WAN 2.2 Text-to-Video 5B
WAN 2.2 5B generates a 720p clip from a text prompt for a flat 20 credits. A low-cost MoE model for drafts, b-roll and rapid prompt testing.
WAN 2.2 Text-to-Video 5B is the low-cost entry point to Alibaba's WAN 2.2 family. It turns a text prompt into one 720p clip for a flat 20 credits per run, regardless of how many times you retry within a session. The 5B Mixture-of-Experts model is built for drafts, b-roll and prompt testing where per-second pricing would add up quickly.
- Price: 20 credits / clip
- Resolution: 720p
- Input: Text prompt
- Audio: Silent
How to use WAN 2.2 Text-to-Video 5B
Use 5B as the cheap exploration tier. It is built for drafts, b-roll and prompt testing, not for hero shots.
- Write one scene per run A single subject with one action gives the 5B model the best chance. Crowded prompts are where the low-capacity model shows its limits.
- Keep the framing landscape The output is fixed at 1280×720. If you need vertical, plan to crop in the edit or use a model with portrait sizes.
- Use the flat price to iterate Each run costs 20 credits. Run several prompt variants back to back and pick the best motion instead of polishing one prompt blind.
- Lock the seed for comparisons When the only variable should be wording, reuse the seed. That makes the differences between runs attributable to the prompt.
- Move the winner to a timed model Once the idea works, re-render it on WAN 2.2 Ultra-Fast or Image-to-Video where duration, aspect ratio and resolution are controllable.
Settings reference
WAN 2.2 Text-to-Video 5B is the flat-price draft model: one prompt, one fixed 720p landscape clip, 20 credits per run.
| Setting | Value | Notes |
|---|---|---|
| Prompt | Required | Describe the scene, motion and style. One clear idea works better than a crowded one. |
| Size | 1280×720 | Fixed landscape 720p. There is no portrait or square option on this endpoint. |
| Price | 20 credits per video | Flat per-run price regardless of prompt or retries; there is no per-second billing. |
| Duration | Fixed | No duration selector. The endpoint returns one fixed-length clip. |
| Seed | Optional | Reuse a seed to compare prompt variants on the same baseline. |
| Audio | Silent | No native audio and no audio upload on this endpoint. |
What the 5B model offers
WAN 2.2 uses a Mixture-of-Experts architecture with high-noise and low-noise experts split by denoising timestep. The 5B variant keeps that structure in a smaller, cheaper model. It understands lighting, colour, composition and camera language well enough for cinematic-looking b-roll, and it handles multi-object scenes better than older budget models.
The trade-off is resolution and control. Output is fixed at 720p in a 16:9 frame, there is no resolution menu, no aspect-ratio menu and no audio. It is also a text-only endpoint: you cannot supply a starting image. Treat it as a fast sketch tool, not a finishing model.
- MoE architecture with 5B parameters
- One 720p landscape clip per run
- Flat price independent of retries
- Text-only input
- Silent output
Flat pricing versus per-second models
At 20 credits per clip, WAN 2.2 5B is the cheapest way to see a prompt move on Conut.ai. WAN 2.2 Ultra-Fast charges 10 credits per second, so a 5-second clip costs 50 and an 8-second clip costs 80. The 5B model costs less than half of a 5-second Ultra-Fast run and roughly a quarter of an 8-second run.
The comparison is not purely financial. Ultra-Fast lets you choose 5 or 8 seconds and landscape or vertical framing; 5B returns a single fixed-format clip. If you need a vertical deliverable or a specific length, the cheaper model may cost more in wasted retries. Use 5B to explore the idea, then move the winning prompt to a model that matches the delivery spec.
- 20 credits per generated clip
- No duration-based price increase
- Ultra-Fast costs 50 credits for 5 seconds
- 5B is best for exploration, not delivery
- Retries are cheap enough to test several prompts
| Model | Format | Price | Best for |
|---|---|---|---|
| WAN 2.2 5B | 720p landscape | 20 credits flat | Drafts and prompt tests |
| WAN 2.2 Ultra-Fast | 720p, 5s or 8s | 10 credits/s | Cheap timed clips |
| WAN 2.2 Image-to-Video | 480p/720p, 5s or 8s | 30–60 credits/s | Animating a still |
Prompting a small model well
Smaller models reward concrete prompts. Instead of a mood, describe visible facts: who is in frame, what they are doing, where the light comes from and how the camera behaves. An old man adjusts his glasses while reading a newspaper in a sunlit park is a better prompt than a peaceful morning, because every word points to something the model can render.
Keep the action within a single beat. The endpoint does not expose a duration control, so a prompt that describes a beginning, middle and end may be compressed or truncated. One clear action, one camera behaviour and one lighting condition produce more consistent results than a short story.
- Describe visible facts, not moods
- One action per prompt
- Name the camera behaviour
- State the light source and direction
- Avoid multi-beat narratives
When the 5B model is the right choice
Use it when you are still deciding what the shot should be. Twenty credits lets you test five directions for the price of one premium second. It is also useful for background b-roll where 720p is enough, and for filling a storyboard with moving thumbnails before committing to a final model.
Move to a larger model when the clip needs 1080p, audio, vertical framing, a specific duration or a supplied starting image. WAN 2.2 Ultra-Fast handles timed 720p clips; WAN 2.6 adds 1080p and sound; HappyHorse adds five aspect ratios and a 3–15 second range.
