Midjourney Prompt Engineering ?????????????????????
Midjourney ???????????? AI image generation tool ?????????????????????????????????????????? text prompts ??????????????????????????????????????????????????????????????????????????????????????? prompt engineering ???????????????????????? ???????????????????????? prompt ????????????????????????????????? Subject (??????????????????????????????), Style (????????????????????????), Details (??????????????????????????????), Parameters (????????? settings ????????? Midjourney)
Feature Flag Management ?????????????????? AI Prompts ????????????????????? ?????????????????????????????? prompts ?????????????????? feature flags ?????? software development ????????????/????????? prompt versions ??????????????????????????????????????? deploy code ???????????? ??????????????? prompt ????????????????????? users ??????????????????????????????????????? rollout ????????????????????? rollback ?????? prompt ??????????????????????????????????????????????????????????????????
???????????????????????? Rapid iteration ??????????????? prompt ?????????????????????????????????, A/B testing ????????????????????????????????? prompt versions, Risk reduction rollback ?????????????????? prompt ??????????????????????????????????????????, Analytics ?????????????????????????????? prompt version, Collaboration ??????????????????????????????????????????????????? prompt library
Feature Flag ?????????????????? AI Prompts
?????????????????? feature flag system ?????????????????? prompts
# === Prompt Feature Flag System ===
cat > prompt_flags.yaml << 'EOF'
prompt_feature_flags:
product_image_generator:
name: "Product Image Generator"
description: "Generate product images for e-commerce"
enabled: true
variants:
control:
weight: 50
prompt_template: |
{product_name}, product photography, white background,
studio lighting, high resolution, 4k, commercial photography
parameters: "--ar 1:1 --style raw --v 6"
treatment_a:
weight: 25
prompt_template: |
{product_name}, professional product shot, minimalist setup,
soft shadows, clean composition, editorial style, 8k detail
parameters: "--ar 1:1 --style raw --v 6 --stylize 200"
treatment_b:
weight: 25
prompt_template: |
{product_name}, lifestyle product photography, natural light,
warm tones, bokeh background, premium feel, magazine quality
parameters: "--ar 4:3 --style raw --v 6 --stylize 300"
targeting:
rules:
- attribute: "product_category"
operator: "in"
values: ["electronics", "fashion", "home"]
- attribute: "user_plan"
operator: "equals"
value: "premium"
metrics:
- "click_through_rate"
- "conversion_rate"
- "user_rating"
rollout:
percentage: 100
sticky: true # Same user always gets same variant
social_media_banner:
name: "Social Media Banner"
enabled: true
variants:
v1:
weight: 70
prompt_template: |
{brand_message}, social media banner, modern design,
vibrant colors, eye-catching, {platform} format
parameters: "--ar 16:9 --v 6"
v2:
weight: 30
prompt_template: |
{brand_message}, digital marketing banner, bold typography,
gradient background, professional, {platform} optimized
parameters: "--ar 16:9 --v 6 --stylize 400"
rollout:
percentage: 50 # Only 50% of users get this feature
EOF
echo "Prompt feature flags configured"??????????????? Prompt Management System
Python system ???????????????????????????????????? prompts
#!/usr/bin/env python3
# prompt_manager.py ??? AI Prompt Management System
import json
import logging
import hashlib
from typing import Dict, List, Any, Optional
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger("prompts")
class PromptManager:
"""Feature Flag-based Prompt Management"""
def __init__(self):
self.prompts = {}
self.versions = {}
self.metrics = {}
def register_prompt(self, name, config):
"""Register a prompt with variants"""
self.prompts[name] = config
self.versions[name] = config.get("version", 1)
def evaluate(self, prompt_name, context):
"""Evaluate which prompt variant to use"""
config = self.prompts.get(prompt_name)
if not config or not config.get("enabled", False):
return None
# Check targeting rules
if not self._check_targeting(config.get("targeting", {}), context):
return None
# Check rollout percentage
rollout = config.get("rollout", {}).get("percentage", 100)
user_id = context.get("user_id", "anonymous")
user_hash = int(hashlib.md5(f"{prompt_name}:{user_id}".encode()).hexdigest(), 16) % 100
if user_hash >= rollout:
return None
# Select variant based on weights
variant = self._select_variant(config["variants"], user_id, prompt_name)
# Render template
template = variant["prompt_template"]
rendered = template.format(**context)
return {
"prompt_name": prompt_name,
"variant_name": variant["name"],
"rendered_prompt": rendered.strip(),
"parameters": variant.get("parameters", ""),
