"AI Image Prompting 2026 — The 8-Element Formula and How Each Tool Differs"

Subject · Scene · Camera · Lighting · Style — one structure that works across Nano Banana, Midjourney, and GPT Image

ํ•ต์‹ฌ ์š”์•ฝ

  • Audience: You read the tool comparison and picked one — but your prompts keep collapsing into the same generic look or missing the brief.
  • What you'll get: 1) The 8-element prompt formula that works in all three tools, 2) tool-specific differences (Midjourney parameters, Nano Banana's natural language, GPT Image's conversational edits), 3) before/after concrete examples, 4) seven common mistakes, 5) weights and negative prompts.
  • Mental model: A prompt isn't a list of words. It's a director's note to a camera crew — lens, light, blocking, mood — not "make it pretty."

1. The 8-element formula (works in all tools)

Synthesizing 2026 official guides from Black Forest Labs, Anthropic, and Midjourney, they all converge on the same order:

[Subject] + [Scene/Environment] + [Composition/Shot] + [Camera/Lens]
+ [Lighting] + [Style/Medium] + [Color/Mood] + [Quality/Negatives]

1.1 What goes where

# Element Example phrasing
1 Subject "Korean woman in her 30s," "black cat," "lighthouse on a cliff"
2 Scene "rainy Tokyo alley," "sunlit cafe window seat"
3 Composition "close-up," "wide-angle," "rule of thirds," "over-the-shoulder"
4 Camera/Lens "35mm film," "85mm portrait lens," "shallow depth of field"
5 Lighting "golden hour," "rim light," "softbox front," "neon back light"
6 Style/Medium "photorealistic," "watercolor," "oil painting," "Studio Ghibli style"
7 Color/Mood "muted pastel palette," "high contrast," "moody, melancholic"
8 Quality/Negatives "8K, sharp focus" / "no text, no watermark"

1.2 Order matters

Diffusion models put more weight on words near the front (2026 guide). Put your primary subject and key action in the first 10–15 words.

1.3 Full example

"A 30-something Korean woman in a beige trench coat [Subject] walking through a rainy Seoul alley at night [Scene] medium shot, slight low angle [Composition] shot on 50mm prime, shallow depth of field [Camera] soft neon backlight from shop signs [Lighting] photorealistic, cinematic still [Style] muted teal and amber palette, melancholic [Mood] 8K, sharp focus, no watermark [Quality]"

This works as-is in Nano Banana 2, Midjourney, and GPT Image 1.5.

2. Where the tools diverge

2.1 Nano Banana 2 (natural-language friendly)

Plain prose works fine. Text rendering is strong, so you can directly request on-image text.

"A book cover for 'AI for Beginners' — minimalist white background, serif title in black, geometric illustration of a circuit board with leaves growing out of it, soft gradient orange-to-yellow accent, clean modern editorial design"

2.2 Midjourney v7 (parameter-driven precision)

Use parameters to fine-tune style strength, diversity, and consistency (official parameters).

Parameter Effect Recommended values
--s (stylize) Aesthetic strength 100 (default), 50 (faithful to prompt), 750 (heavy stylization)
--c (chaos) Diversity across the four outputs 0–50 normal, 50–100 experimental
--ar (aspect ratio) Aspect ratio 16:9, 2:3, 1:1
--seed Lock the seed Vary one element while keeping the base
--sref Style-reference URL Mimic the style of another image
--oref Character-reference URL Keep a person consistent across prompts

Example:

A medieval castle on a cliff, sunrise, cinematic, fog --ar 16:9 --s 250 --c 30

V7 specials: personalization profiles (--p) apply your trained taste; Draft Mode (--draft) gives 10× faster ideation.

2.3 GPT Image 1.5 (conversational)

Its strength isn't the first generation — it's iterative editing. Use the 8-element formula on the first call, then plain conversation afterwards.

