WAN 2.2 Text-to-Video Ultra-Fast
WAN 2.2 Text-to-Video Ultra-Fast renders 720p clips in 5 or 8 seconds at 10 credits per second. A low-cost, high-throughput model for drafts and batches.
WAN 2.2 Text-to-Video Ultra-Fast is the high-throughput variant of WAN 2.2. It produces 720p clips in 5 or 8-second lengths, in landscape or vertical framing, at 10 credits per second. That makes a 5-second clip 50 credits and an 8-second clip 80 — cheap enough to generate several options before choosing one to refine.
- Duration: 5s / 8s
- Resolution: 720p
- Price: 10 credits/s
- Frames: 16:9 / 9:16
How to use WAN 2.2 Text-to-Video Ultra-Fast
Ultra-Fast is for volume. Write short prompts, generate several options and keep the best take.
- Write a compact prompt One subject, one action, one camera move. The fast path rewards clarity and punishes long, layered descriptions.
- Choose the framing 1280×720 for landscape, 720×1280 for vertical. There is no square option, so pick the one closest to the final edit.
- Pick 5 or 8 seconds Five seconds is the default and the cheapest way to explore. Eight seconds is for actions that need a full beat.
- Add a short negative prompt A few artefact terms, such as text or extra fingers, are enough. Long exclusion lists tend to flatten the image.
- Rerun with the same seed When a take works, keep its seed and adjust one clause at a time. That isolates what the prompt change did.
Settings reference
WAN 2.2 Ultra-Fast is the throughput model: two lengths, two framings, a negative prompt and a flat 10 credits per second.
| Setting | Value | Notes |
|---|---|---|
| Prompt | Required | Describe the scene, motion and style. |
| Duration | 5s / 8s | Default is 5 seconds. Longer clips bill linearly. |
| Size | 1280×720 / 720×1280 | Landscape or vertical, both at 720p. Default is 1280×720. |
| Negative prompt | Optional | Name artefacts to suppress, such as text, extra limbs or warping. |
| Seed | Optional | Reuse a seed to keep the motion while changing the prompt. |
| Price | 10 credits per second | A 5-second clip costs 50 credits; an 8-second clip costs 80, in either framing. |
Where Ultra-Fast fits
Ultra-Fast keeps the WAN 2.2 aesthetic and motion model but optimises for throughput. It is the endpoint to use when you need many variations quickly: testing a hook for a short-form video, filling a storyboard, or producing a batch of background clips where 720p is sufficient.
It is not a lower-quality afterthought. The trade is control, not just fidelity: 720p maximum, 5 or 8 seconds, landscape or vertical, no audio and no image input. Within those limits it is one of the cheapest timed video endpoints available.
- 10 credits per second
- 5 or 8-second clips
- 1280×720 or 720×1280
- Text prompt required
- Silent output
Pricing at 720p
Billing is output seconds multiplied by 10. A 5-second landscape clip is 50 credits; an 8-second landscape clip is 80. The same prices apply to vertical output. There is no per-run fee and no resolution tier to choose, so the only cost lever is length.
Compare that with WAN 2.2 Image-to-Video at 60 credits per second at 720p: a 5-second still animation costs 300. Ultra-Fast is a different job, but the gap shows why it is the default for exploration and high-volume b-roll.
- 5 seconds = 50 credits
- 8 seconds = 80 credits
- Same price for vertical
- No per-run fee
- Cheapest timed WAN 2.2 text endpoint
| Duration | Landscape | Vertical | Credits |
|---|---|---|---|
| 5s | 1280×720 | 720×1280 | 50 |
| 8s | 1280×720 | 720×1280 | 80 |
Prompts that survive a fast render
Fast models have less time to resolve ambiguity, so prompts should be front-loaded with the subject and action. Put the most important noun and verb in the first sentence, then add camera and light. A prompt that opens with atmosphere and only reveals the subject at the end often produces a generic frame.
Use the negative prompt for persistent artefacts rather than long lists. If hands are consistently malformed, add deformed hands; if the model keeps adding text, add text, watermark. Keep the negative list short so it does not fight the positive prompt.
- Subject and action first
- Camera and light second
- One main action per clip
- Short, targeted negative prompt
- Avoid contradictory instructions
Seeds, batches and reproducibility
The endpoint supports a seed, which is useful for batch work: fix the seed to compare prompt variants against the same starting noise, or leave it random to explore. Because the clips are cheap, generating four variations is often faster than perfecting one prompt.
