Create models / Consistency

Same person. New photograph. No identity séance.

Consistency is not repeating the same seed and hoping. It is an approved identity reference, a small set of truths you refuse to renegotiate, and a test that reaches beyond the easy beauty crop.

Make it, step by step Copy the prompts

Fashion model in a movement identity consistency test — Motion is where identity systems earn trust. Test it before the campaign depends on it.

The point of view

Lock fewer things, more clearly.

Give the person one explicit FaceID/reference. Repeat age, hair and body proportions only when the shot needs reinforcement; do not describe a different face around the identity token.

Treat hairstyle, makeup and body shape as art-direction decisions with their own approved master. If they change, call it a new look on purpose—not a side effect of a longer prompt.

Make it

From blank page to a production-ready first set.

Work small until the cast, product and photograph agree. Then make the system repeatable.

01
Approve the casting portrait
Choose a close, neutral image with a clear face and no accessory or hairstyle that will contaminate every future scene.
02
Register one identity
Turn the winner into a FaceID/reference and use the exact token every time. Names alone carry no identity.
03
Create the four-view test
Render close, full-length, seated and walking with the same styling. Compare face geometry, age, hair and proportions.
04
Save the passing setup
Move the reference, prompts, background and settings into a template or pack. Consistency improves when the team repeats the system, not just the prose.

Prompt book

Prompts you can actually start from.

Replace the placeholder IDs with your Astria references. Keep the factual product and identity instructions; change the creative language until it sounds like your brand.

Full-length fashion model identity result — Example result

Four-view proof · 4:5

<faceid:MODEL_ID:1> woman shown as a coherent four-image casting test: close portrait, full-length standing, seated three-quarter view, and walking mid-step. Same exact face, age, pulled-back hair, body proportions and calm expression in every image. Simple black T-shirt and trousers, white studio, soft daylight, natural skin. No face drift, no age change, no body reshaping.

Why this worksOne compact battery tests the distances and poses that usually reveal drift.

Same fashion identity in a new seated scene — Example result

New scene, same person · 3:2

<faceid:MODEL_ID:1> woman leaning against a tiled café counter in early morning light, wearing <faceid:COAT_ID:1> coat. Preserve the exact approved face, age, hairstyle and body proportions. Candid 50mm fashion photograph, soft window light, muted city palette, realistic skin and coat texture.

Why this worksThe model stays factual while the setting is allowed to change.

Close crop preserving the same fashion identity — Example result

New crop, same look · 4:5

Waist-up crop of <faceid:MODEL_ID:1> woman from the approved full-length look, wearing <faceid:BLAZER_ID:1> blazer. Keep the same face, hair, makeup, body proportions, blazer fit, light direction and grey background. Camera moves closer; nothing else changes.

Why this worksIt frames a derivative crop as a camera change, not a new generation brief.

Choose the tool by the photograph

There is no single “best model.” There is a best next move.

FaceID / reference

The identity anchor

This is the durable part of the system. The text cannot substitute for the actual chosen face.

Nano Banana 2

Reference-led scene changes

Test identity across the four-view battery before making it your default for a production.

Seedream 5

Multi-reference continuity

A candidate when model, outfit and location all need references; compare against the same identity test.

Model availability changes. Astria added Recraft V4.1 in May 2026; check the current model catalog before production. See Astria changes

Show me the money

Consistency turns a good image into a castable production asset.

Once a model survives close, full, seated and motion views, new products and crops stop reopening the most subjective decision in the shoot.

Fewer identity corrections
One recognizable cast across channels
Templates that survive team handoffs

Full-length identity benchmark photograph

See the work

Compare the hard views.

The Astria benchmark puts current image models through the same identity sequence, so teams can choose from evidence rather than a leaderboard.

Build it with Astria Skills

The craft can live in your workflow.

The open-source Astria skills turn these methods into guided work with the Astria CLI. Read the skill, adapt the judgment, and keep every generation reviewable.