1st: "A young man holding an espresso cup, cafรฉ window seat, morning light, photorealistic, 50mm lens, shallow depth of field" 2nd: "Same image, but change the cup to a glass of orange juice" 3rd: "Now add a dog sleeping under the table"

Each step persists. It tracks the previous image, which gives the highest cross-edit consistency.

3. Before / After

Before (vague)

"Pretty landscape photo"

After 1 (specific)

"A photorealistic landscape of a quiet mountain lake at golden hour, mirror-like water reflection, autumn maple trees at the shore, mist rising from the surface, wide-angle composition, shot on 24mm lens, warm orange and teal palette, sharp focus, 8K"

After 2 (Midjourney parameters added)

A photorealistic landscape of a quiet mountain lake at golden hour, mirror-like water reflection, autumn maple trees at the shore, mist rising from the surface, wide-angle composition, shot on 24mm lens, warm orange and teal palette, sharp focus --ar 21:9 --s 200 --c 20

After 3 (GPT Image conversational)

After the first generation: "Same scene but at twilight with a faint full moon over the mountains."

Same intent, 5–10× quality gap.

4. Seven common mistakes

Mistake Result Fix
1. Stacking adjectives ("amazing, beautiful, stunning") Mostly ignored Replace with concrete description ("misty rim light, gold-tipped autumn leaves")
2. Negatives ("not blurry") Ignored or reversed Use positive form ("sharp focus, fine detail")
3. Too many elements at once Some get dropped Keep 3–5 priorities, push the rest into edits
4. Repeating the same word No effect Use weights: ((emphasis)) or word::2
5. Generic quality tags ("8K, ultra-realistic") Weak signal Describe actual detail ("pores visible on skin, fabric texture")
6. Missing detail anchors for people Hands and eyes break "natural hands, anatomically correct, sharp eyes"
7. Not pinning a seed Can't iterate --seed (Midjourney) or save a generated image to lock in GPT Image

5. Weights and negative prompts

5.1 Weights (Midjourney)

:: followed by a number controls per-token influence.

red sports car::3, urban street::1, neon signs::0.5

→ The car gets 3× weight, the street is baseline, neon is downweighted.

5.2 Negative prompts (Stable Diffusion-family, Midjourney --no)

--no text, watermark, signature, blur, low quality

Midjourney doesn't auto-honor negative phrasing — use the --no parameter. Nano Banana 2 and GPT Image have weaker negative-prompt support; prefer positive phrasing.

5.3 Reference images

Tool How
Nano Banana 2 Attach an image + natural language ("in this style")
Midjourney --sref [URL] for style, --oref [URL] for character consistency
GPT Image Attach an image, then say "in this style"

6. Copy-paste starter templates

Portrait

[Person description] in [Location], [shot type] shot, [Lighting], 
shot on [Lens] with shallow depth of field, photorealistic, 
[Mood] mood, 8K, sharp focus, natural hands and eyes

Landscape

A photorealistic landscape of [Subject] at [Time of day], 
[weather/atmosphere], [composition type] composition, 
shot on [Lens], [color palette] palette, sharp focus, 8K

Illustration / concept art

[Subject] in [Setting], [art style — e.g., Studio Ghibli / Moebius / 
watercolor], [color palette], [lighting], detailed line art, 
[mood], --ar 16:9 --s 400

Product mockup

[Product] on [Surface], studio lighting with softbox front, 
[background — clean white / wooden table], 50mm macro lens, 
shallow depth of field, photorealistic, commercial photography

Book cover / poster

A book cover design for "[Title]" — [layout description], 
[typography — serif/sans, color], [illustration concept], 
[color palette], minimalist editorial design, --ar 2:3

Developer notes

  1. Templatize prompts: Python f-strings or LangChain PromptTemplate with {subject}, {lighting}, etc. — essential when generating 100+ images.
  2. Automate quality scoring: GPT-5 Vision or Claude Vision can score "prompt fidelity." Auto-regenerate anything below threshold.
  3. Save (seed, prompt, model_version): a small DB makes good outputs reproducible.
  4. Nano Banana 2 batch: API supports n=4 per call → choose the best automatically.
  5. Midjourney --sref / --oref automation: not recommended via unofficial bots — ToS and stability concerns. Stick with OpenAI/Gemini for production automation.
  6. IP-safety filter: pre-filter prompts for real-person and brand mentions before submission.