"version": self.versions[prompt_name],
}
def _check_targeting(self, targeting, context):
rules = targeting.get("rules", [])
if not rules:
return True
for rule in rules:
attr = context.get(rule["attribute"])
if rule["operator"] == "equals" and attr != rule["value"]:
return False
elif rule["operator"] == "in" and attr not in rule.get("values", []):
return False
return True
def _select_variant(self, variants, user_id, prompt_name):
total_weight = sum(v.get("weight", 1) for v in variants.values())
user_hash = int(hashlib.md5(f"{prompt_name}:variant:{user_id}".encode()).hexdigest(), 16) % total_weight
cumulative = 0
for name, variant in variants.items():
cumulative += variant.get("weight", 1)
if user_hash < cumulative:
return {**variant, "name": name}
first_name = list(variants.keys())[0]
return {**variants[first_name], "name": first_name}
def record_metric(self, prompt_name, variant, metric_name, value):
"""Record metric for prompt variant"""
key = f"{prompt_name}:{variant}:{metric_name}"
if key not in self.metrics:
self.metrics[key] = []
self.metrics[key].append(value)
def get_analytics(self, prompt_name):
"""Get analytics for prompt"""
results = {}
for key, values in self.metrics.items():
if key.startswith(prompt_name):
parts = key.split(":")
variant = parts[1]
metric = parts[2]
if variant not in results:
results[variant] = {}
results[variant][metric] = {
"count": len(values),
"avg": round(sum(values) / len(values), 3) if values else 0,
"min": min(values) if values else 0,
"max": max(values) if values else 0,
}
return results
# Demo
manager = PromptManager()
# Register product image prompt
manager.register_prompt("product_image", {
"enabled": True,
"version": 3,
"variants": {
"control": {
"weight": 50,
"prompt_template": "{product_name}, product photography, white background, studio lighting, 4k",
"parameters": "--ar 1:1 --v 6",
},
"lifestyle": {
"weight": 50,
"prompt_template": "{product_name}, lifestyle photography, natural light, warm tones, premium feel",
"parameters": "--ar 4:3 --v 6 --stylize 300",
},
},
"rollout": {"percentage": 100},
})
# Evaluate for different users
for user_id in ["user_001", "user_002", "user_003"]:
result = manager.evaluate("product_image", {
"user_id": user_id,
"product_name": "Wireless Headphones AX-500",
})
if result:
print(f"{user_id} ??? {result['variant_name']}: {result['rendered_prompt'][:60]}...")A/B Testing Prompts
??????????????? prompt variants
#!/usr/bin/env python3
# ab_testing.py ??? Prompt A/B Testing Framework
import json
import logging
import random
import math
from typing import Dict, List
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger("ab_test")
class PromptABTest:
"""A/B Testing for AI Prompts"""
def __init__(self, test_name):
self.test_name = test_name
self.variants = {}
self.results = {}
def add_variant(self, name, prompt_config):
self.variants[name] = prompt_config
self.results[name] = {"impressions": 0, "conversions": 0, "ratings": []}
def record_impression(self, variant):
self.results[variant]["impressions"] += 1
def record_conversion(self, variant):
self.results[variant]["conversions"] += 1
def record_rating(self, variant, rating):
self.results[variant]["ratings"].append(rating)
def get_results(self):
"""Calculate A/B test results"""
output = {}
for name, data in self.results.items():
impressions = data["impressions"]
conversions = data["conversions"]
ratings = data["ratings"]
cvr = conversions / impressions if impressions > 0 else 0
avg_rating = sum(ratings) / len(ratings) if ratings else 0
output[name] = {
"impressions": impressions,
"conversions": conversions,
"conversion_rate": round(cvr * 100, 2),
"avg_rating": round(avg_rating, 2),
"sample_size": impressions,
}
return output
def statistical_significance(self, variant_a, variant_b):
"""Check if difference is statistically significant"""
a = self.results[variant_a]
b = self.results[variant_b]
n_a = a["impressions"]
n_b = b["impressions"]
if n_a == 0 or n_b == 0:
return {"significant": False, "reason": "Not enough data"}
p_a = a["conversions"] / n_a
p_b = b["conversions"] / n_b
# Z-test for two proportions
p_pool = (a["conversions"] + b["conversions"]) / (n_a + n_b)
se = math.sqrt(p_pool * (1 - p_pool) * (1/n_a + 1/n_b)) if p_pool > 0 else 1
z_score = (p_b - p_a) / se if se > 0 else 0
significant = abs(z_score) > 1.96 # 95% confidence
return {
"variant_a_cvr": round(p_a * 100, 2),
"variant_b_cvr": round(p_b * 100, 2),
"lift": round((p_b - p_a) / p_a * 100, 2) if p_a > 0 else 0,
"z_score": round(z_score, 3),
"significant": significant,
"confidence": "95%" if significant else "< 95%",
"winner": variant_b if z_score > 1.96 else variant_a if z_score < -1.96 else "No winner yet",
}
# Demo: A/B Test product image prompts
test = PromptABTest("product_image_v3")
