import gradio as gr from openai import OpenAI import openai from pydantic import BaseModel, Field import os import requests from PIL import Image import tempfile import io import markdown import base64 import datetime import json import re from dotenv import load_dotenv load_dotenv() # Clé OpenRouter openrouter_api_key = os.getenv("OPENROUTER_API_KEY") OPENROUTER_TEXT_MODEL = os.getenv("OPENROUTER_TEXT_MODEL", "mistralai/mistral-small-3.1-24b-instruct:free") # Modèles OpenAI OPENAI_TEXT_MODEL = "gpt-4o-mini" OPENAI_IMAGE_MODEL = "dall-e-3" # Modèle Image via OpenRouter OPENROUTER_IMAGE_MODEL = "openai/dall-e-3" # --- Modèles Pydantic --- class BiasInfo(BaseModel): bias_type: str = Field(..., description="Type de biais identifié") explanation: str = Field(..., description="Explication contextuelle") advice: str = Field(..., description="Conseil d'atténuation") class BiasAnalysisResponse(BaseModel): detected_biases: list[BiasInfo] = Field(default_factory=list) overall_comment: str = Field(default="") # --- Fonctions Utilitaires --- posture_mapping = {"": "","Debout": "standing up","Assis": "sitting","Allongé": "lying down","Accroupi": "crouching","En mouvement": "moving","Reposé": "resting"} facial_expression_mapping = {"": "","Souriant": "smiling","Sérieux": "serious","Triste": "sad","En colère": "angry","Surpris": "surprised","Pensif": "thoughtful"} skin_color_mapping = {"": "","Clair": "light","Moyen": "medium","Foncé": "dark","Très foncé": "very dark"} eye_color_mapping = {"": "","Bleu": "blue","Vert": "green","Marron": "brown","Gris": "gray"} hair_style_mapping = {"": "","Court": "short","Long": "long","Bouclé": "curly","Rasé": "shaved","Chauve": "bald","Tresses": "braided","Queue de cheval": "ponytail","Coiffure afro": "afro","Dégradé": "fade"} hair_color_mapping = {"": "","Blond": "blonde","Brun": "brown","Noir": "black","Roux": "red","Gris": "gray","Blanc": "white"} clothing_style_mapping = {"": "","Décontracté": "casual","Professionnel": "professional","Sportif": "sporty"} accessories_mapping = {"": "","Lunettes": "glasses","Montre": "watch","Chapeau": "hat"} gender_mapping = {"Homme": "man", "Femme": "woman", "Non-binaire": "non-binary person"} MAX_LOG_LINES = 150 def update_log(event_description, session_log_state): timestamp = datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S") new_log_entry = f"[{timestamp}] {event_description}" current_log = session_log_state if session_log_state else "" log_lines = current_log.splitlines() if len(log_lines) >= MAX_LOG_LINES: current_log = "\n".join(log_lines[-(MAX_LOG_LINES-1):]) updated_log = current_log + "\n" + new_log_entry if current_log else new_log_entry return updated_log.strip() def clean_json_response(raw_response): match = re.search(r"```json\s*({.*?})\s*```", raw_response, re.DOTALL | re.IGNORECASE) if match: return match.group(1) start = raw_response.find('{'); end = raw_response.rfind('}') if start != -1 and end != -1 and end > start: potential_json = raw_response[start:end+1] try: json.loads(potential_json); return potential_json except json.JSONDecodeError: cleaned = re.sub(r",\s*([}\]])", r"\1", potential_json) try: json.loads(cleaned); return cleaned except json.JSONDecodeError: pass return raw_response.strip() # --- Holder Client API --- active_api_client_holder = {"client": None, "openai_key": None} # --- Fonctions Principales --- def get_active_client(app_config): """Récupère le client stocké globalement.""" api_source = app_config.get("api_source") if not api_source: return None, "Source API non configurée." client = active_api_client_holder.get("client") if not client: print("WARN: Client actif non trouvé, tentative de ré-initialisation.") if api_source == "openai" and active_api_client_holder.get("openai_key"): try: client = OpenAI(api_key=active_api_client_holder["openai_key"]) active_api_client_holder["client"] = client; print("Client OpenAI ré-initialisé.") except Exception as e: return None, f"Échec ré-init OpenAI: {e}" elif api_source == "openrouter" and openrouter_api_key: try: client = OpenAI(base_url="https://openrouter.ai/api/v1", api_key=openrouter_api_key) active_api_client_holder["client"] = client; print("Client OpenRouter ré-initialisé.") except Exception as e: return None, f"Échec ré-init OpenRouter: {e}" else: return None, f"Impossible ré-init client pour '{api_source}'. Clé/config manquante." if not client: return None, f"Client API pour '{api_source}' non disponible." return client, None def analyze_biases(app_config, objective_text, session_log_state): """Analyse les biais dans l'objectif marketing en forçant un format JSON.""" log = session_log_state log = update_log(f"Analyse biais objectif (début): '{objective_text[:50]}...'", log) if not objective_text: return BiasAnalysisResponse(overall_comment="Veuillez fournir un objectif marketing.").model_dump(), update_log("Analyse biais: Objectif vide.", log) active_client, error_msg = get_active_client(app_config) if error_msg: log = update_log(f"ERREUR Analyse biais: {error_msg}", log) return BiasAnalysisResponse(overall_comment=f"Erreur: {error_msg}").model_dump(), log model_name = app_config.get("text_model") api_source = app_config.get("api_source") # --- Génération du Schéma JSON --- bias_schema = None try: bias_schema = BiasAnalysisResponse.model_json_schema() except Exception as schema_e: log = update_log(f"ERREUR Génération schéma Pydantic: {schema_e}", log) return BiasAnalysisResponse(overall_comment=f"Erreur interne génération schéma: {schema_e}").model_dump(), log # --- System Prompt --- system_prompt = f""" Tu es un expert en marketing éthique et en psychologie cognitive, spécialisé dans la création de personas. Analyse l'objectif marketing suivant : "{objective_text}" Identifie les BIAIS COGNITIFS POTENTIELS ou RISQUES DE STÉRÉOTYPES pertinents pour la création de personas. Concentre-toi sur : 1. **Stéréotypes / Généralisations Hâtives :** Suppose-t-on des traits basés sur le genre, l'âge, l'ethnie, le statut socio-économique sans justification ? (Ex: 'tous les jeunes urbains sont écolos') 2. **Biais de Confirmation / Affinité :** L'objectif semble-t-il chercher à valider une idée préconçue ou refléter trop les opinions du concepteur ? (Ex: 'prouver que notre produit est parfait pour CE type de personne') 3. **Simplification Excessive / Manque de Nuance :** Le groupe cible est-il décrit de manière trop monolithique, ignorant la diversité interne ? (Ex: 'les seniors actifs' sans différencier leurs motivations ou capacités) 4. **Autres biais pertinents** (Ex: Oubli de fréquence de base, Biais de normalité si applicable). Pour chaque biais potentiel identifié : - Nomme le type de biais (ex: Stéréotype d'âge). - Explique brièvement POURQUOI c'est un risque DANS CE CONTEXTE de création de persona. - Propose un CONSEIL PRÉCIS pour nuancer l'objectif ou être vigilant lors de la création. Structure TOUTE ta réponse EXCLUSIVEMENT en utilisant le format JSON suivant (basé sur la classe Pydantic BiasAnalysisResponse): {{ "detected_biases": [ {{ "bias_type": "Type de biais identifié", "explanation": "Explication