Models

Wan 2.7 vs Wan 2.6

Model Wan 2.7Format Short dramaMarket Global English
Short answer

They do different jobs rather than the same job at different quality. Wan 2.7 composes video from text or image references — eight references, images, clips or audio. Wan 2.6 restyles footage you already have, takes three references, and **won't accept an image at all.** So the version numbers are misleading: this isn't an upgrade path.

Adjacent version numbers usually mean one replaced the other. Here they mean two tools that happen to share a family name, and the giveaway is in a line of grey text under the reference field.

What are they each for?

Their own descriptions answer it, and they don't overlap.

Wan 2.7
"Reference-composed video, text or image in." A model for building a shot from a description plus references — images, clips or audio, up to eight of them.
Wan 2.6
"Restyle a clip you already have." A model for changing the treatment of existing footage. It accepts clips or audio and does not accept images.

Wan 2.6 won't take an image reference. That single fact tells you it isn't a newer Wan 2.7.

What actually differs?

Four rows, and the reference behaviour is the substantive one.

Wan 2.7Wan 2.6
What it's forReference-composed video, text or image inRestyle a clip you already have
Duration**2–15s****5–15s**
Resolution720 / 1080720 / 1080
References**8****3**
Accepted typesImages, clips or audio**Clips or audio — no images**
Aspect ratios16:9, 9:16, 1:1, 4:3, 3:416:9, 9:16, 1:1, 4:3, 3:4
AudioAlways onAlways on
FramesStart and end**Start only**
Prompt length5,000 charactersNot shown

Source: Hexcoded Creative Studio picker, Video → Create, September 2026.

Four differences, and they all point the same way.

Wan 2.7 starts at two seconds; Wan 2.6 starts at five. Two seconds is the shortest minimum on the entire roster, which makes 2.7 usable for an insert or a single beat in a way 2.6 isn't.

Eight references against three. Enough for a character, a location, a style and a prop on 2.7. Three is enough for a source clip and a look.

Images against no images. This is the one that settles it. A model that won't take a still image isn't a general-purpose generator — it's a model whose input is footage.

Start and end frames against start only. So 2.7 supports keyframing between two anchors and 2.6 doesn't.

Where this falls short. The picker exposes what each accepts, not how well either performs. Whether Wan 2.6's restyling is good is a question the settings can't answer and we haven't measured.

Why does the restyling model sit in Create?

Because it isn't an editing model, and this catches people.

Wan 2.6 is in Video Create, not Video Edit. Its job is restyling footage, which sounds like editing — but the mechanism is generating a new clip conditioned on a video reference rather than modifying an existing one in place.

So if you're looking for restyling in Edit mode, you won't find it. You attach the clip as a reference in Create and describe the treatment you want.

That's a genuine second route to changing an existing clip, alongside Edit mode. Worth knowing, because Edit mode requires a source clip of three to ten seconds and this doesn't.

Restyling happens in Create, not Edit. If you looked for it in Edit mode, that's why you didn't find it.

The other route to changing a clip

What do they have in common?

More than the differences suggest.

Same resolutions — 720p and 1080p, no 4K on either.

Same five aspect ratios — 16:9, 9:16, 1:1, 4:3 and 3:4. Mid-table on the roster, and more than the Veo family's two.

Same audio behaviour — always on, with no toggle on either. Which matters if you've cast a voice separately, since neither model lets you decline the generated sound.

And both cap at fifteen seconds.

What do they cost?

Neither model appears on Hexcoded's published render costs panel.

That panel covers eight video models and neither Wan is among them. So there's no published figure to compare, and we're not going to derive one.

The authoritative figure is the cost shown before you render, which appears per model. For a comparison between these two specifically, that's the only source.

So when do you use which?

Three cases, and the input decides it.

1

Building a shot from a description or stills

Wan 2.7. Eight references, images accepted, start and end frames, and a two-second minimum for short inserts. Wan 2.6 can't do this — it won't take an image.

2

Changing the look of footage you already have

Wan 2.6, in Create mode, with the clip attached as a reference. This is a second route alongside Edit mode, and it has no three-to-ten-second source restriction.

3

Anything needing a voice you cast separately

Neither. Both have audio always on with no toggle. Use a model with a toggle — the Seedance family, Vidu Q3-Turbo or Kling 3.0 Omni.

Wan 2.7 is the only model on the roster with a two-second minimum. If you need a very short insert — a single gesture, a reaction, a beat — that's a real advantage, and nothing else goes below three seconds except Grok Imagine 1.5 and Vidu Q3-Turbo at one.

All settings read from Hexcoded's Creative Studio picker in September 2026, both models observed individually. Neither model appears on the published render costs panel. Settings change with model versions. The cost shown before you render is the authoritative figure.

The bottom line
  • These aren't two versions of one model. Wan 2.7 composes; Wan 2.6 restyles
  • Wan 2.6 won't accept an image reference at all. Its input is footage
  • Wan 2.7 takes eight references; Wan 2.6 takes three
  • Wan 2.7 starts at two seconds — the shortest minimum on the roster. Wan 2.6 starts at five
  • Wan 2.7 supports start and end frames. Wan 2.6 supports start only
  • Restyling happens in Create mode, not Edit. That's why you couldn't find it in Edit
  • Both share 720p and 1080p, five aspect ratios, and audio always on with no toggle
  • Neither appears on the published cost panel. Check the figure shown before you render

No. They do different jobs. Wan 2.7 composes video from a description plus references including images. Wan 2.6 restyles footage you already have and won't accept an image reference at all. The adjacent version numbers are misleading.

Because its input is footage. Its reference field reads "add clips or audio," and its description is "restyle a clip you already have." A restyling model works from existing video rather than from stills.

In Create mode, not Edit. Wan 2.6 is in the Video Create list, and you attach the clip you want restyled as a reference. That's a second route to changing an existing clip alongside Edit mode — and unlike Edit, there's no three-to-ten-second source restriction.

Wan 2.7 takes eight, accepting images, clips or audio. Wan 2.6 takes three, accepting clips or audio only. Eight is enough for a character, a location, a style and a prop; three is enough for a source clip and a look.

No. Both have audio always on with no toggle. If you've cast a voice separately, use a model with a toggle — the Seedance family, Vidu Q3-Turbo or Kling 3.0 Omni are the six on the roster that have one.

Wan 2.7 starts at two seconds, which is the shortest minimum of any model on the roster except Grok Imagine 1.5 and Vidu Q3-Turbo at one. Wan 2.6 starts at five.

Two models, two jobs

What each model accepts shows in the picker before you attach anything — reference count, accepted types, duration range and audio behaviour, per model. Eighteen video models on one credit balance.

Open Creative Studio

More on model capability, access and rights in Models.