WAN 2.7 Video is Alibaba's flagship video model on PixelDojo — text-to-video, image-to-video, and a video-extend mode that continues an existing clip. 720p (2.5 credits/s) and 1080p (3 credits/s), durations 2-15 seconds, 5 aspect ratios, with prompt-enhancement built in.
WAN 2.7 Video is the top-of-line in the WAN family — Alibaba's flagship video generation model. Where it stands out: three distinct modes in one model (text-to-video for composing from a prompt, image-to-video for animating a starting frame, video-extend for continuing an existing clip), durations from 2 to 15 seconds, and built-in prompt enhancement.
Two resolution tiers (720p at 2.5 credits/s, 1080p at 3 credits/s) and five aspect ratios — 16:9 for cinematic, 9:16 for vertical, 1:1 for square social, 4:3 / 3:4 for traditional photo proportions. Up to 15 seconds in a single clip — longer than most video models on PixelDojo.
3
Modes (T2V · I2V · Extend)
2–15s
Duration range
720p · 1080p
Resolution tiers
Text-to-video composes from a prompt alone. Image-to-video animates a starting frame from your prompt. Video-extend takes an existing clip and writes a continuation — useful for stretching short clips into longer narratives without re-rendering.
720p (2.5 credits/s) for iteration and social. 1080p (3 credits/s) for hero shots, atmospheric scenes, and material-heavy product work. 20% credit premium for 1080p — usually worth it for the final render.
WAN 2.7 supports up to 15 seconds in a single clip — most video models cap at 6 or 10. Useful for atmospheric scenes that need breathing room, multi-shot sequences, or narrative beats that don't fit in 5 seconds.
Pass a source video and prompt — WAN 2.7 writes a continuation that picks up from the final frame. Great for stretching a 5-second clip you like into a 10-15 second sequence without re-rendering the original.
Each example shows the exact prompt that produced the result. Copy any prompt with one click.
1080p · 16:9 · 8s · 24 credits
Vibrant cinematic establishing shot drifting forward over a futuristic coastal city at golden hour, neon-lined boardwalks, glass towers reflecting amber light, distant ocean waves crashing, warm sunset color grade, the camera moves at a very slow forward dolly, hyper-detailed 8k
Cinematic establishing shots reward 1080p — atmospheric detail (waves, neon reflections, distant towers) compresses badly at 720p. Pair the 16:9 widescreen with ONE dominant camera move ('slow forward dolly', 'crane up', 'pull back') and let WAN handle pacing.
720p · 4:3 · 5s · 12.5 credits
Medium close-up of a young woman with curly auburn hair sitting at a sunlit cafe window, she lifts a porcelain teacup to her lips, takes a slow sip, then turns her head toward the camera with a soft smile, shallow depth of field, warm afternoon light, hyper-detailed skin texture and fabric weave
4:3 aspect lends portrait close-ups a slightly classical feel. WAN 2.7 reads micro-beats well — "lifts cup, takes a sip, turns head, smiles" plays cleanly. 720p is plenty for character close-ups; save the credit premium for atmospheric work.
1080p · 1:1 · 6s · 18 credits
Studio product shot of a matte-black ceramic coffee mug on a polished walnut surface, steam curling upward, the camera does a slow 90-degree rotational pan around the mug, soft top-down rim lighting against a charcoal gradient backdrop, photoreal advertising quality
1:1 square at 1080p with one explicit camera move (rotational pan, dolly in, push back) is the canonical product spot. WAN 2.7 handles material textures (matte ceramic, polished walnut, brushed metal) cleanly at 1080p.
1080p · 21:9 cropped to 16:9 · 10s · 30 credits
Wide cinematic landscape of a misty Norwegian fjord at dawn, mountain peaks emerging through low fog, a single distant fishing boat leaves a thin wake on glassy water, the camera does a very slow forward dolly, cool blue-green palette with warm amber highlight on one peak, sounds of distant water and gentle wind
10-second atmospheric clips reward 1080p — fog, water surface, distant detail all preserved. One slow camera move plus a "warm spot" element to break the cool palette. 16:9 widescreen is the natural fit; ultra-wide 21:9 isn't natively supported, so 16:9 is your widest.