contextuelle du risque.", "advice": "Conseil spécifique d'atténuation." }} // ... autres biais détectés ... ], "overall_comment": "Bref commentaire général. Indique si aucun biais majeur n'est détecté." }} Réponds en français. S'il n'y a pas de biais clair, retourne une liste 'detected_biases' vide et indique-le dans 'overall_comment'. Assure-toi que la sortie est un objet JSON unique et valide correspondant exactement à cette structure. Ne retourne AUCUN texte avant ou après le JSON. """ response_content_str = "" try: # --- Choix dynamique du response_format --- response_format_config = None if api_source == "openai": response_format_config = {"type": "json_object"} log = update_log(f"INFO: Utilisation response_format=json_object pour OpenAI ({model_name})", log) elif api_source == "openrouter" and bias_schema: response_format_config = { "type": "json_schema", "json_schema": { "name": "bias_analysis", "strict": True, "description": "Analyse des biais potentiels dans un objectif marketing.", "schema": bias_schema } } log = update_log(f"INFO: Utilisation response_format=json_schema pour OpenRouter ({model_name})", log) else: log = update_log(f"WARN: Aucun response_format spécifique appliqué pour {api_source}", log) # --- Appel API --- completion = active_client.chat.completions.create( model=model_name, messages=[{"role": "user", "content": system_prompt}], temperature=0, max_tokens=2400, response_format=response_format_config, ) raw_response_content = completion.choices[0].message.content # --- Parsing de la réponse --- try: parsed_response = BiasAnalysisResponse.model_validate_json(raw_response_content) log = update_log(f"Analyse biais objectif (fin): Biais trouvés - {len(parsed_response.detected_biases)}", log) return parsed_response.model_dump(), log except (json.JSONDecodeError, TypeError, ValueError) as direct_parse_error: format_type = response_format_config.get('type', 'inconnu') if response_format_config else 'inconnu' log = update_log(f"ERREUR Parsing direct réponse JSON (mode {format_type}): {direct_parse_error}. Contenu brut: {raw_response_content!r}", log) cleaned_response_str = clean_json_response(str(raw_response_content)) try: parsed_response = BiasAnalysisResponse.model_validate_json(cleaned_response_str) log = update_log(f"Analyse biais objectif (fin après clean): Biais trouvés - {len(parsed_response.detected_biases)}", log) return parsed_response.model_dump(), log except Exception as final_parse_error: error_msg_detail = f"Erreur parsing JSON final: {final_parse_error}. Nettoyé: '{cleaned_response_str[:200]}...'" print(error_msg_detail) log = update_log(f"ERREUR Analyse biais parsing final (mode {format_type}): {final_parse_error}", log) return BiasAnalysisResponse(overall_comment=f"Erreur technique parsing réponse JSON (mode {format_type}): {final_parse_error}").model_dump(), log # --- Gestion des erreurs API --- except openai.BadRequestError as e: error_type = type(e).__name__; error_details = repr(e) user_error_msg = f"Erreur Requête API ({error_type}). Vérifiez param/modèle." log_msg_prefix = f"ERREUR API Call ({api_source}, {model_name})" if "response_format" in str(e): user_error_msg += f" Problème format réponse ({response_format_config.get('type', '?') if response_format_config else '?'})." log_msg = f"{log_msg_prefix}: Problème format réponse. Détails: {error_details}" elif "model" in str(e): user_error_msg += " Modèle invalide ou non trouvé." log_msg = f"{log_msg_prefix}: Modèle invalide. Détails: {error_details}" else: log_msg = f"{log_msg_prefix}: {str(e)}. Détails: {error_details}" print(log_msg); log = update_log(log_msg, log) return BiasAnalysisResponse(overall_comment=user_error_msg).model_dump(), log except openai.AuthenticationError as e: error_msg = f"Erreur auth API ({api_source}). Vérifiez clé."; print(error_msg) log = update_log(f"ERR API Auth ({api_source}): {error_msg}", log) return BiasAnalysisResponse(overall_comment=error_msg).model_dump(), log except openai.RateLimitError as e: error_msg = f"Erreur API ({api_source}): Limite taux atteinte."; print(error_msg) log = update_log(f"ERR API RateLimit ({api_source}): {error_msg}", log) return BiasAnalysisResponse(overall_comment=error_msg).model_dump(), log except Exception as e: error_type = type(e).__name__; error_details = repr(e) user_error_msg = f"Erreur technique analyse ({error_type}). Vérifiez connexion/modèle." log_msg = f"ERR Analyse biais API Call ({error_type} sur {api_source}, {model_name}): {str(e)}. Détails: {error_details}" print(log_msg); log = update_log(log_msg, log) return BiasAnalysisResponse(overall_comment=user_error_msg).model_dump(), log """Analyse les biais dans l'objectif marketing en forçant un schéma JSON.""" log = session_log_state log = update_log(f"Analyse biais objectif (début): '{objective_text[:50]}...'", log) if not objective_text: return BiasAnalysisResponse(overall_comment="Veuillez fournir un objectif marketing.").model_dump(), update_log("Analyse biais: Objectif vide.", log) active_client, error_msg = get_active_client(app_config) if error_msg: log = update_log(f"ERREUR Analyse biais: {error_msg}", log) return BiasAnalysisResponse(overall_comment=f"Erreur: {error_msg}").model_dump(), log model_name = app_config["text_model"] # --- Génération et MODIFICATION du Schéma JSON --- try: bias_schema = BiasAnalysisResponse.model_json_schema() except Exception as schema_e: log = update_log(f"ERREUR Génération schéma Pydantic: {schema_e}", log) return BiasAnalysisResponse(overall_comment=f"Erreur interne génération schéma: {schema_e}").model_dump(), log # --- System Prompt --- system_prompt = f""" Tu es un expert en marketing éthique et en psychologie cognitive, spécialisé dans la création de personas. Analyse l'objectif marketing suivant : "{objective_text}" Identifie les BIAIS COGNITIFS POTENTIELS ou RISQUES DE STÉRÉOTYPES pertinents pour la création de personas. Concentre-toi sur : 1. **Stéréotypes / Généralisations Hâtives :** Suppose-t-on des traits basés sur le genre, l'âge, l'ethnie, le statut socio-économique sans justification ? (Ex: 'tous les jeunes urbains sont écolos') 2. **Biais de Confirmation / Affinité :** L'objectif semble-t-il chercher à valider une idée préconçue ou refléter trop les opinions du concepteur ? (Ex: 'prouver que notre produit est parfait pour CE type de personne') 3. **Simplification Excessive / Manque de Nuance :** Le groupe cible est-il décrit de manière trop monolithique, ignorant la diversité interne ? (Ex: 'les seniors actifs' sans différencier leurs motivations ou capacités) 4. **Autres biais pertinents** (Ex: Oubli de fréquence de base, Biais de normalité si applicable). Pour chaque biais potentiel identifié : - Nomme le type de biais (ex: Stéréotype d'âge). - Explique brièvement POURQUOI c'est un risque DANS CE CONTEXTE de création de persona. - Propose un CONSEIL PRÉCIS pour nuancer l'objectif ou être vigilant lors de la création. Structure ta réponse en utilisant le format JSON suivant (avec la classe Pydantic BiasAnalysisResponse): {{ "detected_biases": [ {{ "bias_type": "Type de