumbnail
• Audio mixing: Combine music, voice, and sound effects with individual volume controls

Credit Costs Breakdown

Video generation costs more credits than images because of the computational complexity involved. Here is a quick reference.

Model	Credits per Clip
LTX 2 Pro	50
WAN v2.2	60
Kling v3 Standard	60
Hailuo 02	70
Kling O3 Standard	80
Seedance v1 Pro	80
Kling v3 Pro	100
Veo 3.1	400

Cost-Effective Video Workflow

1. Concept phase: Use LTX 2 Pro (50 credits) or WAN v2.2 (60 credits) to test ideas quickly
2. Refinement: Switch to Kling v3 Standard (60 credits) or Seedance v1 Pro (80 credits) for better quality
3. Production: Use Kling v3 Pro (100 credits) for final clips that need features like lip sync or multi-shot
4. Hero shots: Reserve Veo 3.1 (400 credits) for your most important scenes where 4K quality and native audio matter

Visit the

page for current credit packages.

Tips for Great Video Results

Start with image-to-video: Generating an image first and then animating it gives you far more control than text-to-video alone. Iterate on cheap images before spending video credits.

One action per clip: Keep your prompts focused on a single movement or action. Multi-action prompts often produce confused results.

Use camera presets when available: LTX 2 Pro's Dolly, Jib, and Static presets produce smoother camera motion than prompt-based direction.

Plan your shots like a real film: Think in terms of establishing shots, close-ups, and reaction shots. Generate each as a separate clip and combine them in the export.

Match models to your needs: Do not use Veo 3.1 for every clip. Use affordable models for B-roll and transition shots, and save premium models for key moments.

Ready to create your first video? Open the

and start generating. For still frames and character design, check out the guide.