Generate a professional headshot prompt first, then stay on this page to verify evidence, boundaries, and risk before you publish.
Set core variables and generate a prompt pack in under one minute.
Drag and drop or click to browse
Supported formats: JPG, PNG, WEBP (Max 10MB)
Quick presets
4Deterministic mode enabled: core prompt builder works without external model calls.
Review your prompt, negative prompt, quality band, and next action.
Run the builder to generate your prompt pack
You will see model variants, realism checks, and go/no-go actions here.
Core conclusions, measurable indicators, and applicable boundaries.
Applicability balance
Based on defined audience fit rows including explicit non-fit boundaries.
Prompt pack snapshot
Core variables: Corporate trust / Neutral studio / Business formal
Target: LinkedIn profile, tone: Confident and calm
Readiness: 83, guardrail coverage: 40%
5 core variables
Style, background, attire, platform, and tone explain most quality variance in first-pass outputs.
Boundary: If any variable is implicit, artifact risk increases and extra iterations are needed.
400 x 400 + 8MB
LinkedIn Help states profile photos should be within 400 x 400 to 7680 x 4320 and under 8MB.
Boundary: Prompt alone cannot guarantee final file size; export settings still need manual checks.
n = 610
A 2017 open-access study found other-selected profile images produced more favorable impressions than self-selected ones across two internet studies (n=610).
Boundary: This finding guides selection workflow, not model architecture.
No AI filters
U.S. passport guidance explicitly disallows changing photos with software, filters, or AI.
Boundary: This page targets business identity photos only, not government ID workflows.
Model rewrite + lifecycle
OpenAI documentation notes DALL·E 3 prompts may be automatically rewritten and model support windows can change, so teams should keep variant prompts plus version logs.
Boundary: Evidence is from OpenAI model behavior; non-OpenAI platforms can differ and still require replay testing.
| Persona / scenario | Applicable | Not applicable | Suggested action |
|---|---|---|---|
| Job seeker updating LinkedIn | Needs fast prompt baseline and profile-safe framing. | If user expects passport-compliant output from the same prompt. | Generate prompt pack, run 3 iterations, then external review. |
| Medical expert team page | Needs clean attire cues and conservative trust signal. | If visual branding requires artistic or cinematic rendering. | Use medical preset and enforce natural skin texture checks. |
| Founder media kit owner | Needs both credibility and narrative personality. | If all outputs must be uniform across a large team directory. | Use creative preset but review policy-safe boundaries before publish. |
| HR team managing 50+ profiles | Needs repeatable prompt templates and copy packs. | If each profile requires fully custom storytelling prompts. | Lock one template per role family and standardize negative prompt rules. |
| Government ID applicant | Not applicable for regulated ID photo workflows. | Passport, visa, and official identity documentation scenarios. | Use official photo requirements instead of AI prompt generation. |
How inputs map to prompt quality, guardrail checks, and interpretation.
Tool reasoning flow
Flow goes from normalized inputs to decision output with explicit step inputs/outputs.
Scoring rubric weights
Facial structure and micro-expression remain close to the subject.
Framing and tone match where the photo will be used.
Lighting is plausible, skin texture stays natural, no waxy effect.
Attire and background align with role and industry expectations.
Negative prompt prevents extra fingers, warped eyes, and fake logos.
Map user intent to style, platform, and tone to avoid ambiguous prompts.
Input: Style + platform + tone selections
Output: Role-aware prompt skeleton
Append negative prompt rules to suppress common realism failures.
Input: Prompt skeleton + negative rule library
Output: Main prompt + negative prompt pair
Calculate deterministic readiness score with rubric weights and boundary penalties.
Input: Prompt pair + rubric weights + toggles
Output: Readiness band + remediation hints
Show applicable boundaries, evidence confidence, and next action CTA.
Input: Score + evidence map + risk table
Output: Go/no-go decision with action path
Source-backed facts, observed timestamps, and unresolved evidence gaps.