biais identifié", "explanation": "Explication contextuelle du risque.", "advice": "Conseil spécifique d'atténuation." }} ], "overall_comment": "Bref commentaire général. Indique si aucun biais majeur n'est détecté." }} Réponds en français. S'il n'y a pas de biais clair, retourne une liste 'detected_biases' vide et indique-le dans 'overall_comment'. Ne retourne PAS de texte brut ou d'explications supplémentaires. Utilise uniquement le format JSON ci-dessus. """ response_content_str = "" try: # --- Appel API avec Structured Output --- completion = active_client.chat.completions.create( model=model_name, messages=[{"role": "user", "content": system_prompt}], temperature=0.2, max_tokens=2400, response_format={ "type": "json_schema", "json_schema": { "name": "bias_analysis", "strict": True, "description": "Analyse des biais potentiels dans un objectif marketing.", "schema": bias_schema } }, ) raw_response_content = completion.choices[0].message.content try: parsed_response = BiasAnalysisResponse.model_validate_json(raw_response_content) log = update_log(f"Analyse biais objectif (fin): Biais trouvés - {len(parsed_response.detected_biases)}", log) return parsed_response.model_dump(), log except (json.JSONDecodeError, TypeError, ValueError) as direct_parse_error: log = update_log(f"ERREUR Parsing direct réponse JSON Schema: {direct_parse_error}. Contenu brut: {raw_response_content!r}", log) cleaned_response_str = clean_json_response(str(raw_response_content)) try: parsed_response = BiasAnalysisResponse.model_validate_json(cleaned_response_str) log = update_log(f"Analyse biais objectif (fin après clean): Biais trouvés - {len(parsed_response.detected_biases)}", log) return parsed_response.model_dump(), log except Exception as final_parse_error: error_msg = f"Erreur parsing JSON final: {final_parse_error}. Nettoyé: '{cleaned_response_str[:200]}...'" print(error_msg); log = update_log(f"ERREUR Analyse biais parsing final: {final_parse_error}", log) return BiasAnalysisResponse(overall_comment=f"Erreur technique parsing réponse JSON Schema: {final_parse_error}").model_dump(), log except openai.BadRequestError as e: error_type = type(e).__name__; error_details = repr(e) user_error_msg = f"Erreur Requête API ({error_type}). Vérifiez les paramètres/schéma." if "Invalid schema for response_format" in str(e): user_error_msg += " Problème avec le format de réponse demandé." log_msg = f"ERREUR API Call: Schéma JSON invalide selon l'API. Détails: {error_details}" else: log_msg = f"ERREUR API Call ({error_type}): {str(e)}. Détails: {error_details}" print(log_msg); log = update_log(log_msg, log) return BiasAnalysisResponse(overall_comment=user_error_msg).model_dump(), log except openai.AuthenticationError as e: error_msg = f"Erreur auth API ({app_config.get('api_source', '?')}). Vérifiez clé."; print(error_msg); log = update_log(f"ERR API Auth: {error_msg}", log); return BiasAnalysisResponse(overall_comment=error_msg).model_dump(), log except openai.RateLimitError as e: error_msg = f"Erreur API ({app_config.get('api_source', '?')}): Limite taux."; print(error_msg); log = update_log(f"ERR API RateLimit: {error_msg}", log); return BiasAnalysisResponse(overall_comment=error_msg).model_dump(), log except Exception as e: error_type = type(e).__name__; error_details = repr(e) user_error_msg = f"Erreur technique analyse ({error_type}). Vérifiez connexion/modèle." log_msg = f"ERR Analyse biais API Call ({error_type}): {str(e)}. Détails: {error_details}" print(log_msg); log = update_log(log_msg, log) return BiasAnalysisResponse(overall_comment=user_error_msg).model_dump(), log # --- display_bias_analysis --- def display_bias_analysis(analysis_result): if not analysis_result: return [("Aucune analyse effectuée.", None)] biases = analysis_result.get("detected_biases", []) overall_comment = analysis_result.get("overall_comment", "") highlighted_data = [] if "Erreur" in overall_comment: highlighted_data.append((overall_comment, "ERROR")) elif not biases: highlighted_data.append((overall_comment or "Aucun biais majeur détecté.", "INFO")) else: if overall_comment: highlighted_data.append((overall_comment + "\n\n", "COMMENT")) for bias_info in biases: highlighted_data.append((f"⚠️ {bias_info.get('bias_type', '?')}: ", "BIAS_TYPE")) highlighted_data.append((f"{bias_info.get('explanation', '-')}\n", "EXPLANATION")) highlighted_data.append((f"💡 Conseil: {bias_info.get('advice', '-')}\n", "ADVICE")) return highlighted_data # --- generate_persona_image --- def generate_persona_image(app_config, *args): """Génère l'image du persona via OpenAI ou OpenRouter.""" inputs = args[:-1] session_log_state = args[-1] log = session_log_state (first_name, last_name, age, gender, persona_description_en, skin_color, eye_color, hair_style, hair_color, facial_expression, posture, clothing_style, accessories) = inputs api_source = app_config.get("api_source") image_gen_enabled = app_config.get("image_generation_enabled", False) if not image_gen_enabled: log = update_log("Génération image: Désactivée (API non configurée ou non supportée).", log) return None, log, "Génération d'image désactivée ou non supportée par la configuration API actuelle." active_client, client_error_msg = get_active_client(app_config) if client_error_msg: log = update_log(f"ERREUR Génération image (Client): {client_error_msg}", log) return None, log, f"Erreur client API pour génération image: {client_error_msg}" if not first_name or not last_name or not age or not gender: return None, log, "Veuillez remplir prénom, nom, âge et genre pour générer l'image." # --- Construction du Prompt --- gender_en = gender_mapping.get(gender, "person") lens_aperture = "Kodak Portra 400" lighting = "soft natural light" photo_style_details = f"portrait {lighting}, shot on {lens_aperture}" base_description = ( f"{photo_style_details} of {first_name} {last_name}, " f"a {age}-year-old {gender_en}. " ) # Ajout des détails optionnels (moins d'emphase sur "skin texture") details = "" if skin_color_mapping.get(skin_color): details += f"Skin tone: {skin_color_mapping[skin_color]}. " if eye_color_mapping.get(eye_color): details += f"Eye color: {eye_color_mapping[eye_color]}. " if hair_style_mapping.get(hair_style): details += f"Hairstyle: {hair_style_mapping[hair_style]}. " if hair_color_mapping.get(hair_color): details += f"Hair color: {hair_color_mapping[hair_color]}. " if facial_expression_mapping.get(facial_expression): details += f"Facial expression: {facial_expression_mapping[facial_expression]}. " if accessories_mapping.get(accessories): details += f"Wearing: {accessories_mapping[accessories]}. " if clothing_style_mapping.get(clothing_style): details += f"Wearing {clothing_style_mapping[clothing_style]} clothing. " # Le contexte peut être utile pour l'environnemental if persona_description_en: details += f"Context: {persona_description_en}. " # Négatifs (garder ce qui est pertinent) final_prompt = f"{base_description}{details}" log = update_log(f"Génération image (début via {api_source}): Prompt='{final_prompt[:100]}...'", log) pil_image = None try: if