WAN 2.6 Text-to-Video
WAN 2.6 text-to-video: 5–15s at 720p/1080p, native sound, seed, negative prompt and prompt expansion. 30–40 credits per second, audio upload supported.
WAN 2.6 Text-to-Video turns a text prompt into a cinematic clip at 720p or 1080p. It is the control-oriented WAN generation: seed, negative prompt and prompt expansion are all exposed, and you can add sound or upload an audio track up to 15 MB. Clips run 5, 10 or 15 seconds. On Conut.ai pricing is per second by resolution: 30 credits at 720p and 40 at 1080p.
- Resolution: 720p / 1080p
- Duration: 5s / 10s / 15s
- Price: 30–40 credits/s
- Audio: Sound + upload
How to use WAN 2.6 Text-to-Video
Start with the shot structure, then add the controls. Multi-shot is what separates a scene from a single moving image.
- Write the scene and decide the shot count Single-shot keeps one continuous camera move. Multi-shot lets the clip move between beats, which suits a 10- or 15-second mini-scene.
- Choose the size for the destination 1280×720 for landscape feeds, 720×1280 for vertical. 1080p sizes cost 40 credits per second instead of 30.
- Set duration to fit the beats Five seconds for one action, ten for a short sequence, fifteen for a scene with a cut. Do not stretch one idea across fifteen seconds.
- Guide the audio Leave the sound generated, or upload a WAV/MP3 up to 15 MB when the clip must follow a specific track or voice.
- Tune with negative prompt, expansion and seed Use the negative prompt for recurring artefacts, disable expansion for exact scripts, and lock the seed when comparing wording.
Settings reference
WAN 2.6 Text-to-Video is the control-oriented WAN text model: seed, negative prompt and prompt expansion are exposed, and audio can be generated or uploaded.
| Setting | Value | Notes |
|---|---|---|
| Prompt | Required | Describe scene, characters, motion, camera moves and style. |
| Audio | Generated + upload up to 15 MB | Native sound is generated; an uploaded track guides the audio when provided. |
| Duration | 5s / 10s / 15s | Default is 5 seconds. |
| Size | 1280×720 / 720×1280 / 1920×1080 / 1080×1920 | Landscape and portrait at 720p and 1080p. Default is 1280×720. |
| Shot type | Single / Multi | Multi-shot lets the model cut between beats inside one clip. |
| Negative prompt | Optional | Name artefacts to avoid, such as warped faces or text. |
| Prompt expansion | Optional toggle | Expands sparse prompts; disable it when the wording is fixed. |
| Seed | Optional | Reuse a seed to compare prompt changes against the same baseline. |
What WAN 2.6 text-to-video is good at
WAN 2.6 focuses on believable motion and professional cinematography. It handles weight, momentum and environmental interaction well, and it produces a cleaner 1080p image than the 2.5 generation. The result is less stylized than Kling and more grounded than the average text-to-video model, which makes it a good fit for product films, documentary-style inserts and realistic scenes.
The control surface is the other reason to choose it. A seed makes a good generation repeatable, a negative prompt suppresses unwanted elements, and prompt expansion fills in cinematic detail when your prompt is short. Those three controls together give more reproducibility than most endpoints in this price range.
- Realistic physics and natural motion
- 720p and 1080p output
- Seed for repeatable results
- Negative prompt for unwanted elements
- Prompt expansion for short prompts
WAN 2.6 pricing by resolution
Pricing is straightforward: 30 credits per second at 720p and 40 at 1080p. The difference between the two tiers is small enough that 1080p is practical for final delivery, while 720p remains the cheaper review format.
- Billed per second by output resolution
- No 480p tier; 720p is the minimum
- 15-second maximum
- Four size presets: 1280×720, 720×1280, 1920×1080, 1080×1920
| Resolution | Credits/s | 5s | 10s | 15s |
|---|---|---|---|---|
| 720p | 30 | 150 | 300 | 450 |
| 1080p | 40 | 200 | 400 | 600 |
Seed, negative prompt and prompt expansion
The seed is the most practical control. Generate a clip you like, note the seed, and reuse it when you change one variable, such as a camera move or a lighting direction. That turns prompt writing into a process of controlled comparisons instead of starting from scratch each time.
Negative prompts are best kept short and concrete. List the things that would ruin the shot, such as text, extra fingers, watermark or distorted faces, rather than abstract qualities. Prompt expansion is useful when you have a simple idea and want the model to add camera and lighting detail; turn it off when you need the prompt followed literally.
- Reuse the seed when testing one change at a time
- Keep negative prompts to concrete unwanted elements
- Turn off prompt expansion for literal prompts
- Use expansion to add camera and light detail to short prompts
- Single-shot and multi-shot modes are selectable
- Change one variable per generation when comparing
Audio options and prompting
WAN 2.6 can generate sound with the clip or accept an uploaded audio track up to 15 MB. Generated sound is best for ambience and effects that match the action. An uploaded track is useful when you already have a voiceover, a music bed or a sound design pass that the clip must follow.