References

This is part 5-2 of 11 in the AI Basics series. Next: AI voice/video — Suno, Runway, and Sora.

๋Œ“๊ธ€

๋Œ“๊ธ€ ์“ฐ๊ธฐ

์ด ๋ธ”๋กœ๊ทธ์˜ ์ธ๊ธฐ ๊ฒŒ์‹œ๋ฌผ

Agent Memory Engine (2/10) — Building an AI Agent Memory System with SQLite Alone

← 1/10 SQLite Memory Engine f… ๐Ÿ“š Series Index 3/10 memcore Completed → A memory engine that runs without daemons, without dependencies, anywhere ํ•ต์‹ฌ ์š”์•ฝ This post covers how to design an AI agent memory engine (memcore) on a single-file SQLite backend. - Structure: 25 modules, ~6,464 lines, 22 DB tables - Principles: no daemon, single-file portability, no required external dependencies, host-agnostic - Core techniques: 4-layer hybrid search (topic · vector · FTS5 · LIKE), U-tag dialectic-based confidence evolution, ingestion-layer bias rejection rules What This Post Covers One approach to building AI agent memory without a server or cloud vector DB. Specifically: (1) the limitations of text-file-based memory, (2) vector DB alternatives compared and selection criteria, (3) how to layer search tiers, (4) how to evolve confidence scores over time, and (5) how to block biased memory accumulation at the ingestion stage. Design Principles: Why Single-File SQLite Text-file-based ...
Read more »

"ML Foundations (9/9) — PyTorch vs TensorFlow, and the Road to Local LLMs"

์ด๋ฏธ์ง€
← 8/9 Deep Learning Architec… ๐Ÿ“š Series Index (series end) The final part. So far we've focused on models. Now we focus on the tools that actually run them . We'll lay out the philosophical difference between PyTorch and TensorFlow, and trace how Hugging Face transformers , llama.cpp , MLX , and Ollama built the bridge to running large language models on your own machine. By the end you should have the full mental model of "download a pretrained LLM and serve it locally." 0. Learning Objectives Explain the eager-vs-graph difference between PyTorch and TensorFlow. Explain, in graph terms, how autograd automates the backward pass. Use Hugging Face transformers ' AutoModel , AutoTokenizer , and pipeline abstractions. Describe llama.cpp's GGUF quantization, INT4 inference, and CPU-first flow. Describe Apple MLX's use of unified memory and how it differs from PyTorch. Run a local LLM with Ollama and call it through an OpenAI-compatible API. ...
Read more »

"RAG Core Study (14/26) — Evaluation Sets with RAGAS & DeepEval"

์ด๋ฏธ์ง€
Series overview: Series index ๐Ÿ“š Series Index Without an evaluation set, every RAG improvement is a story, not evidence. RAG systems fail in more than one place: retrieval, context selection, answer generation, and grounding. That means "looks better" is not a metric. Part 14 explains how to build a golden dataset , how RAGAS and DeepEval fit into the workflow, and why teams must separate retrieval quality from answer quality if they want tuning decisions to hold up. 0. Prerequisites Part 13 reranking — system quality now depends on multi-stage ranking. Part 1 grounding — a correct-looking answer is not enough. Part 12 Hybrid — multiple retrieval settings need comparison on fixed data. 1. Learning Objectives Build a small but useful golden dataset for RAG. Distinguish retrieval metrics from answer metrics . Use RAGAS and DeepEval in the right roles. Avoid the common traps in judge-model-based evaluation. 2. ํ•ต์‹ฌ ์š”์•ฝ An evaluation set for RAG is ...
Read more »

"ML Foundations (8/9) — Deep Learning Architectures: CNN, RNN, Attention"