api_source == "openai": response = active_client.images.generate( model=OPENAI_IMAGE_MODEL, prompt=final_prompt, size="1024x1024", n=1, response_format="url", quality="standard" ) image_url = response.data[0].url img_response = requests.get(image_url) img_response.raise_for_status() pil_image = Image.open(io.BytesIO(img_response.content)) elif api_source == "openrouter": # --- Appel via OpenRouter Chat Completions --- response = active_client.chat.completions.create( model=OPENROUTER_IMAGE_MODEL, messages=[{"role": "user", "content": final_prompt}], ) headers = { "Authorization": f"Bearer {openrouter_api_key}", "Content-Type": "application/json", } payload = { "model": OPENROUTER_IMAGE_MODEL, "messages": [{"role": "user", "content": final_prompt}], "max_tokens": 150, "n": 1, } api_url = "https://openrouter.ai/api/v1/chat/completions" http_response = requests.post(api_url, headers=headers, json=payload) http_response.raise_for_status() response_data = http_response.json() image_base64_data = None if response_data.get("choices"): message = response_data["choices"][0].get("message", {}) content = message.get("content") if isinstance(content, list): for part in content: if part.get("type") == "image_url": image_url_obj = part.get("image_url", {}) image_base64_data = image_url_obj.get("url") break if not image_base64_data: log = update_log(f"ERREUR Image OpenRouter: Image non trouvée dans la réponse. Réponse: {str(response_data)[:500]}", log) raise ValueError("Réponse OpenRouter ne contient pas d'URL d'image base64.") if not image_base64_data.startswith("data:image"): raise ValueError(f"URL d'image invalide reçue: {image_base64_data[:100]}...") image_base64_string = image_base64_data.split(',', 1)[1] image_bytes = base64.b64decode(image_base64_string) pil_image = Image.open(io.BytesIO(image_bytes)) else: raise ValueError(f"Source API non supportée pour la génération d'image: {api_source}") log = update_log(f"Génération image (fin via {api_source}): Succès.", log) return pil_image, log, None except (openai.AuthenticationError, openai.RateLimitError, openai.BadRequestError, requests.exceptions.RequestException, ValueError, KeyError) as e: error_type = type(e).__name__ error_msg_detail = str(e) if hasattr(e, 'response') and e.response is not None: try: error_msg_detail += f" | Détail API: {e.response.text[:200]}" except: pass user_error_msg = f"Erreur génération image via {api_source} ({error_type})." full_log_msg = f"ERREUR Génération image via {api_source} ({error_type}): {error_msg_detail}" print(full_log_msg) log = update_log(full_log_msg, log) return None, log, user_error_msg except Exception as e: error_type = type(e).__name__ error_msg = f"Erreur inattendue génération image via {api_source} ({error_type}): {str(e)}" print(error_msg); log = update_log(error_msg, log) return None, log, f"Erreur inattendue ({error_type}) lors de la génération d'image." # --- refine_persona_details --- def refine_persona_details(app_config, first_name, last_name, age, field_name, field_value, bias_analysis_dict, marketing_objectives, session_log_state): log = session_log_state log = update_log(f"Raffinement (début): Champ='{field_name}', Valeur='{field_value[:50]}...'", log) active_client, error_msg = get_active_client(app_config) if error_msg: log = update_log(f"ERREUR Raffinement: {error_msg}", log); return log, f"ERREUR: {error_msg}", field_name model_name = app_config["text_model"] biases_text = "Aucune analyse de biais précédente." if bias_analysis_dict: try: detected = bias_analysis_dict.get("detected_biases", []) biases_text = "\n".join([f"- {b.get('bias_type','?')}: {b.get('explanation','-')}" for b in detected]) if detected else bias_analysis_dict.get("overall_comment", "Aucun biais majeur détecté.") except Exception as e: biases_text = f"Err lecture biais: {e}"; log = update_log(f"ERR Lecture Biais Dict: {e}", log) system_prompt = f""" Tu es un assistant IA expert en marketing éthique, aidant à affiner le persona marketing pour '{first_name} {last_name}' ({age} ans). L'objectif marketing initial était : "{marketing_objectives}" L'analyse initiale de cet objectif a soulevé les points suivants : {biases_text} Tâche: Concentre-toi UNIQUEMENT sur le champ '{field_name}' dont la valeur actuelle est '{field_value}'. Propose 1 à 2 suggestions CONCISES et ACTIONNABLES pour améliorer, nuancer ou enrichir cette valeur. Tes suggestions doivent viser à : - Rendre le persona plus réaliste et moins cliché. - ATTÉNUER spécifiquement les biais potentiels listés ci-dessus s'ils sont pertinents pour ce champ. - Rester cohérent avec l'objectif marketing général. - Éviter les généralisations excessives. Si la valeur actuelle semble bonne ou si tu manques de contexte pour faire une suggestion pertinente, indique-le simplement (ex: "La valeur actuelle semble appropriée." ou "Difficile de suggérer sans plus de contexte."). Réponds en français. Ne fournis QUE les suggestions ou le commentaire d'approbation/manque de contexte. Ne répète pas la question. Ne fournis pas d'explications supplémentaires ou de texte brut. Utilise un format clair et concis.""" suggestions = "" try: response = active_client.chat.completions.create( model=model_name, messages=[{"role": "user", "content": system_prompt}], temperature=0.6, max_tokens=800, ) suggestions = response.choices[0].message.content.strip() log = update_log(f"Raffinement (fin): Champ='{field_name}'. Suggestions: '{suggestions[:50]}...'", log) return log, suggestions, field_name except openai.AuthenticationError as e: error_msg = f"Erreur auth API ({app_config.get('api_source', '?')}) raffinement."; print(error_msg); log = update_log(f"ERR API Auth (Refine): {error_msg}", log); return log, f"ERREUR: {error_msg}", field_name except openai.RateLimitError as e: error_msg = f"Erreur API ({app_config.get('api_source', '?')}) (Refine): Limite taux."; print(error_msg); log = update_log(f"ERR API RateLimit (Refine): {error_msg}", log); return log, f"ERREUR: {error_msg}", field_name except Exception as e: error_msg = f"Erreur raffinement '{field_name}': {str(e)}"; print(error_msg); log = update_log(f"ERR Raffinement '{field_name}': {str(e)}", log); return log, f"ERREUR: {error_msg}", field_name # --- generate_summary --- def generate_summary(persona_image_pil, *args): session_log_state = args[-1]; inputs = args[:-1]; log = session_log_state (first_name, last_name, age, gender, persona_description_en, skin_color, eye_color, hair_style, hair_color, facial_expression, posture, clothing_style, accessories, marital_status, education_level, profession, income, personality_traits, values_beliefs, motivations, hobbies_interests, main_responsibilities, daily_activities, technology_relationship, product_related_activities, pain_points, product_goals, usage_scenarios, brand_relationship, market_segment, commercial_objectives, visual_codes, special_considerations, daily_life, references) = inputs log = update_log(f"Génération résumé: Pour '{first_name} {last_name}'.", log) summary = ""; image_html = "
\n" if not first_name or not last_name or not age: summary += "

