Costume consistency

AI Costume & Outfit Consistency

Updated November 3, 2026

Quick answer: clothing is established once on a character's reference sheet and inherited by every shot that refers to them — shots never re-list the outfit, because doing so competes with the face and pulls the camera back until the head is a dot. A real change of garment becomes its own reference image; weather, dirt and a loosened collar are written as text, because reference-conditioned models copy what is in the picture and accept effects layered over it.
Still from “Crimson Vow”, an AI short drama generated end-to-end by SceneMixer (4 characters, 1 episode)
SampleCrimson VowDigital HD1 episodeProduced end-to-end by SceneMixer from the scriptWatch the episodes
SceneMixer at a glanceWhat you get
Video priceFrom $0.06 a second (10 credits) on the default tier, up to $0.47 a second on the top tier; every price is shown before you press the button
Clip lengthUp to 30 seconds per generated segment; an episode is composited from its segments at 1 credit a second
One episode, worked outA 60-second episode on the default tier is about 660 credits (video plus compositing), about $4.11 at the lowest credit price; a 10-episode series of 90-second episodes from a 30,000-character novel is about 10,026 credits, about $62
DialogueSpoken in the video itself, one voice per character, in any of 15 languages
Formats9:16 vertical or 16:9 landscape; 480P to 1080P, plus a 2K tier
Free to start50 credits on signup, 15 free setup images, a free storyboard for the first episode and a free 5-second video preview
Who uses it2,900+ creators have generated with SceneMixer (accounts with at least one generation, counted November 2026)
InputA script, a novel or a one-line idea; .txt, .docx and .md uploads

Wardrobe is established once, then referred to

The instinct is to describe the clothes in every shot. It is the wrong move, and expensively so: a paragraph of wardrobe in a shot prompt competes with the face for the model's attention, and in a tall frame it pulls the camera back far enough to shrink the head until identity is gone. A pale overcoat filling seventy percent of the frame with a tiny face on top is a real failure mode, not a hypothetical.

So SceneMixer keeps clothing on the character's reference sheet, and shots refer to the person rather than re-describing the outfit. What the sheet establishes, every shot inherits.

At a glance

Where wardrobe is decided
The reference sheetOne outfit, described in cause terms — cut, fabric, construction, closure — and rendered as a casting photo on white.
The costume axisA project-level register (period, tailoring tradition, fabric and craft) injected once during analysis, so the whole cast is dressed in one world.
A second outfitBecomes its own reference image — a state of the character, not a sentence in a shot.
Small changesRain, dust, a loosened collar, something held: written as text in the shot. No second image needed.
Shot promptsRefer to the character. They do not re-list the clothing, and they do not pull back to display it.

The line between a state and a sentence

This is the distinction that decides whether you need a second image, and it comes from how reference-conditioned video models behave: they copy what is in the picture — face, garments, accessories, structure — almost verbatim, and they accept as text anything that sits on top of the picture as a global effect — weather, time of day, light, dirt.

  • Make a reference image when the garment changes: a complete change of outfit, a transformation, armour on instead of robes, a uniform instead of civilian clothes, burned or torn beyond recognition.
  • Write it as text when the garment is the same and something is happening to it: soaked, snow-covered, bloodstained, sleeves rolled, coat carried instead of worn.
  • Two forms that can appear in the same frame are not states — they are two characters, or a character and a prop.

Getting this line wrong in the direction of too many images is the common error: early on, most auto-detected “states” were things like holding a cup or a lamp being lit. Those are text. SceneMixer now caps how many states it will propose per character precisely because over-generating them costs money and buys nothing.

Two outfits written into one description

Scripts routinely write a character's wardrobe as an arc — early episodes in one thing, later in another, plus a wedding look. Handed all three as a single description, the render draws a figure wearing half of each. So the description is split structurally before any image is generated: the first outfit stays on the base sheet, and the rest become separate state images. Part labels (footwear, accessories, fabric) are recognised as parts of one outfit rather than as a second one.