Write the visual prompt as a shot description and keep the audio request separate. State the subject, action, camera and light, then add the sounds the scene produces. If you are uploading audio, describe the visual timing so cuts and actions land against the track.
- Generated sound for ambience and effects
- Audio upload up to 15 MB
- Describe visual timing when using an uploaded track
- Keep sound cues physical rather than musical
- Use 5-second clips to test sync before a longer render
Limitations and alternatives
There is no 480p tier and no 4K. Clips stop at 15 seconds, so longer sequences need multiple generations. The endpoint is more expensive per second than WAN 2.5 at 480p and 720p, but it adds 1080p, stronger physics and a fuller control set.
WAN 2.5 is the budget option for talking shots and one-pass A/V sync. Wan 3.0 adds 2–30 second durations, audio and thinking mode at 23/45/90 credits per second. Seedance 2.5 is the choice for 4K.
- 15-second maximum
- 720p minimum; no 4K
- No aspect-ratio selector beyond the four size presets
- For budget A/V sync, use WAN 2.5
- For 30-second clips, use Wan 3.0
Prompt examples for WAN 2.6 Text-to-Video
- Automotive reveal: A black sports car sits in a concrete garage as overhead lights switch on one by one, camera tracks slowly around the front quarter, reflections sliding across the paint, cool industrial light. Relay clicks, deep room tone, distant city hum. Use a fixed seed while testing light direction.
- Documentary portrait: A farmer stands at the edge of a wheat field at golden hour, wind moving the crop, camera slowly pushes in, natural light, shallow depth of field, documentary color grade. Wind, insects, distant tractor. A short prompt plus expansion gives a natural result.
- Kitchen macro: Butter melts in a hot pan, garlic sizzling, steam rising, macro camera slowly pushes in, warm tungsten light, shallow depth of field. Sizzle, knife on wood, quiet kitchen ambience. Negative prompt: text, hands, utensils warping.
- Storm coastline: Waves crash against a rocky coastline under a storm sky, spray rising, camera tracks along the cliff edge, cold blue and grey palette, volumetric clouds. Wind, crashing waves, distant thunder. 15 seconds gives the wave cycle room to complete.
- Night market: A crowded night market seen from eye level, steam rising from food stalls, neon signs reflecting on wet pavement, camera walks forward through the crowd, warm and cool mixed light. Chatter, sizzling, distant music. Walking camera plus crowd motion is a strong realism test.
- Product macro: A watch lies on dark stone as the camera orbits slowly, a narrow light sweeping across the dial and bracelet, dust particles in the beam, luxury commercial look. Soft mechanical ticking, deep room tone. Use 1080p for fine product detail.
Pro tips for WAN 2.6 Text-to-Video
2.6 is the most controllable WAN text model. Use the controls to run experiments, not to decorate every prompt.
- Multi-shot for narrative, single for product A product rotation should stay in one continuous shot. A story beat benefits from a cut. Pick the structure before writing.
- 1080p is a 33 percent premium Forty credits per second versus thirty. Draft at 720p and re-render only the selected take at 1080p.
- Upload audio when the track is fixed Generated sound is convenient but variable. A fixed music bed or voiceover should be uploaded so the picture follows it.
- Keep the negative prompt short A handful of artefact terms works better than a long exclusion list. Too many negatives flatten useful detail.
- Use the seed for A/B tests Change one prompt clause at a time with the same seed. That is the only way to learn what the wording actually does.
Frequently asked questions
How much does WAN 2.6 text-to-video cost?
It costs 30 credits per second at 720p and 40 credits per second at 1080p. A 5-second 720p clip is 150 credits; a 15-second 1080p clip is 600.
How long can a WAN 2.6 clip be?
5, 10 or 15 seconds. Fifteen seconds is the maximum.
Does WAN 2.6 support a seed?
Yes. A seed is exposed, which makes a generation repeatable when you change one prompt variable at a time.
Does WAN 2.6 support negative prompts?
Yes. Use a short list of concrete unwanted elements such as text, watermarks or distorted faces.
Can I upload my own audio?
Yes. The endpoint accepts an audio track up to 15 MB, and it can also generate sound with the clip.
What resolutions are available?
720p and 1080p, in landscape or portrait. There is no 480p or 4K tier.
What is prompt expansion?
It lets the model add cinematic detail to a short prompt. Turn it off when you need the prompt followed literally.
What is the difference from Wan 3.0?
Wan 3.0 adds 2–30 second durations, optional audio and deep-thinking mode, but it costs more per second. WAN 2.6 caps at 15 seconds and focuses on control and 1080p value.
Compare with other text-to-video models
- WAN 2.6 Image-to-Video
- WAN 2.5 Text-to-Video
- Wan 3.0 Text-to-Video
- Kling O3 Pro Text-to-Video
- Seedance 2.5 Text-to-Video
- Browse all AI video models
Specifications
- Provider: Wavespeed
- Model type: Text to video
- Credit pricing: 720p: 30 credits · 1080p: 40 credits
- Capabilities: new, cinematic, starter, 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.