CartVisualCreate account

How do FASHN Model Create’s reference roles and face_reference_mode divide identity from composition?

Use `image_reference` for full-image composition and pose inspiration, and `face_reference` for a specific identity. With a face reference, `face_reference_mode` changes how that face influences pose and expression: `match_base` favors the prompt and base generation, while `match_reference` more closely follows the face reference’s pose, gaze, and expression. `match_reference` is the documented default. Combined-reference precedence is unspecified.

Published · By CartVisual editorial team

Choose what each image is authorized to guide

An image_reference is an optional composition-and-pose guide; FASHN says the output uses it as inspiration rather than copying every detail. This may convey a full-body arrangement, camera framing, or spatial relationship. A face_reference is an optional portrait intended to guide a recurring identity. Neither field by itself verifies clothing construction, and identity continuity is not proof of an unchanged overall pose.

Decide whether the approved requirement is a recognizable face, a particular full-image arrangement, or both. Write the requirement beside the selected field, not merely beside the filename. If image_reference is present, Model Create inherits its aspect ratio unless you set aspect_ratio explicitly. With no image reference, the documented default is 1:1; a ratio fixes frame proportions, not a guaranteed pixel rectangle.

Set Model Create’s own face mode deliberately

face_reference_mode is a Model Create control, not just a Model Swap setting. match_base prioritizes the prompt and base generation while adapting the reference face to those instructions. match_reference aligns the generated model more closely with the reference face’s pose, gaze, and expression; FASHN documents it as the default. This is not a contract to preserve an entire supplied composition.

This illustrative POST /v1/run body selects an identity-led, prompt-directed presentation. The example.invalid address is deliberately unusable. Keep endpoint-specific fields inside inputs.

{
  "model_name": "model-create",
  "inputs": {
    "prompt": "front-facing model, coat visible, relaxed arms",
    "face_reference": "https://assets.example.invalid/approved-face.png",
    "face_reference_mode": "match_base",
    "aspect_ratio": "4:5"
  }
}

Do not invent joint-reference precedence

The supplied page documents image_reference and face_reference separately but does not say which wins when both are provided, or how face mode resolves disagreement with image-reference composition. Mark a brief requiring both exact identity and a particular body pose as unresolved; obtain vendor clarification or inspect authorized trial outputs before relying on a combination. Do not present that unknown as mutual exclusivity either.

Using face_reference also imposes a documented 2K output ceiling, adds three credits per image, and typically adds around 20 seconds. For full resolution and ratio diagnostics, consult the existing Model Create face-resolution cap and aspect-ratio inheritance guides rather than treating those conditions as new findings here. FASHN labels this endpoint Experimental.

Illustrative contrast: identity-led versus composition-led coat image

A seller has an approved face portrait with a three-quarter head angle and a separate full-body front-view reference showing the coat unobstructed. For an identity-led asset whose front-facing instruction matters more than copying the portrait’s angle, the seller plans face_reference with match_base and a 4:5 destination. For a composition-led draft without mandatory identity, the seller instead uses image_reference and explicitly selects 4:5 to avoid inherited framing.

If the final brief requires both that exact identity and the supplied full-body arrangement, the docs do not establish combined-input precedence. Next action: record which requirement is mandatory, choose the relevant single-reference configuration, or hold the combined-input request for clarification and review.

Where CartVisual fits

A separate CartVisual model-reference planning route; it does not implement FASHN Model Create, identity locking, or the described reference controls.

See the current CartVisual workflow and its limitations

Source and evidence scope

Prepared by the CartVisual editorial team with AI assistance. Competitor behavior is documented rather than independently executed for this guide.

FASHN: Model Create

Official source snapshot checked .

The linked Model Create documentation supports the reference roles, face-mode default, ratio inheritance, 2K cap, three-credit surcharge and typical latency addition. The supplied page does not specify joint-reference precedence. Configurations and scenarios are illustrative; they do not establish successful endpoint execution or CartVisual integration.

Related decisions

Browse ecommerce creative workflows