Stage1b gap audit and remediation status
Closed items include traceable sources; rows marked pending/no reliable public data remain hypotheses instead of hard conclusions.
| Gap | Impact | Action | Status |
|---|---|---|---|
| Model-variant recommendation lacked first-party lifecycle evidence. | Could mislead teams into assuming one prompt stays stable across model upgrades. | Added OpenAI Image Generation docs covering DALL·E 3 prompt rewriting, moderation controls, and the 2026-05-12 support sunset for DALL·E 2/3. | Closed in stage1b |
| Cross-platform authenticity boundaries were under-specified. | Teams might reuse heavily edited headshots where platform rules require realistic photos. | Added LinkedIn likeness policy and Google Business “represent reality / no excessive alteration” guidance with explicit dates. | Closed in stage1b |
| Selfie distance distortion risk had no quantitative evidence. | Close-range references can produce unprofessional facial proportions even when prompts look correct. | Added JAMA 2018 camera-distance evidence (12-inch vs 5-foot distortion) as a hard boundary for reference inputs. | Closed in stage1b |
| Automated quality scoring lacked demographic-risk caveats. | Without quality controls, teams may overtrust low-quality captures in sensitive review flows. | Added NIST FRTE notes on image-quality dependence and camera pitch misadjustment risk. | Closed in stage1b |
| No reliable public causal dataset maps headshot prompt score to hiring conversion uplift. | Any direct “score => conversion%” claim would be overconfident. | Marked as pending; keep as experiment hypothesis and require controlled A/B tracking before external claims. | Pending: no reliable public data |
Evidence coverage
9 validated facts and 4 open evidence gaps.
Research updated: 2026-02-18
Official platform documentation · Accessed 2026-02-18 · page label: Last updated 1 year ago
LinkedIn profile photos must be 400x400 to 7680x4320, under 8MB, JPG/PNG.
Open sourceOfficial platform policy · Accessed 2026-02-18 · page label: Last updated 1 year ago
Profile photos must reflect your likeness; non-compliant profile images may be removed.
Open sourceOfficial platform policy · Last updated 2025-07-16 · accessed 2026-02-18
Photos should be at least 720x720, between 10KB and 5MB, and should represent reality with no excessive alterations.
Open sourceOfficial model documentation · Accessed 2026-02-18
The guide notes DALL·E 3 can automatically rewrite prompts and supports moderation controls (auto/low).
Open sourceOfficial model lifecycle notice · Accessed 2026-02-18 · deprecation date listed as 2026-05-12
OpenAI states support for DALL·E 2 and DALL·E 3 in the API will stop on 2026-05-12.
Open sourceOfficial government guideline · Last updated 2025-12-15 · accessed 2026-02-18
Passport guidance prohibits changing photos with software, apps, filters, or artificial intelligence.
Open sourcePeer-reviewed study · Published 2018-03-01 · accessed 2026-02-18
At 12-inch distance, perceived nasal size increases ~30% (male) / ~29% (female) compared with 5-foot portrait distance.
Open sourcePeer-reviewed study · Published 2017-04-14 · corrected 2021-08-13 · accessed 2026-02-18
Across two internet studies (n=610), other-selected profile images produced more favorable impressions than self-selected images.
Open sourceGovernment benchmark · Table last updated 2025-03-05 · accessed 2026-02-18
NIST notes false negatives depend strongly on image quality and camera misadjustment can induce pitch-angle variation.
Open sourceSERP competitor content · Accessed 2026-02-18
Contains static prompt lists and tips, useful as baseline comparison but not policy-grade evidence.
