Choose the reference that actually shows the identity details your result needs. Brightness, file size and apparent sharpness are not enough. Blur can obscure facial edges; clipped highlights can erase features completely. If neither photograph contains dependable evidence, a new authorized photograph is usually a better starting point than asking generation to invent the missing face.
Inspect information rather than attractiveness
Compare eye shape, nose contour, jawline and hairline at a useful viewing size. Check whether a bright region retains texture or is simply featureless. An enlarged blurry image may reveal no additional detail. Sharpening can emphasize edges without recovering trustworthy identity information.
Do not treat a generated enhancement as proof of what was hidden. It can create a plausible eye or skin detail that was not observable in the source. Keep the original available and be honest about which features you can verify.
A fictional headshot choice
An invented applicant has two photographs. The indoor one is slightly soft but shows the full face evenly. The outdoor one is crisp around the jacket but the sun has washed out one eye and part of the cheek. For a frontal professional portrait, the indoor image may provide more useful identity evidence despite looking less polished.
If both eyes are unreadable in the indoor source as well, neither is a good identity reference. A new evenly lit photograph can solve the input problem. Repeatedly generating prettier faces from inadequate sources risks producing a convincing stranger rather than a useful headshot.
Make a reference admission checklist
Check visibility of the required features, expression, angle, obstruction and relevant costume details. Record a reason for accepting or rejecting each candidate. Keep the task specific: a full-body pose study may need different evidence from a close professional portrait.
Do not assume more references are always better. Where a workflow accepts several images, contradictory age, hairstyle or angle cues may complicate the request. Use the actual permitted inputs and choose a clear main reference rather than promising an unsupported multi-image identity lock.
Review the result against the surviving evidence
In the Image Skill Market, select an appropriate available portrait method and inspect the returned image beside the source. Judge identity before flattering light or background. Use controlled photo editing for modest corrections where appropriate, but do not claim it can recover clipped pixels or missing anatomical detail faithfully.
For important professional use, ask the depicted person to approve the result. A third party's impression that the picture looks realistic is not the same as the person's recognition. If the source is inadequate, stop and improve the input instead of accumulating more speculative variations.
Does a larger file always provide a better reference?
No. A large image can still be out of focus or clipped. Evaluate the visible evidence relevant to the intended crop.
Can I brighten a dark reference before uploading?
A controlled adjustment may help reveal existing detail. Inspect noise and color afterward, and do not treat newly invented enhancement detail as recovered fact.
Sources and related reading
See the input examples in Creating with Gen-4 Image References for another provider’s reference guidance. Do not transfer its input limits or controls to a different platform.