WAN AI: Text-to-Video and Image-to-Video Made Simple

What Is WAN AI?

WAN is a powerful family of AI video generation models that excels at creating high-quality 1080p video clips from text descriptions or still images. In FP AI Studio, you get access to two distinct WAN models: WAN Text-to-Video (wan-2-5-t2v-1080p) and WAN Image-to-Video (wan-2-5-i2v-1080p). Both models share a common set of controls but serve different creative purposes, making them a versatile pair for any video project.

What sets WAN apart from other video models in the app is its combination of prompt expansion, negative prompt support, and clean 1080p output. Whether you are generating a video purely from a text concept or animating an existing photograph, WAN gives you the tools to produce polished results with minimal trial and error.

Text-to-Video Mode

The WAN Text-to-Video model (wan-2-5-t2v-1080p) generates a video clip entirely from your written prompt. You do not need any reference image — just a description of the scene you want to see.

Step-by-Step

1. Open FP AI Studio and navigate to the Video Generation tab.
2. Select WAN Text-to-Video (wan-2-5-t2v-1080p) from the model dropdown.
3. Write your prompt. Be specific about the scene, camera movement, lighting, and mood. For example: "A golden retriever running along a sandy beach at sunset, waves crashing in the background, warm cinematic lighting, slow motion."
4. Optionally add a negative prompt to exclude unwanted elements.
5. Choose your duration: 5 seconds or 10 seconds.
6. Toggle prompt expansion on or off depending on your preference.
7. Set a seed value if you want reproducible results, or leave it blank for a random seed.
8. Tap Generate and wait for your video.

Text-to-Video is ideal when you have a concept in mind but no reference visual. The model handles composition, color, and motion entirely based on your description, giving you a complete video from scratch.

Image-to-Video Mode

The WAN Image-to-Video model (wan-2-5-i2v-1080p) takes a still image and brings it to life with AI-generated motion. This mode requires an image URL as input in addition to your text prompt.

Step-by-Step

1. Select WAN Image-to-Video (wan-2-5-i2v-1080p) from the model dropdown.
2. Paste the image URL of your reference image into the image input field. This can be an image you previously generated in the app or any publicly accessible image URL.
3. Write a prompt describing the motion you want. Focus on what should move and how, rather than describing the image itself. For example: "Camera slowly zooms in, hair blowing in the wind, subtle smile appears."
4. Add a negative prompt, choose duration (5s or 10s), set prompt expansion, and configure seed as needed.
5. Tap Generate.

Image-to-Video is perfect when you already have a strong visual — perhaps an AI-generated image from Flux or Mystic, a stock photo, or even a personal photograph — and you want to add cinematic motion to it. The model preserves the composition and style of your input image while intelligently adding movement.

Prompt Expansion Explained

Both WAN models include a prompt expansion toggle that can dramatically improve your results. When enabled, the AI takes your original prompt and expands it with additional descriptive detail before generating the video. This means a short prompt like "cat sleeping on a couch" might be internally expanded to include details about lighting, camera angle, fur texture, and ambient atmosphere.

Prompt expansion is especially useful if you tend to write brief prompts or are not sure how to describe cinematic qualities. It acts as a built-in prompt engineer that fills in the gaps. However, if you have already written a highly detailed prompt and want precise control over every aspect, you may want to turn prompt expansion off to avoid the model adding unwanted details.

Our recommendation: Start with prompt expansion on. If the results add elements you did not intend, turn it off and write a more detailed prompt manually.

Negative Prompts for Better Quality

Negative prompts tell the model what to avoid in the generated video. Both WAN models support this feature, and using it effectively can significantly improve output quality.

• For general quality: "blurry, low resolution, distorted, flickering, artifacts, watermark"
• For human subjects: "deformed hands, extra fingers, unnatural face, morphing body parts"
• For landscapes: "oversaturated, flat lighting, repetitive textures, static sky"

Think of the negative prompt as a guardrail. It does not guarantee the exclusion of every listed element, but it steers the generation away from common problems. Even a simple negative prompt like "blurry, low quality, distorted" makes a noticeable difference in most generations.

Seed Control & Reproducibility

Both WAN models support seed control. A seed is a numerical value that initializes the random generation process. When you use the same seed with the same prompt and settings, you get the same (or very similar) output. This is invaluable for iterative workflows.

For example, if you generate a video you mostly like but want to tweak the prompt slightly, keeping the same seed ensures the overall composition stays consistent while the changes you made take effect. Without a fixed seed, every generation is completely random, making it hard to iterate on a concept.

To use seed control, simply enter any integer value in the seed field. Leave it empty or set to 0 for a random seed each time.

Practical Examples

Example 1: Cinematic Landscape (Text-to-Video)

Prompt: "Aerial drone shot of a misty mountain valley at dawn, rays of sunlight breaking through clouds, a river winding through evergreen forests, cinematic 4K quality"

Negative prompt: "blurry, shaky camera, oversaturated, artificial looking"

Duration: 10 seconds | Prompt expansion: On

Example 2: Product Animation (Image-to-Video)

Image: A still product photo of a wristwatch on a marble surface

Prompt: "Camera slowly rotates around the watch, light reflections glide across the glass face, shallow depth of field"

Negative prompt: "blurry, flickering, distorted reflections"

Duration: 5 seconds | Prompt expansion: Off

Example 3: Character Portrait (Image-to-Video)

Image: An AI-generated portrait of a fantasy warrior

Prompt: "Hair blowing gently in the wind, eyes shift to look at camera, subtle breathing motion, dramatic lighting"

Negative prompt: "morphing face, extra limbs, flickering"

Duration: 5 seconds | Prompt expansion: On

WAN's dual-model approach gives you the flexibility to generate videos from pure imagination or from existing visuals. Combined with prompt expansion, negative prompts, seed reproducibility, and flexible duration options, these two models cover a wide range of creative video needs right inside FP AI Studio.