Which image a given shot uses

Per shot, in priority: a state chosen for that specific segment, then a per-episode override image, then a state assigned for that episode, then the base sheet. Only states that have actually been rendered count — if a state exists on paper but has no image, the shot falls back to the base sheet and carries the difference as text instead, so the change is still described even when there is no picture of it.

SceneMixer assigns states automatically: after a shot list is generated, a pass decides which segments use which state and writes it down. It is done per segment rather than per episode on purpose — a flashback outfit applied to a whole episode would put the character in the wrong clothes for every shot in it.

What it will not do

  • Dress a character from a photo of a real garment. You can upload a reference image for a character, but there is no garment-transfer feature.
  • Change an outfit in a rendered shot. Wardrobe is fixed at generation. Changing it means rendering again.
  • Keep a state you never rendered. States without images fall back to the base sheet plus text.
  • Preserve manual per-segment state picks through a shot-list regeneration. Those are recalculated; episode-level choices survive.
  • Hold up under a long coat in a vertical full-length frame. Known limit: the garment wins and the face shrinks. Use closer framing.

What it costs

Each state image on SceneMixer is a generated image: 7 credits. Your first project includes 15 free reference images across characters, locations and props, and state images have their own separate free allowance in that first project — so trying one costs nothing. Text-only changes cost nothing at all, which is the main reason the state-versus-sentence line is worth learning. See character sheets for how the base image is built, and the pricing page for rates.

Languages

Native dialogue in 15 languages

Every series below was produced by the same SceneMixer pipeline, with the cast speaking that language natively: the interface, the working documents and the spoken lines are all in one language, and nothing is dubbed. Open one to hear it.

See all 15 languages

In SceneMixer vs. stitching tools together

Most AI drama workflows are five or six tools in a row. This is where each step happens here, and what it replaces.

StepIn SceneMixerDoing it by hand with separate tools
Script and episode planPaste a script or a novel, or type one line: it is read into a cast, locations, props and an episode outline you can editA chat model writes the script; you split it into episodes and carry the text from tool to tool
Character faces and placesEvery character, location and prop is drawn once as a reference sheet and attached to every shot it appears inAn image model, one sheet at a time; you upload the right reference to every clip yourself
ShotsA storyboard per episode with numbered shots and seconds; video is generated segment by segment in 9:16 or 16:9 against the same sheetsA video model, one clip per prompt, describing the same people again each time
Voices and dialogueLines are performed in the video itself, one voice per character, in any of 15 languagesA text-to-speech tool for each line, then a lip-sync pass on every clip
Fixing one shotEdit that shot's text and regenerate the segment alone; the rest of the episode is untouchedRegenerate the clip, then re-cut the timeline
FinishingSegments are composited into one episode file with its score; members download without the AI badgeA video editor to join the clips, add music and captions, then export
PublishingYou upload the finished episode to TikTok, YouTube, Instagram or anywhere else; nothing is posted for youUpload by hand, the same as here

Frequently asked questions

Why is the outfit not described in every shot?

Because re-describing it competes with the face for attention and pushes the camera back to display the clothes, shrinking the head until the character is unrecognisable. SceneMixer establishes wardrobe once on the reference sheet, and every shot that refers to that character inherits it.

When do I need a second image for a costume?

When the garment itself changes — a full change of outfit, armour, a uniform, something burned or torn beyond recognition. If the same garment is just wet, dusty, bloodied or unbuttoned, write it in the shot text; reference-conditioned models accept those as effects layered over the picture.

My script describes a character's wardrobe changing over the season. What happens?

SceneMixer splits the description before anything is drawn: the first outfit stays on the base sheet and the later ones become their own state images. Handed all of them at once, a single render draws a figure wearing parts of each.

Which image gets used in a given shot?

On SceneMixer, a state chosen for that segment wins, then a per-episode override, then a state assigned to the episode, then the base sheet. States that have no rendered image do not count — the shot uses the base sheet and describes the difference in words.

Can I change the outfit in a shot I already rendered?

No. SceneMixer fixes wardrobe when the shot is generated, so a change means generating that segment again at the usual per-second price.

Dress the cast once

Fifteen free reference images is a whole small cast.

Try SceneMixer free

Credit packages from $1.49 · Pro $7.99/mo · No credit card to start