test.add_variant("control", {"prompt": "product photography, white background"})
test.add_variant("lifestyle", {"prompt": "lifestyle photography, natural light"})
# Simulate data
random.seed(42)
for _ in range(500):
test.record_impression("control")
if random.random() < 0.12: # 12% CVR
test.record_conversion("control")
test.record_rating("control", random.uniform(3.5, 5.0))
for _ in range(500):
test.record_impression("lifestyle")
if random.random() < 0.15: # 15% CVR
test.record_conversion("lifestyle")
test.record_rating("lifestyle", random.uniform(3.8, 5.0))
results = test.get_results()
print("A/B Test Results:")
for name, data in results.items():
print(f" {name}: CVR={data['conversion_rate']}%, Rating={data['avg_rating']}, n={data['sample_size']}")
sig = test.statistical_significance("control", "lifestyle")
print(f"\nStatistical Significance:")
print(f" Lift: {sig['lift']}%")
print(f" Significant: {sig['significant']} ({sig['confidence']})")
print(f" Winner: {sig['winner']}")Version Control ????????? Rollback
?????????????????? prompt versions
# === Prompt Version Control ===
cat > prompt_versioning.yaml << 'EOF'
prompt_versions:
product_image:
current_version: 3
versions:
v1:
created: "2024-01-15"
author: "john"
status: "archived"
prompt: "product photo, white background, studio"
parameters: "--ar 1:1 --v 5.2"
performance:
conversion_rate: 8.5
avg_rating: 3.8
v2:
created: "2024-03-01"
author: "jane"
status: "archived"
prompt: "product photography, white background, studio lighting, high resolution, 4k"
parameters: "--ar 1:1 --style raw --v 6"
change_log: "Upgraded to v6, added style raw"
performance:
conversion_rate: 12.1
avg_rating: 4.2
v3:
created: "2024-05-15"
author: "john"
status: "active"
prompt: "product photography, white background, studio lighting, 4k, commercial photography"
parameters: "--ar 1:1 --style raw --v 6"
change_log: "Added commercial photography keyword"
performance:
conversion_rate: 13.8
avg_rating: 4.4
rollback_policy:
auto_rollback: true
conditions:
- metric: "conversion_rate"
operator: "drops_below"
threshold: 10
window: "24h"
- metric: "error_rate"
operator: "exceeds"
threshold: 5
window: "1h"
rollback_to: "previous_version"
EOF
# Git-based prompt versioning
cat > .gitattributes << 'EOF'
prompts/*.yaml diff
prompts/*.json diff
EOF
cat > prompts/product_image.json << 'EOF'
{
"name": "product_image",
"version": 3,
"variants": {
"default": {
"template": "{product_name}, product photography, white background, studio lighting, 4k, commercial photography",
"parameters": "--ar 1:1 --style raw --v 6",
"negative_prompt": "blurry, low quality, distorted, watermark"
}
},
"metadata": {
"author": "john",
"created": "2024-05-15",
"tags": ["product", "e-commerce", "photography"]
}
}
EOF
echo "Version control configured"Monitoring ????????? Analytics
????????????????????????????????? prompts
#!/usr/bin/env python3
# prompt_analytics.py ??? Prompt Analytics Dashboard
import json
import logging
from typing import Dict, List
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger("analytics")
class PromptAnalytics:
def __init__(self):
pass
def dashboard(self):
return {
"overview": {
"total_prompts": 15,
"active_ab_tests": 3,
"total_generations_24h": 12500,
"avg_rating": 4.3,
},
"top_prompts": [
{"name": "product_image_v3", "generations": 4500, "cvr": 13.8, "rating": 4.4},
{"name": "social_banner_v2", "generations": 3200, "cvr": 8.5, "rating": 4.1},
{"name": "blog_illustration", "generations": 2800, "cvr": 11.2, "rating": 4.3},
{"name": "avatar_generator", "generations": 1200, "cvr": 22.5, "rating": 4.6},
],
"ab_tests": [
{
"test": "product_image_lifestyle",
"status": "running",
"days_active": 7,
"control_cvr": 12.1,
"treatment_cvr": 15.3,
"lift": "+26.4%",
"significant": True,
"recommendation": "Deploy treatment (lifestyle variant)",
},
{
"test": "banner_bold_typography",
"status": "running",
"days_active": 3,
"control_cvr": 8.5,
"treatment_cvr": 9.1,
"lift": "+7.1%",
"significant": False,
"recommendation": "Need more data (min 500 samples per variant)",
},
],
"cost_tracking": {
"monthly_api_cost": 450,
"cost_per_generation": 0.036,
"generations_per_conversion": 7.2,
"cost_per_conversion": 0.26,
},
}
analytics = PromptAnalytics()
dash = analytics.dashboard()
print("Prompt Analytics Dashboard:")
print(f" Active Prompts: {dash['overview']['total_prompts']}")
print(f" Generations (24h): {dash['overview']['total_generations_24h']}")
print("\nTop Prompts:")
for p in dash["top_prompts"]:
print(f" {p['name']}: {p['generations']} gens, CVR={p['cvr']}%, Rating={p['rating']}")
print("\nA/B Tests:")
for t in dash["ab_tests"]:
print(f" {t['test']}: {t['lift']} lift, Significant={t['significant']}")
print(f" ??? {t['recommendation']}")
print(f"\nCost: /conversion")FAQ ??????????????????????????????????????????