Vertical output is selected through the size parameter rather than an aspect-ratio menu. If the same concept must ship in both formats, generate each separately; do not crop a landscape render, because the composition was built for the wider frame.
- Seed support for controlled comparisons
- Generate several variants cheaply
- Vertical is a separate render
- Do not crop landscape to vertical
- No audio track
Limits and upgrade paths
Ultra-Fast stops at 720p and 8 seconds. It has no audio, no reference image and no first-and-last-frame control. When a clip needs any of those, move to WAN 2.6 Text-to-Video for 5/10/15 seconds with audio and 1080p, or HappyHorse for a continuous 3–15 second range and five ratios.
If the output is right but the resolution is not, regenerate the winning prompt on a higher-tier model rather than upscaling. Upscaling a 720p fast render can sharpen edges, but it cannot add the detail a 1080p native render would have produced.
- 8-second maximum
- 720p maximum
- No audio or image input
- WAN 2.6 for 1080p and sound
- Regenerate rather than upscale for delivery
Prompt examples for WAN 2.2 Text-to-Video Ultra-Fast
- Short-form hook: A chef flips a bright green herb salad in a smoking wok, flames leaping toward the ceiling, camera pushes in fast, warm kitchen light, steam and sparks in the air. Lead with the action for a 5s hook.
- Vertical product: A smartphone rotates above a reflective black podium, screen glowing with a blue interface, camera orbits slowly, dark studio with soft rim light, clean commercial look. Render 720×1280, do not crop.
- Atmosphere b-roll: Rain streaks down a train window at night, city lights blur into long streaks, camera holds steady, cool blue and amber reflections, quiet mood. Simple ambient motion is reliable.
- Nature loop: Waves roll onto a black sand beach under a grey sky, sea foam spreading across the shore, camera slowly tilts down, cold muted palette. Use 8s for a calmer loop.
- Character moment: A detective in a trench coat lights a cigarette under a streetlamp, rain falling through the beam, camera slowly pushes in, high contrast noir lighting. Keep the character action single.
- Abstract background: Sheets of translucent blue and silver fabric ripple in slow motion, light refracting through the folds, camera drifts forward, dark void background. Good for text-overlay backgrounds.
Pro tips for WAN 2.2 Text-to-Video Ultra-Fast
At 10 credits per second, iteration is cheap. The constraint is prompt discipline, not budget.
- Ten credits per second is the floor A 5-second clip is 50 credits and an 8-second clip is 80. Generate a batch of options for less than one premium-model run.
- Vertical is a native option 720×1280 is not a crop. Generate vertical for shorts and reels rather than reframing landscape footage.
- Keep prompts under two sentences The fast model has less room to reconcile competing instructions. Short prompts produce more consistent motion.
- Use the negative prompt for text Garbled lettering is the most common artefact in fast generations. Adding text to the negative prompt is usually enough.
- Batch, then select Generate three or four variations with different seeds, watch them side by side and keep one. This is faster than tuning a single prompt.
Frequently asked questions
How much does WAN 2.2 Ultra-Fast cost?
10 credits per second. A 5-second clip is 50 credits and an 8-second clip is 80.
What durations are available?
Only 5 or 8 seconds.
Can I generate vertical video?
Yes. Choose 720×1280 for vertical or 1280×720 for landscape. Both cost the same per second.
Does it support negative prompts?
Yes. Use a short list of persistent artefacts rather than a long paragraph of exclusions.
Does it generate audio?
No. Use WAN 2.6 Text-to-Video for native audio.
Can I use an image as the first frame?
No. Use WAN 2.2 Image-to-Video instead.
Is it suitable for final delivery?
For 720p social or web delivery, yes. For 1080p or larger screens, regenerate on a higher-tier model rather than upscaling.
How does it compare with WAN 2.2 5B?
5B costs a flat 20 credits for a fixed 720p clip; Ultra-Fast costs 10 credits per second but lets you choose 5 or 8 seconds and vertical framing.
Compare with other fast text-to-video models
- WAN 2.2 Text-to-Video 5B
- 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: 10 credits per second
- Capabilities: popular, cinematic, realistic, starter, ultra, fast, multi-shot
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.