์ด๋ฏธ์ง€
← 7/9 Deep Learning Training ๐Ÿ“š Series Index 9/9 PyTorch vs TensorFlow → What did we keep adding on top of an MLP? Vision went CNN, sequences went RNN/LSTM, and eventually both converged on Attention and the Transformer. Each architecture is easier to remember if you read it as what it gave up to gain something else . This part is the one-line summary of the decisive inflection points: LeNet → AlexNet → ResNet → LSTM → Attention → Transformer. 0. Learning Objectives Explain why CNN's convolution, pooling, and weight sharing match images so well. Trace what got unblocked from LeNet → AlexNet → VGG → ResNet as depth grew. State BPTT for RNNs and the vanishing/exploding gradient problem. Write the LSTM gate equations (forget, input, output) with intuition. Write the attention formula in query/key/value form and the scaled dot-product variant. State the precise reasons Transformers replaced RNNs (parallelism + long-range dependencies). 1. ํ•ต์‹ฌ ์š”์•ฝ CNN : weight sharin...
Read more »

"ML Foundations (7/9) — Deep Learning Training: Optimizers, Regularization, Initialization"

← 6/9 Neural Networks ๐Ÿ“š Series Index 8/9 Deep Learning Architec… → Building an MLP, as we did in Part 6, is not the same as getting it to train well . The right optimizer, regularization, initialization, and learning rate — when these line up, deep networks converge. When they don't, the network refuses to learn at all. This part is the catalog of those crafts, with formulas, paper citations, and code in one place. 0. Learning Objectives Compare and write the update rules for SGD, Momentum, Nesterov, and Adam. Explain how Dropout, BatchNorm, and LayerNorm work and where they belong in a model. Derive the variance formulas for Xavier and He initialization and match them to activations. Implement step, cosine, and warmup learning-rate schedules in PyTorch. Explain why gradient clipping is effectively required for RNNs and Transformers. Diagnose the most common training failures (NaN, plateau, overfitting) and apply first-line fixes. 1. ํ•ต์‹ฌ ์š”์•ฝ SGD : \(w \leftarro...
Read more »

OpenClaw to Hermes Migration (2/13) — What to Preserve, Partially Port, or Discard

← 1/13 Current Structure Inve… ๐Ÿ“š Series Index 3/13 What Moves to Hermes → A code review record classifying operational scripts into three tiers: preserve, partial-preserve, and discard. What This Post Covers An asset classification framework (preserve / partial-preserve / discard) for migrating a legacy agent system to a successor A concrete case applying the Strangler Fig + Subset Migration strategy to a memory pipeline Classification results evaluated across LOC, external dependencies, and replaceability axes A method for slimming down a bloated single file (1,310 LOC) using the migration as a natural entry point Problem Definition: Port Everything or Cut Deep? The OpenClaw inventory comprises multiple agents, multiple scheduler/daemon processes, and multiple operational scripts. When moving to the successor system, Hermes, the choice is not binary. A full port carries accumulated technical debt along with it; a minimal port discards hard-won domain logic. Code-revie...
Read more »

AI Agents I Built (5/7) — Building an Automated Blogger API Publishing System

← 4/7 My Life Organizer Agent ๐Ÿ“š Series Index 6/7 Dual → Designing a Markdown-Based Auto-Publish Pipeline with Blogger API v3 + OAuth 2.0 Key Summary Blogger API v3 + OAuth 2.0 enables a pipeline that converts Markdown files into publishable HTML posts blogger_publish.py handles the full flow — frontmatter parsing → body cleanup → HTML conversion → CSS injection → API post — in a single script For bulk publishing, quota management (delay, retry, TOC suppression) is the critical design concern Platform Selection Criteria API access is mandatory for AI agent-driven content publishing. A comparison of platforms: Platform API Cost AdSense Decision WordPress.com REST API Paid plan required Paid only Cost overhead Ghost REST API Self-hosted or paid Manual setup Ops overhead Blogger v3 API Free Native integration Adopted Blogger is free, exposes a REST API, and integrates natively with AdSense. For personal blog automation, this combination i...
Read more »