Open source| Claim | Data point | Boundary | Source |
|---|---|---|---|
| LinkedIn profile photos have explicit size and format constraints. | LinkedIn Help states 400x400 to 7680x4320, under 8MB, JPG/PNG. | Prompt quality cannot replace final export validation for dimensions and file size. | LinkedIn Help (a549049) Accessed 2026-02-18 · page label: Last updated 1 year ago |
| LinkedIn requires profile photos to reflect your likeness. | The policy allows removal of profile images that violate photo conditions. | This is a moderation rule, not a guarantee that any realistic image is brand-appropriate. | LinkedIn Help (a1377087) Accessed 2026-02-18 · page label: Last updated 1 year ago |
| Google Business photo policy requires authenticity and basic technical thresholds. | Guidance includes minimum 720x720 resolution, 10KB-5MB size, and no excessive alterations. | This applies to Google Business Profile; teams must still verify each destination platform. | Google Business Profile Help Last updated 2025-07-16 · accessed 2026-02-18 |
| Prompt text can be transformed by the model stack before image generation. | OpenAI docs note DALL·E 3 “automatically rewrites your prompt”, and moderation controls can be set to auto or low. | This behavior is OpenAI-specific evidence and should not be over-generalized to all providers. | OpenAI image generation guide Accessed 2026-02-18 |
| Model lifecycle changes can invalidate stable prompt assumptions. | OpenAI changelog states API support for DALL·E 2 and DALL·E 3 stops on 2026-05-12. | Lifecycle notice is vendor-specific; keep version tags and replay tests across all providers. | OpenAI Image API changelog Accessed 2026-02-18 · deprecation date listed as 2026-05-12 |
| Regulated ID workflows disallow AI-edited images. | U.S. passport guidance explicitly prohibits software, apps, filters, and AI alterations. | This applies to U.S. passport workflows and should be treated as a hard exclusion boundary. | U.S. Department of State Last updated 2025-12-15 · accessed 2026-02-18 |
| Short camera distance can materially distort perceived facial proportions. | JAMA model reports ~30% (male) / ~29% (female) perceived nasal enlargement at 12 inches versus 5 feet. | This is geometric distortion evidence, not a direct hiring or conversion outcome metric. | JAMA Facial Plastic Surgery (PMC) Published 2018-03-01 · accessed 2026-02-18 |
| Self-selection bias affects final profile-image choice quality. | Two internet studies (n=610) reported more favorable impressions for other-selected profile photos. | This supports review workflow design, not model architecture ranking. | Cognitive Research, 2017 Published 2017-04-14 · corrected 2021-08-13 · accessed 2026-02-18 |
| Image-quality and camera-angle issues can increase automated decision errors. | NIST FRTE notes false negatives are strongly tied to image quality and camera misadjustment can induce pitch-angle variation. | Evidence comes from biometric evaluation; use as risk guardrail, not a direct business KPI predictor. | NIST FRTE demographics Table last updated 2025-03-05 · accessed 2026-02-18 |
| Evidence gap | Status | Impact | Minimum path |
|---|---|---|---|
| Prompt performance benchmark under identical prompts across model vendors | Pending: no reliable public apples-to-apples benchmark | Variant ordering can change after model updates and safety policy shifts. | Replay top variants monthly, log model/version/policy settings, and archive failures. |
| Causal relationship between prompt score and hiring/reply conversion | Pending: no reliable public causal dataset | Direct percentage-lift claims would likely overstate certainty. | Treat score as internal quality proxy and run controlled A/B tests before external claims. |
| Cross-region disclosure requirements for AI-generated profile images | Open: no unified global disclosure standard found in public sources | Teams may under-disclose synthetic edits in regulated or high-trust sectors. | Add legal checkpoint per target market before publishing externally. |
| Demographic-specific quality thresholds for headshot scoring | Partial evidence only; no open threshold standard | A single universal threshold may underperform for diverse populations. | Track anonymized review outcomes by cohort and tune thresholds quarterly. |
Compare this hybrid workflow with static prompt lists, chat-only, and studio-only paths.
Decision utility chart
Illustrative scores show why a hybrid tool-plus-report path balances speed and explainability.