- Prompt exploration and storyboard thumbnails
- Low-cost b-roll at 720p
- Not for 1080p, audio or vertical delivery
- Not for image-led shots
- Upgrade once the concept is approved
Settings, limits and reproducibility
The endpoint accepts a prompt and an optional seed. A fixed seed plus an unchanged prompt is the only way to reproduce a result; changing either produces a new clip. There is no negative prompt on this variant, so avoid describing things you do not want and instead describe the frame you do want.
Output is 720p, landscape, silent and fixed in length. There is no upscale step built in. If the final delivery needs 1080p, plan to regenerate on a higher-tier model rather than upscaling a 720p draft, because upscaling cannot recover detail the model never generated.
- Prompt plus optional seed
- No negative prompt
- 720p landscape only
- No built-in upscale
- Regenerate at a higher tier for 1080p delivery
Prompt examples for WAN 2.2 Text-to-Video 5B
- Street portrait: An old man adjusts his glasses while reading a newspaper on a sunlit park bench, pigeons on the path, soft morning light from the left, camera slowly pushes in. Concrete action and light direction.
- Product on a table: A white ceramic coffee cup steams on a wooden table beside a folded newspaper, warm window light, shallow depth of field, camera locked off, quiet morning atmosphere. Locked-off shots are reliable at 720p.
- City b-roll: A red tram crosses a stone bridge at dusk, city lights reflecting in the river, camera tracks parallel to the tram, cool blue hour with warm windows. One moving subject keeps the frame readable.
- Nature detail: Raindrops slide down a green leaf, a small insect clings to the edge, soft overcast light, macro lens, camera holds steady with a slight drift. Macro shots hide the resolution limit.
- Interior scene: A woman writes in a notebook at a cafe window, steam rising from her cup, afternoon light falling across the page, camera slowly pans right. State the camera move explicitly.
- Action beat: A cyclist rounds a corner on a wet city street, water spraying from the tyres, camera tracks from the side, overcast light, motion blur in the background. Keep the action to one beat.
Pro tips for WAN 2.2 Text-to-Video 5B
The 5B model is a sketchbook. Treat its output as a decision aid, not a deliverable.
- Twenty credits per idea is the point The flat price makes broad exploration cheap. Generate four or five directions before committing to a final model.
- Avoid fine detail in the prompt Text, hands and complex interactions are the first things a 5B model loses. Keep the scene readable at thumbnail size.
- 720p landscape only There is no resolution or aspect control. Build the brief around landscape 16:9 rather than fighting the endpoint.
- Use it for b-roll Abstract backgrounds, weather, crowds and simple motion hold up well. Character acting and dialogue do not.
- Graduate to Ultra-Fast for length control When the clip needs 5 or 8 seconds, vertical framing or a negative prompt, WAN 2.2 Ultra-Fast is the next step up.
Frequently asked questions
How much does WAN 2.2 5B cost?
20 credits per generated clip, regardless of retries or prompt changes. There is no per-second charge.
What resolution does it output?
720p in a 16:9 landscape frame. There is no resolution or aspect-ratio selector.
How long is the clip?
The endpoint returns one short fixed-length clip per run. Duration is not exposed as a setting, unlike Ultra-Fast, which offers 5 or 8 seconds.
Does it generate audio?
No. The output is silent. Use WAN 2.6 Text-to-Video if you need generated sound.
Can I upload a starting image?
No. This is a text-to-video endpoint. Use WAN 2.2 Image-to-Video or HappyHorse Image-to-Video for image-led shots.
Does it support a seed?
Yes. Reuse the seed with the same prompt to reproduce a result.
Is it better than WAN 2.2 Ultra-Fast?
It is cheaper per clip, but Ultra-Fast gives you control over duration and vertical framing. Choose based on whether you need a specific length or format.
Can I upscale the result to 1080p?
You can run it through Video Upscaler, but upscaling cannot recover detail the model never generated. For true 1080p, regenerate on a higher-tier model.
Compare with other WAN text-to-video models
- WAN 2.2 Ultra-Fast Text-to-Video
- WAN 2.2 Image-to-Video
- WAN 2.5 Text-to-Video
- WAN 2.6 Text-to-Video
- HappyHorse 1.0 Text-to-Video
- Browse all AI video models
Specifications
- Provider: Wavespeed
- Model type: Text to video
- Credit pricing: 20 credits per video
- Capabilities: popular, cinematic, realistic
About Conut.ai
Conut.ai is an AI image, video and audio generation platform with 50+ creative tools. It supports text-to-image, image-to-video, face swap, motion control, relighting, inpainting and video effects through models including Nano Banana Pro, Seedream, GPT Image, Sora 2, Kling, WAN and Seedance. Start free — no credit card required.
Explore more AI tools
- Pixverse V5.5 Effects
- Seedance 2.5 Text-to-Video
- Seedance 2.5 Turbo Text-to-Video
- Seedance 2.5 Image-to-Video
- Seedance 2.5 Turbo Image-to-Video
- Seedance 2.0 Mini Text-to-Video
- Seedance 2.0 Mini Image-to-Video
- MiniMax H3 Text-to-Video
- MiniMax H3 Image-to-Video
- Photo Colorizer
- Age Transform
- Virtual Try-On