Infos base manquantes

Prénom, nom, âge requis (Étape 2).

"; image_html += "

Image non générée.

" else: if persona_image_pil and isinstance(persona_image_pil, Image.Image): try: buffered = io.BytesIO(); img_to_save = persona_image_pil.copy() if img_to_save.mode == 'RGBA' or 'transparency' in img_to_save.info: img_to_save = img_to_save.convert('RGB') img_to_save.save(buffered, format="JPEG", quality=85); img_bytes = buffered.getvalue() img_base64 = base64.b64encode(img_bytes).decode(); img_data_url = f"data:image/jpeg;base64,{img_base64}" image_html += f"Persona {first_name}\n" except Exception as e: img_err_msg = f"Erreur encodage image: {e}"; image_html += f"

{img_err_msg}

"; log = update_log(f"ERR Encodage Image Résumé: {e}", log) else: image_html += "

Aucune image disponible.

" summary += f"

{first_name} {last_name}, {age} ans ({gender})

" def add_section(title, fields): content = "" for label, value in fields.items(): should_add = (label == "Revenus annuels (€)" and value is not None) or (label != "Revenus annuels (€)" and value) if should_add: if label == "Revenus annuels (€)" and isinstance(value, (int, float)): try: value_str = f"{int(value):,} €".replace(",", " ") except ValueError: value_str = str(value) + " €" else: value_str = str(value) value_str_html = markdown.markdown(value_str).replace('

', '').replace('

', '').strip().replace("\n", "
") content += f"{label}: {value_str_html}
\n" return f"