| Dimension | This page | Prompt list | Chat only | Studio only |
|---|---|---|---|---|
| Policy boundary clarity | Includes LinkedIn, Google Business, and U.S. passport boundaries with dates. | Mostly style tips with limited policy mapping. | Policy coverage depends on user prompts and model answers. | Strong capture guidance, but digital-disclosure boundaries may be absent. |
| Model portability risk control | Variant prompts + version tagging + lifecycle reminders. | One-size-fits-all text with no version trace. | Can adapt quickly, but history and reproducibility are weak. | Low model dependency, high scheduling dependency. |
| Evidence traceability | Source list + observed dates + open evidence-gaps table. | Usually no timestamps or uncertainty markers. | Citation quality varies by session and prompt rigor. | Portfolio-heavy evidence, often without structured citations. |
| Failure-mode handling | Explicit guardrails for over-stylization, crop mismatch, and compliance misuse. | Mostly positive prompts; negative guardrails are sparse. | Depends on operator discipline to remember edge cases. | Physical setup risks are managed, but AI-edit misuse is out of scope. |
| Team governance readiness | Prompt pack, checklist, and risk matrix support repeatable reviews. | Manual copy/edit flow with weak audit trail. | Fast ideation, harder to standardize across reviewers. | Consistent with one vendor, expensive to scale frequent updates. |
| Causal outcome certainty | Explicitly marks conversion-lift claims as pending without reliable public causal data. | Often uses implied outcome claims without method notes. | May generate confident percentages without reproducible backing. | Can show portfolio outcomes, but cross-channel causal attribution remains difficult. |
Need a fast go/no-go decision first?
Generate the prompt pack now. If the score is below 80, review method and risk sections before publishing.
Concrete misuse, compliance, and mismatch risks with executable mitigation steps.
Risk heatmap
Red points indicate high-impact risks that should be mitigated first.
| Risk | Probability | Impact | Mitigation |
|---|---|---|---|
| Over-stylized output can violate professional authenticity expectations. | Medium | High | Cap stylization tokens, enforce realism negatives, and cross-check LinkedIn/Google reality rules before publish. |
| Close-range selfie references can produce distorted facial proportions. | Medium | High | Prefer portrait-distance captures over 5 feet for reference images and reject extreme close-up inputs. |
| Using headshot prompts for passport/visa workflows creates compliance exposure. | Low | High | Hard-block regulated ID use cases and route to official government photo requirements. |
| Owner-only final selection can lock in self-selection bias. | Medium | Medium | Require at least one external reviewer and keep rationale for final picks. |
| Model lifecycle changes can break previously stable prompts. | Medium | Medium | Tag outputs with model/version, monitor changelogs, and replay critical prompt packs monthly. |
| Claiming direct hiring-conversion uplift without causal evidence can mislead stakeholders. | Medium | High | Mark as pending by default and publish outcome claims only after controlled A/B validation. |
Scenario playbooks
Practical examples with premise, process, and measurable outcomes.
Premise: Needed to replace a 3-year-old profile image before fundraising outreach.
Process: Used founder-modern preset, ran 4 iterations, and applied external reviewer selection.
Outcome: Readiness improved from 74 to 88 after tuning tone and background consistency.
Premise: Hospital marketing team needed consistent profile photos for 30 clinicians.
Process: Locked one medical preset, standardized negative prompts, and reused copy pack.
Outcome: Reduced review revisions by using one deterministic rubric across all profiles.
Premise: Founder needed one polished profile for LinkedIn and one expressive variant for media outreach.
Process: Generated one conservative and one expressive variant with separate publish boundaries.
Outcome: Avoided channel mismatch by assigning each variant to a specific platform.
Use these examples to calibrate natural texture and professional framing thresholds.
Example 1: realism reference
A four-step flow from input to publish decision.
Choose style, background, attire, platform, and tone before writing any free-form notes.
Run the builder to get the main prompt, negative prompt, and model variants.
Read readiness score, guardrail coverage, and low-score remediation guidance.
Keep one external reviewer in the final gate and map each variant to its target platform.
Answering operational, strategy, and compliance questions.
Continue to generation, editing, and policy validation workflows.
Apply your generated prompt pack directly in a LinkedIn-focused workflow.
Validate platform policy and team-level governance before publishing.
Check posture, framing, and confidence cues to pair with your prompt.
Tune light direction and contrast so prompt intent survives rendering.
Apply subtle post-processing without breaking likeness fidelity.
Estimate budget impact when choosing between AI and studio workflows.
Keep the prompt pack, run one final quality pass, then move into full image generation.