Q: ????????????????????????????????? feature flags ?????????????????? AI prompts?
A: AI prompts ?????????????????????????????????????????????????????????????????????????????????????????? prompt ???????????????????????????????????????????????????????????????????????????????????????????????????????????? Feature flags ???????????? ??????????????? prompt ????????????????????? users ??????????????????????????????????????? (?????? risk), Rollback ???????????????????????? prompt ?????????????????????????????????????????? (????????????????????? deploy code), A/B test ????????????????????????????????? prompt versions ???????????? data ????????????, Gradual rollout ?????????????????????????????? % users ?????????????????? prompt ????????????, Audit trail ??????????????????????????????????????????????????? ????????????????????? prompt ???????????? ??????????????????????????? ??????????????????????????? feature flags ???????????? hardcode prompts ?????? code ?????????????????????????????????????????????????????? prompt ???????????? deploy ???????????? ??????????????????????????? A/B test ?????????????????????
Q: Midjourney parameters ???????????????????????????????????????????????????????
A: Parameters ???????????? --ar (aspect ratio) ????????????????????????????????????????????? ???????????? --ar 16:9, --ar 1:1, --ar 9:16, --v (version) ??????????????? model version ???????????? --v 6, --v 5.2, --style (style mode) raw ??????????????????????????? prompt ?????????????????????, --stylize (stylization) 0-1000 ?????????????????? = artistic ????????????????????? default 100, --chaos (variety) 0-100 ?????????????????? = ?????????????????????????????????????????????, --quality (render quality) 0.25, 0.5, 1 ?????????????????? = detail ?????????????????????, --no (negative prompt) ?????????????????????????????????????????????????????????????????? ?????????????????? product photography ??????????????? --style raw --v 6 --ar ?????????????????????????????? ?????????????????? creative artwork ????????? --stylize 300-500
Q: ??????????????????????????????????????? prompt library ??????????????????????
A: ?????????????????????????????????????????? Level 1 (Simple) ?????????????????? JSON/YAML files ?????? Git repository version control ???????????? Git commits ???????????????????????????????????? Level 2 (Structured) ????????? database (PostgreSQL) ???????????? prompts + versions + metrics ??????????????? admin UI ???????????????????????????????????? ???????????????????????????????????? Level 3 (Platform) ????????? prompt management platform ???????????? PromptLayer, Humanloop, LangSmith ?????? built-in versioning, analytics, A/B testing ????????????????????????????????????????????? ?????????????????????????????????????????? Git + YAML ????????????????????? ??????????????? prompts ????????????????????? (50+) ?????????????????????????????? database ???????????? platform
Q: Midjourney ????????? DALL-E ????????? Stable Diffusion ??????????????????????????????????
A: Midjourney ??????????????????????????????????????????????????? (???????????????????????? artistic style) ????????????????????? Discord ??????????????? API ????????????????????????????????????????????? (????????????????????? workaround) ???????????? $10-60/month ??????????????? creative work, marketing DALL-E 3 (OpenAI) ?????? API ???????????????????????? integrate ????????? ChatGPT ????????? text rendering ?????????????????? ??????????????? automated workflows ???????????? $0.04-0.12/image Stable Diffusion open source self-host ????????? (?????????) customize ??????????????????????????? (fine-tune, ControlNet, LoRA) ???????????? GPU ?????????????????? local inference ??????????????? developers ?????????????????????????????? control ????????????????????? ?????????????????? feature flag management ??????????????? DALL-E 3 (API ??????) ???????????? Stable Diffusion (self-host, API ???????????????????????????) Midjourney ??????????????????????????? automation ?????????????????????????????? official API
