ForensicFaceGen: AI-Powered Suspect Sketch Generation Using Stable Diffusion for Criminal Investigations
Bibliographic record
Abstract
This paper demonstrates the entire process to construct forensic face composites from witness statements through advanced AI and image processing techniques. The system begins with text input from witnesses. This input is subjected to natural language processing (NLP) for extracting primary features. It achieves this by disassembling words to determine their base forms and comparing similar words. The system then converts the processed data into an organized prompt. It improves this question more with a large language model(LLM) to ensure that it's precise and comprehensive.The improved prompt guides an AI image generation tool (like Stable Diffusion or GANs) to make an initial face composite. This picture gets better through techniques to boost resolution and cut down noise making it clearer. A user screen lets people give feedback and make changes over and over so the result matches what the witness said. Law enforcement checks the final face before it goes into a safe database (PostgreSQL or Firebase) and is put on cloud platforms (AWS or Google Cloud) for approved users to access.This method speeds up the old forensic sketch process. It uses AI to be more accurate work faster, and grow , while still letting humans step in to make it better. The system is built in parts, so it can change as NLP and AI that makes new things get better in the future.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.004 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.014 | 0.004 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".