Wan 2.7 Review: Uncensored Edits, a Softer Text Side

Our test batch ran 68 prompts across both endpoints, judged blind by three independent reviewers who didn't know which model made which image. On careful studio and portrait prompts, Wan 2.7 is genuinely strong: a brutalist tower photograph, a backlit vineyard scene and a still life of four distinct materials all came back close to the brief, and a "GOOD MORNING" mug label and a kraft-paper honey jar both printed their text correctly. One clear editor result was a straight relight — magenta from one side, cyan from the other — that held the subject's face, hair and sweater exactly while completely changing the mood of the shot.

The weak spots cluster around two things: anything that needs to be counted or measured precisely, and busy scenes with more than one person in contact or motion. Reviewers flagged a clock face stuck on ten-ten regardless of the time requested, a wine glass that never quite reached the brim we asked for, and a mirror reflection that didn't track the logic of what should and shouldn't match. None of this is unique to Wan 2.7 among the models in our batch, but it's consistent enough here to plan around: use it for atmosphere, wardrobe, lighting and composition, and double-check anything that has to add up to a specific number.

The gallery holds published, safe-for-work examples from the Wan 2.7 model page:

Three depths, one shaft of light. A post-apocalyptic scrapyard scene stages three figures across three distinct depth planes, with focus held on the middle one — the kind of layered composition the reasoning pass is built for.

Coffee, rain, a look across the table. A first-person café shot with no reference image and no content toggle to set — about as far as a single line of text goes on this model.

Four references, one cover. A singer, a light source, a microphone and stage haze, each named to its own source image, composed into a single album-cover frame.

One dress, two references. A face plate and a garment plate combine so the silk dress falls exactly where the original window light does.

Wan 2.7 rewards a plain shot list: subject, wardrobe, one action, light source, lens, mood — in that order. On the editor, give every reference image a job in the sentence, or say plainly what has to stay the same.

1. A café table, one line of text. The exact prompt behind the café image above; no reference photo, no extra fields.

First-person POV photograph across a small cafe table: the viewer's own hand wrapped around an espresso cup in the near foreground, and opposite, an adult woman curled sideways on the bench seat in an unbuttoned oversized white shirt over a black bralette, bare legs tucked up, meeting the camera's gaze with a slow amused smile, rain on the window, warm afternoon light, 35mm, shallow depth of field.

Run this prompt on Wan 2.7 →

2. Two labelled references, one dress (edit). Upload the woman first, the garment second, then paste this.

Image 1 is the woman — keep her face, hair and pose. Image 2 is the dusty-rose silk slip dress; dress her in it, matching image 1's window-light direction and colour temperature so the silk sheen falls where the light actually is.

Run this edit on Wan 2.7 →

3. Four references, each with a job (edit). Upload the singer, the light reference, the microphone and the haze source in that order.

Cast: one man. Compose one album cover from these four references: the singer from the first under the deep red side light of the second, sweat-soaked shirt stuck to the back, microphone from the third, stage haze from the fourth. Cinematic, shallow depth of field, heavy film grain.

Run this edit on Wan 2.7 →

Weak: "Woman in a slip dress by a window." No reference, no light direction, no limit — Wan 2.7 fills in all three itself. Strong: prompt 2. Every choice named, one reference for the person and one for the garment.

Our content ladder runs from L1 Suggestive (lingerie, a slipping strap) through L2 Partial and L3 Strong to L4 Intense (an intimate scene between two adults) and L5 Extreme, tested here as two separate probes: L5a (consensual adult restraint and impact play) and L5b (cinematic gore). We ran these on both endpoints between 16 and 24 September 2026.

L1 and L2 rendered as written on both endpoints, every time. L3 Strong was mixed and only tested on text-to-image, where one run rendered and one was softened; we haven't run it on the editor yet. At L4 Intense, the editor rendered about a third of its runs and softened the rest; the text endpoint softened every single one. The L5a probe followed the same pattern: the editor rendered it occasionally, text-to-image softened it every time. The L5b probe was the one place both endpoints agreed, rendering it every time on both sides.

The page labels reflect that split: Uncensored on image-to-image, Softened on text-to-image. Neither endpoint refused anything in our tests — the fallback everywhere was a milder version of the scene, not a rejection.

Not tested yet: L3 Strong on the editor; scenes with two women or two men above L1; a full nine-reference edit (we used up to four).

Adults only, in every image. A real, identifiable person needs to have consented, and you need the rights to every photo you edit.

• Sharper detail and up to 4K? Wan 2.7 Pro shares the same reference-editing shape at a higher price; its text side softened intimate scenes even more often than the base model's in our tests.
• Need video instead, fully uncensored? Wan 2.7 Spicy turns a still into a clip on the same family line, at $0.1235 a second for 720p.
• Cheaper uncensored stills and edits? Qwen Image 2.1 passed our L4 Intense prompt on both text and editing from $0.024 at 1K.
• A finished, photographic look? Seedream 5.0 Pro leans further into skin and light realism, at $0.036 per image. Our Seedream 5.0 Pro review.