{title}

\n{content}\n" if content else "" summary += add_section("Infos socio-démographiques", {"État civil": marital_status, "Niveau d'éducation": education_level, "Profession": profession, "Revenus annuels (€)": income}) summary += add_section("Psychographie", {"Traits de personnalité": personality_traits, "Valeurs et croyances": values_beliefs, "Motivations intrinsèques": motivations, "Hobbies et intérêts": hobbies_interests}) summary += add_section("Relation au produit/service", {"Relation technologie": technology_relationship, "Tâches liées": product_related_activities, "Points de douleur": pain_points, "Objectifs produit": product_goals, "Scénarios d'utilisation": usage_scenarios}) summary += add_section("Contexte pro/quotidien", {"Responsabilités principales": main_responsibilities, "Activités journalières": daily_activities, "Journée type/Citation": daily_life}) summary += add_section("Marketing & considérations", {"Relation marque": brand_relationship, "Segment marché": market_segment, "Objectifs commerciaux": commercial_objectives, "Codes visuels": visual_codes, "Considérations spéciales": special_considerations, "Références/Sources": references}) image_html += "
" final_html = f"
{summary}
{image_html}
" return final_html, log # --- Interface Gradio --- css = ".suggestion-box {border: 1px solid #e0e0e0; border-radius: 5px; padding: 10px; margin: 10px 0; background-color: #f9f9f9;} .suggestion-box h4 { margin-top: 0; margin-bottom: 5px; }" with gr.Blocks(theme=gr.themes.Default(), css=css) as demo: gr.Markdown("# PersonaGenAI : Assistant de création de persona marketing") gr.Markdown("Outil d'aide à la création de personas, intégrant un système d'IA générative (OpenRouter ou OpenAI) pour stimuler la créativité et la réflexivité face aux biais.") # --- États Globaux --- app_config_state = gr.State(value={"api_source": None, "text_model": None, "image_generation_enabled": False, "openai_key_provided": False, "openrouter_key_provided": bool(openrouter_api_key)}) bias_analysis_result_state = gr.State(value={}) persona_image_pil_state = gr.State(value=None) session_log_state = gr.State(value="") status_message_state = gr.State(value="") last_refinement_suggestion_state = gr.State(value=None) # --- Affichage Statut Global --- status_display = gr.Markdown(value="", elem_classes="status-message") def update_status_display(new_message, current_log): if new_message and any(k in new_message for k in ["ERREUR", "WARN", "Configuration"]): current_log = update_log(f"STATUS: {new_message}", current_log) return new_message, current_log # --- Onglets --- with gr.Tabs() as tabs: # --- Onglet 0 : Configuration API --- with gr.Tab("🔑 Configuration API", id=-1): gr.Markdown("### Configuration des clés API") gr.Markdown("Cet outil utilise un système d'IA. Choisissez votre fournisseur. En l'absence de saisie d'une clé API, un mode par défaut sera utilisé.") gr.Markdown("**Note :** Si vous avez une clé OpenAI valide, elle sera utilisée pour la génération d'images et de texte. Sinon, OpenRouter sera utilisé pour le texte uniquement (images désactivées).") if openrouter_api_key: gr.Markdown("✅ Clé API **OpenRouter** trouvée.") else: gr.Markdown("❌ **Clé API OpenRouter non trouvée.** Mode OpenRouter indisponible sans clé.") openai_api_key_input = gr.Textbox(label="Clé API OpenAI (optionnelle)", type="password", placeholder="Entrez clé OpenAI pour DALL-E 3 / GPT", info="Si valide: utilisée pour images ET texte. Sinon: OpenRouter (si clé dispo) pour texte.") configure_api_button = gr.Button("Appliquer la configuration") api_status_display = gr.Markdown("Statut API : Non configuré.") def configure_api_clients(openai_key, current_config, current_log): """Configure le client API et met à jour l'état.""" openai_key_provided = bool(openai_key); openrouter_key_available = current_config["openrouter_key_provided"] status_msg = ""; config = current_config.copy(); active_api_client_holder["client"] = None; active_api_client_holder["openai_key"] = None api_source = None; text_model = None; image_enabled = False; client_to_store = None if openai_key_provided: try: temp_client = OpenAI(api_key=openai_key); temp_client.models.list() # Test client_to_store = temp_client; active_api_client_holder["openai_key"] = openai_key api_source = "openai"; text_model = OPENAI_TEXT_MODEL; image_enabled = True status_msg = f"✅ Config **OpenAI** active (Texte: `{text_model}`, Images: {OPENAI_IMAGE_MODEL} direct)."; config["openai_key_provided"] = True current_log = update_log("Config: Client OpenAI OK.", current_log) except openai.AuthenticationError: status_msg = "⚠️ Clé OpenAI **invalide**."; current_log = update_log("ERR Config OpenAI: Clé invalide.", current_log); config["openai_key_provided"] = False; openai_key_provided = False except Exception as e: status_msg = f"⚠️ Clé OpenAI fournie mais erreur: {str(e)}."; current_log = update_log(f"ERR Config OpenAI: {e}", current_log); config["openai_key_provided"] = False; openai_key_provided = False elif openrouter_key_available: try: temp_client = OpenAI(base_url="https://openrouter.ai/api/v1", api_key=openrouter_api_key) client_to_store = temp_client; api_source = "openrouter"; text_model = OPENROUTER_TEXT_MODEL image_enabled = False status_msg = f"✅ Config **OpenRouter** active (Texte: `{text_model}`)."; config["openai_key_provided"] = False current_log = update_log("Config: Client OpenRouter OK (Images désactivées).", current_log) except Exception as e: status_msg = f"❌ Erreur init OpenRouter: {e}."; current_log = update_log(f"ERR Config OpenRouter: {e}", current_log); client_to_store = None; api_source = None; text_model = None; image_enabled = False; config["openai_key_provided"] = False else: status_msg = "❌ Aucune clé API valide disponible/configurée." client_to_store = None; api_source = None; text_model = None; image_enabled = False; config["openai_key_provided"] = False active_api_client_holder["client"] = client_to_store config["api_source"] = api_source; config["text_model"] = text_model; config["image_generation_enabled"] = image_enabled log_msg = f"Config API. Source: {api_source or 'Aucune'}, Images: {'Actif' if image_enabled else 'Inactif'}." if "OK." not in current_log.splitlines()[-1]: current_log = update_log(log_msg, current_log) return config, status_msg, current_log configure_api_button.click(configure_api_clients, [openai_api_key_input, app_config_state, session_log_state], [app_config_state, api_status_display, session_log_state]) # --- Onglet 1 : Objectif & Analyse Biais --- with gr.Tab("🎯 Étape 1 : Objectif & analyse biais", id=0): gr.Markdown("### 1. Définissez l'objectif marketing") gr.Markdown("Pourquoi créez-vous ce persona ? Le système d'IA analysera l'objectif pour identifier des biais potentiels.") with gr.Row(): objective_input = gr.Textbox(label="Objectif marketing", lines=4, scale=3) with gr.Column(scale=1): gr.Markdown("Suggestions :") suggestion_button1 = gr.Button("Ex 1 : Service éco urbain", size="sm") suggestion_button2 = gr.Button("Ex 2 : App fitness seniors", size="sm") analyze_button = gr.Button("🔍 Analyser l'objectif (biais)") gr.Markdown("---"); gr.Markdown("### Analyse des biais potentiels") bias_analysis_output_highlighted = gr.HighlightedText(label="Biais détectés et conseils", show_legend=True, color_map={"BIAS_TYPE":"coral", "EXPLANATION":"lightgray", "ADVICE":"green", "INFO":"blue", "COMMENT":"orange", "ERROR":"red"}) gr.Markdown("---"); gr.Markdown("### 🤔 Réflexion") user_reflection_on_biases = gr.Textbox(label="Comment utiliser cette analyse ?", lines=2, placeholder="Ex: Attention au stéréotype X...") log_reflection_button = gr.Button("📝 Enregistrer réflexion", size='sm') suggestion1_text = "Créer un persona pour promouvoir un nouveau service de livraison écologique destiné aux jeunes professionnels