An AI portrait API should ship with a default production size, clear identity rules, and a path from draft variants to premium finals.
“Fictional editorial portrait with geometric copper hair, amber rim light, dark navy backdrop, fashion magazine art direction, expressive direct gaze”
FLUX Dev · premium · 960 × 1472Generate in portrait orientation instead of cropping after the fact
The 960×1472 preset gives the model a tall canvas from the start. That supports head-and-shoulders framing, editorial full-body compositions, mobile cards, and negative space above or beside the subject. Both dimensions are multiples of 64, so the preset remains compatible with the provider fleet.
Ask for framing, camera distance, gaze direction, background depth, and light. Avoid contradictory instructions such as close-up and full-body in the same prompt. For a set, keep a structured shot list and vary one composition variable at a time.
- Head-and-shoulders editorial
- Three-quarter environmental portrait
- Full-body fashion concept
- Fictional character sheet
- Anonymous silhouette or avatar
Separate fictional portraits from real-person workflows
Text-to-image portraits should default to fictional adults. A real-person workflow introduces consent, identity, publicity-right, impersonation, and harassment risk. The verified 18+ route prohibits explicit depictions of real or identifiable people, even when the prompt claims consent.
Do not add face upload, face swap, or likeness training as a casual feature. Those capabilities require a dedicated consent record, uploader and subject verification, age checks, revocation, retention controls, and abuse review. The verified 18+ service does not accept face inputs or adult image-to-image requests.
| Workflow | Launch status | Required control |
|---|---|---|
| Fictional adult portrait | Allowed after standard moderation | Prompt and output review |
| Consented professional headshot | Future dedicated workflow | Identity and consent evidence |
| Celebrity or public-figure sexualization | Prohibited | No exception |
| Age-ambiguous sexualized portrait | Prohibited | Fail closed |
Use fast drafts to settle framing before premium rendering
Portrait failures are often compositional rather than purely technical: the crop is wrong, hands enter unexpectedly, or the background competes with the face. Resolve those choices on the fast lane. Move to premium only after a reviewer accepts framing and visual direction.
Keep the accepted draft. A premium model is a new interpretation, not a guaranteed upscale. For exact preservation, add a purpose-built upscale or image-to-image workflow with separate safety review.
{
"prompt": "Fictional adult ceramic artist in a sunlit studio, three-quarter editorial portrait, calm expression, textured linen apron",
"quality": "fast",
"width": 960,
"height": 1472
}Review anatomy, age presentation, and context before use
Portrait QA should check age presentation, anatomy, hands, eyes, teeth, cultural details, unintended text or marks, stereotypes, and whether the output could be mistaken for a real person. Uses involving employment, identity documents, political persuasion, or deceptive profiles need stricter or prohibited handling.
Store a provenance record and consider durable content credentials or watermarking where appropriate. Do not advertise synthetic portraits as documentary photographs.
Questions about AI portrait API
Why 960×1472?
It is the operator’s portrait benchmark size and both dimensions are divisible by 64, making it a practical cross-provider preset.
Can I generate a celebrity portrait?
Ordinary non-deceptive artistic requests require legal and policy review. Explicit or non-consensual depictions of identifiable people are prohibited.
Does the API accept a face photo?
Not in the current API. Identity-bearing uploads need a separate verified consent workflow.
Can I use portraits for profile accounts?
Synthetic identity and impersonation risks require clear labeling and platform-specific review. Deceptive use is not supported.