720p · 16:9 · 12s · 30 credits
12-second action sequence. Shot 1: low-angle of a parkour runner sprinting toward a rooftop edge, wind whipping their jacket. Shot 2: mid-air shot of the runner leaping across a gap between buildings, city lights blurred behind. Shot 3: landing roll into a sprint that continues offscreen. Dynamic handheld camera throughout.
12s gives WAN 2.7 room for a true multi-shot sequence — three distinct beats with clean cuts. 720p is enough for action; cinematic atmosphere is where 1080p shines. Name each shot explicitly ("Shot 1: ... Shot 2: ...") for cleaner cuts.
Text-to-video when you have only a prompt. Image-to-video when you have a strong starting frame you want to animate. Video-extend when you have a 5-10 second clip you want to stretch to 15s without re-rendering. Same prompt language across all three.
1080p preserves fog, water, foliage, and material textures. 720p is plenty for human action, character close-ups, and most prompts where motion matters more than fine detail. 20% credit premium for 1080p — pick by use case.
WAN 2.7's 2-15s range is wider than most video models. Use 8-12 seconds for atmospheric scenes that need breathing room; 10-15s for genuine multi-shot sequences. Don't default to 5s if your story needs more.
'Slow forward dolly', 'crane up', '90-degree rotational pan', 'handheld follow' — WAN reads these literally. Combining moves ("dolly in then pan left") usually produces confused output. One move per shot is the rule.
10-15s clips benefit from explicit "Shot 1: ... Shot 2: ... Shot 3: ..." structure. WAN 2.7 renders cuts cleanly when you name them. Stuffing one beat into 15s wastes the duration — use multi-shot or pick a shorter clip.
WAN was tuned against an official Chinese negative prompt, and it works better left exactly as is. It already blocks subtitles. If you see stray text or watermarks, append 文字，水印，logo to the end. The full default string is in the negative prompt section of this guide.
The starting image already carries the subject, scene, and style. Spend your prompt on motion and camera only. Re-describing appearance wastes tokens and invites morphing away from your image.
WAN was tuned against an official default negative prompt, and that prompt is written in Chinese. The model learned those exact phrases during training, so it responds to them more reliably than to English translations. Leave the default in place and leave it in Chinese.
Here is the full default string: 色调艳丽，过曝，静态，细节模糊不清，字幕，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走
It already covers the common failure modes: oversaturation, overexposure, static frames, blurry detail, bad hands, extra fingers, deformed limbs, cluttered backgrounds, and subtitles (字幕 is in there). If you still see stray text, watermarks, or logos in your output, append 文字，水印，logo to the end. Do not rewrite the rest.
WAN 2.7 generates a soundtrack with the video. Describe the soundscape you want in the positive prompt: dialogue, ambient tone, specific sound effects. "Sounds of distant water and gentle wind" or "the espresso machine hisses in the background" both read cleanly.
Music is a positive-prompt control, not a negative-prompt one. Keep music terms out of the negative prompt. To keep a scene music-free, add the phrase "No background music." to your prompt. To get music, ask for it explicitly, like "suspenseful background music."
For spoken dialogue with lip-sync, write the action sentence first, then the line in this format: [Name, tone]: "line". Example: The agent slams his hand on the table. [Black-suited Agent, angrily]: "Where is the truth?"
Give every speaker a unique name and never tag a line with a pronoun. WAN matches each line to a character by name. Short single-speaker lines under 10 words sync best.
| Setting | Values | Notes |
|---|---|---|
| Mode | text-to-video · image-to-video · video-extend | I2V needs first_frame_image; video-extend needs source video. |
| Resolution | 720p (2.5 cr/s) · 1080p (3 cr/s) | 20% credit premium for 1080p. Worth it for atmospheric / material-heavy clips. |
| Duration | 2-15 seconds (integer) | Longer than most video models. Use 8-12s for atmospheric, 10-15s for multi-shot. |
| Aspect ratio | 16:9 · 9:16 · 1:1 · 4:3 · 3:4 | 5 presets covering common output frames. |
| Negative prompt | string (optional, max 500 chars) | Default works for most prompts. |
| Prompt enhancement | On / Off | Optional WAN-side prompt expansion. Try both; sometimes off produces cleaner output. |
720p (2.5 cr/s) for iteration, social posts, character work, and human action — plenty of fidelity for these. 1080p (3 cr/s) for atmospheric scenes (fog, water, foliage), product shots with detailed materials, and hero clips. The 20% credit premium is usually worth it for the final render.