urbains soucieux de l'environnement (25-35 ans)." suggestion2_text = "Développer une application mobile de fitness personnalisée pour les seniors actifs (+65 ans) cherchant à maintenir une vie saine et sociale." suggestion_button1.click(lambda: suggestion1_text, outputs=objective_input) suggestion_button2.click(lambda: suggestion2_text, outputs=objective_input) analyze_button.click( fn=lambda: gr.update(interactive=False), inputs=None, outputs=[analyze_button] ).then( fn=analyze_biases, inputs=[app_config_state, objective_input, session_log_state], outputs=[bias_analysis_result_state, session_log_state] ).then( fn=display_bias_analysis, inputs=bias_analysis_result_state, outputs=bias_analysis_output_highlighted ).then( fn=lambda r, l: update_status_display(r.get("overall_comment", "") if "Erreur" in r.get("overall_comment", "") else "", l), inputs=[bias_analysis_result_state, session_log_state], outputs=[status_display, session_log_state] ).then( fn=lambda: gr.update(interactive=True), inputs=None, outputs=[analyze_button] ) def log_user_reflection(r, l): return update_log(f"Réflexion (1): '{r}'", l) if r else l log_reflection_button.click(log_user_reflection, [user_reflection_on_biases, session_log_state], [session_log_state]) # --- Onglet 2 : Image & Infos Base --- with gr.Tab("👤 Étape 2 : Image & infos de base", id=1): gr.Markdown("### 2. Identité visuelle et informations de base") with gr.Row(): with gr.Column(scale=1): first_name_input = gr.Textbox(label="Prénom") last_name_input = gr.Textbox(label="Nom") age_input = gr.Slider(label="Âge", minimum=18, maximum=100, step=1, value=30) gender_input = gr.Radio(label="Genre", choices=["Homme", "Femme", "Non-binaire"], value="Homme") persona_description_en_input = gr.Textbox(label="Contexte image (optionnel, anglais)", lines=1, info="Ex: 'reading book', 'working on laptop'") with gr.Accordion("🎨 Détails visuels (optionnel)", open=False): with gr.Row(): skin_color_input = gr.Dropdown(label="Teint", choices=list(skin_color_mapping.keys()), value="") ; eye_color_input = gr.Dropdown(label="Yeux", choices=list(eye_color_mapping.keys()), value="") with gr.Row(): hair_style_input = gr.Dropdown(label="Coiffure", choices=list(hair_style_mapping.keys()), value="") ; hair_color_input = gr.Dropdown(label="Cheveux (couleur)", choices=list(hair_color_mapping.keys()), value="") with gr.Row(): facial_expression_input = gr.Dropdown(label="Expression", choices=list(facial_expression_mapping.keys()), value="") ; posture_input = gr.Dropdown(label="Posture", choices=list(posture_mapping.keys()), value="") with gr.Row(): clothing_style_input = gr.Dropdown(label="Style vêtements", choices=list(clothing_style_mapping.keys()), value="") ; accessories_input = gr.Dropdown(label="Accessoires", choices=list(accessories_mapping.keys()), value="") reset_visuals_button = gr.Button("Réinitialiser détails", size="sm") with gr.Column(scale=1): persona_image_output = gr.Image(label="Image du persona", type="pil", interactive=False) generate_image_button = gr.Button("🖼️ Générer / Mettre à jour l'image", interactive=False) gr.Markdown("💡 **Attention :** Les systèmes d'IA générative peuvent reproduire des stéréotypes. Clé OpenAI requise.", elem_classes="warning") visual_inputs = [skin_color_input, eye_color_input, hair_style_input, hair_color_input, facial_expression_input, posture_input, clothing_style_input, accessories_input] reset_visuals_button.click(lambda: [""] * len(visual_inputs), outputs=visual_inputs) def handle_image_generation(*args): app_config = args[0]; log_state = args[-1]; persona_inputs = args[1:-1] pil_image, updated_log, error_message = generate_persona_image(app_config, *persona_inputs, log_state) status_update_msg = ""; info_popup_msg = None if error_message: if any(k in error_message for k in ["Veuillez remplir", "désactivée"]): info_popup_msg = error_message else: status_update_msg = error_message if info_popup_msg: gr.Info(info_popup_msg) return pil_image, updated_log, status_update_msg generate_image_inputs = [app_config_state, first_name_input, last_name_input, age_input, gender_input, persona_description_en_input] + visual_inputs + [session_log_state] generate_image_outputs = [persona_image_pil_state, session_log_state, status_message_state] generate_image_button.click(handle_image_generation, generate_image_inputs, generate_image_outputs).then(lambda img: img, persona_image_pil_state, persona_image_output).then(update_status_display, [status_message_state, session_log_state], [status_display, session_log_state]) app_config_state.change(lambda cfg: gr.update(interactive=cfg.get("image_generation_enabled", False)), app_config_state, generate_image_button) # --- Onglet 3 : Profil Détaillé & Raffinement --- with gr.Tab("📝 Étape 3 : Profil détaillé & raffinement", id=2): gr.Markdown("### 3. Complétez les détails du persona") gr.Markdown("Utilisez '💡' pour obtenir des suggestions du système d'IA afin de nuancer ce champ.") refinement_suggestion_display = gr.Markdown("*Cliquez sur '💡' à côté d'un champ pour une suggestion.*", elem_classes="suggestion-box") with gr.Row(): with gr.Column(): gr.Markdown("#### Infos socio-démographiques") marital_status_input = gr.Dropdown(label="État civil", choices=["", "Célibataire", "En couple", "Marié(e)", "Divorcé(e)", "Veuf(ve)"]) education_level_input = gr.Dropdown(label="Niveau d'éducation", choices=["", "Secondaire", "Bac", "Licence", "Master", "Doctorat", "Autre"]) profession_input = gr.Textbox(label="Profession") income_input = gr.Number(label="Revenus annuels (€)", minimum=0, step=1000) gr.Markdown("#### Psychographie") with gr.Row(equal_height=False): personality_traits_input = gr.Textbox(label="Traits personnalité", lines=2, scale=4); refine_personality_traits_button = gr.Button("💡", scale=1, size='sm') with gr.Row(equal_height=False): values_beliefs_input = gr.Textbox(label="Valeurs, croyances", lines=2, scale=4); refine_values_beliefs_button = gr.Button("💡", scale=1, size='sm') with gr.Row(equal_height=False): motivations_input = gr.Textbox(label="Motivations", lines=2, scale=4); refine_motivations_button = gr.Button("💡", scale=1, size='sm') with gr.Row(equal_height=False): hobbies_interests_input = gr.Textbox(label="Loisirs, intérêts", lines=2, scale=4); refine_hobbies_interests_button = gr.Button("💡", scale=1, size='sm') with gr.Column(): gr.Markdown("#### Relation produit/service") with gr.Row(equal_height=False): technology_relationship_input = gr.Textbox(label="Relation technologie", lines=2, scale=4, info="Ex: adopte vite, prudent..."); refine_technology_relationship_button = gr.Button("💡", scale=1, size='sm') with gr.Row(equal_height=False): product_related_activities_input = gr.Textbox(label="Tâches liées produit/service", lines=2, scale=4); refine_product_related_activities_button = gr.Button("💡", scale=1, size='sm') with gr.Row(equal_height=False): pain_points_input = gr.Textbox(label="Points de douleur", lines=2, scale=4); refine_pain_points_button = gr.Button("💡", scale=1, size='sm') with gr.Row(equal_height=False): product_goals_input = gr.Textbox(label="Objectifs avec produit/service", lines=2, scale=4); refine_product_goals_button = gr.Button("💡", scale=1, size='sm') with gr.Row(equal_height=False): usage_scenarios_input = gr.Textbox(label="Scénarios d'utilisation", lines=2, scale=4); refine_usage_scenarios_button = gr.Button("💡", scale=1, size='sm') with gr.Accordion("Autres informations (optionnel)", open=False): with gr.Row(): with gr.Column(): gr.Markdown("#### Contexte pro/quotidien") with gr.Row(equal_height=False): main_responsibilities_input = gr.Textbox(label="Responsabilités", lines=2, scale=4); refine_main_responsibilities_button = gr.Button("💡", scale=1, size='sm') with gr.Row(equal_height=False): daily_activities_input = gr.Textbox(label="Activités journalières", lines=2, scale=4); refine_daily_activities_button = gr.Button("💡", scale=1, size='sm') with gr.Row(equal_height=False): daily_life_input = gr.Textbox(label="Journée type/Citation", lines=2, scale=4); refine_daily_life_button = gr.Button("💡", scale=1, size='sm') with gr.Column(): gr.Markdown("#### Marketing & considérations") with gr.Row(equal_height=False): brand_relationship_input = gr.Textbox(label="Relation marque", lines=2, scale=4); refine_brand_relationship_button = gr.Button("💡", scale=1, size='sm') with gr.Row(equal_height=False): market_segment_input = gr.Textbox(label="Segment marché", lines=2, scale=4); refine_market_segment_button = gr.Button("💡", scale=1, size='sm') with gr.Row(equal_height=False): commercial_objectives_input = gr.Textbox(label="Objectifs commerciaux", lines=2, scale=4); refine_commercial_objectives_button = gr.Button("💡", scale=1, size='sm') with gr.Row(equal_height=False): visual_codes_input = gr.Textbox(label="Codes visuels/Marques", lines=2, scale=4); refine_visual_codes_button = gr.Button("💡", scale=1, size='sm') with gr.Row(equal_height=False): special_considerations_input = gr.Textbox(label="Considérations spéciales", lines=2, scale=4); refine_special_considerations_button = gr.Button("💡", scale=1, size='sm') with gr.Row(equal_height=False): references_input = gr.Textbox(label="Références/Sources", lines=2, scale=4); refine_references_button = gr.Button("💡", scale=1, size='sm') def display_refinement_suggestion(suggestion_state): if suggestion_state: field_name, suggestion_text = suggestion_state if "ERREUR:" not in suggestion_text: return f"#### Suggestion pour '{field_name}' :\n\n{suggestion_text}" else: return "*Erreur lors du dernier raffinement (voir statut/log).*" return "*Cliquez sur '💡' pour une suggestion.*" def handle_refinement_request(app_config, fname, lname, age_val, field_name_display, field_val, bias_state_dict, objectives, log_state): updated_log, result_text, field_name_ctx = refine_persona_details(app_config, fname, lname, age_val, field_name_display, field_val, bias_state_dict, objectives, log_state) status_update_msg = ""; suggestion_details = None if result_text: if "ERREUR:" in result_text: status_update_msg = result_text; gr.Warning(f"Erreur raffinement '{field_name_display}'. Voir log.") else: suggestion_details = (field_name_display, result_text) else: gr.Warning(f"Pas de suggestion pour '{field_name_display}'.") return updated_log, status_update_msg, suggestion_details def create_refine_handler(f_name, i_comp): return lambda app_c, fn, ln, age, f_val, bias_s, obj, log_s: handle_refinement_request(app_c, fn, ln, age, f_name, f_val, bias_s, obj, log_s) common_ref_inputs = [app_config_state, first_name_input, last_name_input, age_input] state_ref_inputs = [bias_analysis_result_state, objective_input, session_log_state] refine_handler_outputs = [session_log_state, status_message_state, last_refinement_suggestion_state] refine_buttons_map = { refine_personality_traits_button: ("Traits personnalité", personality_traits_input), refine_values_beliefs_button: ("Valeurs, croyances", values_beliefs_input), refine_motivations_button: ("Motivations", motivations_input), refine_hobbies_interests_button: ("Loisirs, intérêts", hobbies_interests_input), refine_technology_relationship_button: ("Relation technologie", technology_relationship_input), refine_product_related_activities_button: ("Tâches liées", product_related_activities_input), refine_pain_points_button: ("Points de douleur", pain_points_input), refine_product_goals_button: ("Objectifs produit", product_goals_input), refine_usage_scenarios_button: ("Scénarios utilisation", usage_scenarios_input), refine_main_responsibilities_button: ("Responsabilités", main_responsibilities_input), refine_daily_activities_button: ("Activités journalières", daily_activities_input), refine_daily_life_button: ("Journée type/Citation", daily_life_input), refine_brand_relationship_button: ("Relation marque", brand_relationship_input), refine_market_segment_button: ("Segment marché", market_segment_input), refine_commercial_objectives_button: ("Objectifs commerciaux", commercial_objectives_input), refine_visual_codes_button: ("Codes visuels/Marques", visual_codes_input), refine_special_considerations_button: ("Considérations spéciales", special_considerations_input), refine_references_button: ("Références/Sources", references_input), } for btn, (label, input_comp) in refine_buttons_map.items(): btn.click( fn=create_refine_handler(label, input_comp), inputs=common_ref_inputs + [input_comp] + state_ref_inputs, outputs=refine_handler_outputs ).then(update_status_display, [status_message_state, session_log_state], [status_display, session_log_state] ).then(display_refinement_suggestion, [last_refinement_suggestion_state], [refinement_suggestion_display]) # --- Onglet 4 : Résumé Persona --- with gr.Tab("📄 Étape 4 : Résumé du persona", id=3): gr.Markdown("### 4. Visualisez le persona complet") summary_button = gr.Button("Générer le résumé") summary_content = gr.Markdown(elem_classes="persona-summary", value="Cliquez sur 'Générer'...") all_summary_inputs = [persona_image_pil_state, first_name_input, last_name_input, age_input, gender_input, persona_description_en_input, skin_color_input, eye_color_input, hair_style_input, hair_color_input, facial_expression_input, posture_input, clothing_style_input, accessories_input, marital_status_input, education_level_input, profession_input, income_input, personality_traits_input, values_beliefs_input, motivations_input, hobbies_interests_input, main_responsibilities_input, daily_activities_input, technology_relationship_input, product_related_activities_input, pain_points_input, product_goals_input, usage_scenarios_input, brand_relationship_input, market_segment_input, commercial_objectives_input, visual_codes_input, special_considerations_input, daily_life_input, references_input, session_log_state] summary_button.click(generate_summary, all_summary_inputs, [summary_content, session_log_state]) # --- Onglet 5 : Journal de Bord --- with gr.Tab("📓 Journal de bord", id=4): gr.Markdown("### Suivi du processus") gr.Markdown("Historique des actions, réflexions et erreurs.") log_display_final = gr.Textbox(label="Historique session", lines=20, interactive=False, max_lines=MAX_LOG_LINES) download_log_button = gr.DownloadButton(label="Télécharger journal", visible=False) export_log_button_final = gr.Button("Préparer export journal") session_log_state.change(fn=lambda log: log, inputs=session_log_state, outputs=log_display_final) def prep_log_dl(log): if not log: return gr.update(visible=False) try: with tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt', encoding='utf-8') as tf: tf.write(log); fp = tf.name print(f"Log prêt: {fp}"); return gr.update(value=fp, visible=True) except Exception as e: print(f"Err création log DL: {e}"); return gr.update(visible=False) export_log_button_final.click(prep_log_dl, session_log_state, download_log_button) # --- Lancement App --- if not openrouter_api_key: print("\n"+"="*60+"\nWARN: Clé OpenRouter manquante. Fonctionnement limité à OpenAI si clé fournie.\n"+"="*60+"\n") initial_api_status = "Statut API : Non configuré." if openrouter_api_key: print("Clé OR trouvée, config initiale...") try: initial_config, initial_api_status, initial_log = configure_api_clients(None, app_config_state.value, "") app_config_state.value = initial_config; session_log_state.value = initial_log print(initial_api_status); api_status_display.value = initial_api_status except Exception as init_e: print(f"ERR config initiale OR: {init_e}"); initial_api_status = f"❌ Err config initiale OR: {init_e}"; api_status_display.value = initial_api_status demo.queue().launch(debug=False, share=False, pwa=True)