GPT-Image2-Skill

🧭 About this project

Browse the images, open a prompt and try it with your own references. The gpt-image skill generates and edits images; get-prompt-from-image extracts prompts from images. The CLI and generation skill default to GPT Image 2.6 Flare. Image 2 prompts and the 2.5 reading material and task guides remain available.

For the CLI, export the relevant PDF pages as PNG, WebP or JPG, then attach them with -i. See the supported image-reference formats. Keep required text and edit constraints explicit.

For PPT work, try vector-style diagrams, icons and slide layouts. Save still images as PNG, JPEG or WebP; editable SVG or PowerPoint shapes need a separate authoring step.

After the image: animation and web

The image is the starting asset. On the left, the local tool slices an eight-cell sheet into a GIF. On the right, the brain office becomes an animation and a webpage with clickable rooms and a clock-in/out toggle.

See creative workflows for the commands and local demos. GitHub displays HTML source; download the repo and open the demos in a browser, or run a local server. The still images and full gallery continue below.

A · Fire sword: sheet → GIF · No. 171

An animation sprite sheet for a fantasy desktop cursor, inspired by a flaming-sword mouse pointer. Exactly 8 frames in a 4-column by 2-row uniform grid, read left to right then top to bottom. Canvas 2048 by 1024, each cell exactly 512 by 512. Transparent background, no gutters, no borders, no text, no labels. In every cell the same small ornate obsidian sword points diagonally toward the upper-left. Its sharp tip is fixed at cell coordinate (100,100), blade geometry and position identical in every frame. Bright amber and electric-blue flame curls around the blade, progressing smoothly through eight phases and returning to the initial shape at the loop boundary. The blade stays still; only flames, a few sparks and reflected light move. Keep all effects within the middle 360 by 360 pixels of each cell. Crisp illustrated game-UI asset, luminous edges, convincing metallic highlights, strong silhouette, no character, no duplicate swords within a cell. Do not draw checkerboards; use actual transparency.

gpt-image --model gpt-image-2.6-flare -p "<PROMPT ABOVE>" --size 2048x1024 --quality high --background transparent --format png --moderation auto -f sheet.png
gpt-image animate --sheet sheet.png --columns 4 --rows 2 --duration 160 \
  -f cursor.gif --preview player.html --title "Emberglass cursor"

B · Brain office: image → website · No. 167

Animate this exact illustrated brain-office cutaway as a short seamless looping GIF. Keep the camera, room layout, all signs and lettering fixed. Use small motions: the hamster turns the coffee wheel, tiny workers move papers along the conveyor, one paper airplane drifts near the roof, the red DEADLINE lamp gently blinks while the committee keeps eating biscuits. Preserve the drawing style and character identities. Most of the image stays still. End in the starting state. Return a single animated scene, not a grid or storyboard. No new text.

gpt-image --model gpt-image-2.6-flare \
  -i docs/infographics-field-guides/overthinking-departments-26.png \
  -p "<ANIMATION PROMPT ABOVE>" --format gif --frames 9 --frame-duration 200 \
  --size 512x768 --quality medium -f brain-office.gif

The sword tip drifts slightly, so it is an animation asset rather than a registered system cursor. Small lettering changes in the brain animation; the webpage uses real HTML for room names and descriptions. The local player supports pause, frame scrubbing, speed changes and download. It does not repair inconsistent generated frames.

✨ Made with GPT Image 2.6

We tried five ideas from X and Reddit with gpt-image-2.6-flare at high quality. Click an image for the original PNG, or open the prompts below to see the sources and our adaptations.

A and B use No. 31 · Ling Yuan's character sheet, with the reference prompt credited to @MANISH1027512. Some source posts predate 2.6. We generated the images shown here.

A · Failed camera roll · No. 164

Use the adult 26-year-old character in the reference sheet. 素人の失敗写真の数々、3x3、9:16。

gpt-image --model gpt-image-2.6-flare -i docs/character-design/character-sheet.png \
  -p "<EXACT PROMPT ABOVE>" --size 864x1536 --quality high -f docs/photography/failed-camera-roll-26.png

B · Cyber engineer's inventory · No. 166

Use the fictional adult 26-year-old character in the reference sheet. このキャラのカバンの中身を紹介する画像を作成して。キャラの性格が伝わる構成、モダンなソシャゲのインベントリ画面のようなデザイン。 Keep this tasteful and non-explicit. Use English for item labels. Show a densely packed inventory with visually distinct items and short humorous descriptions. Portrait 2:3.

gpt-image --model gpt-image-2.6-flare -i docs/character-design/character-sheet.png \
  -p "<EXACT PROMPT ABOVE>" --size 1024x1536 --quality high -f docs/gaming/engineer-inventory-26.png

C · Inside an overthinker's brain · No. 167

An elaborate illustrated cutaway of a fictional overthinker’s brain shaped like a chaotic six-storey office building. Title: INSIDE MY BRAIN. Each room has its own tiny employees, machines, incident and readable short sign. Top: IDEA DEPARTMENT launches paper airplanes into neighbouring offices. FACT CHECKING uses forklifts to move citation folders. REPLAYING ONE AWKWARD MOMENT has a cinema showing the same handshake. COFFEE POWER PLANT is a hamster-wheel espresso engine. FIVE-MINUTE TASK has construction workers expanding into adjacent rooms. SLEEP OFFICE is locked while employees nap in the corridor. A red deadline alarm is ignored by a committee eating biscuits. Draw absurd connections between these departments with pneumatic tubes, stairs and conveyor belts. Crisp editorial pen illustration with rich muted colour, numerous tiny visual jokes, large legible headings, portrait composition, fully illustrated rather than a chart of empty boxes.

gpt-image --model gpt-image-2.6-flare --moderation auto \
  -p "<EXACT PROMPT ABOVE>" --size 1024x1536 --quality high -f docs/infographics-field-guides/overthinking-departments-26.png

D · The five-minute fix · No. 168

A witty six-panel editorial comic on one portrait page, two columns by three rows. Title: JUST A FIVE-MINUTE FIX. Same fictional adult developer, age 30, round glasses, mustard hoodie in every panel. Panel 1: relaxed at a tidy laptop, a tiny task card says CHANGE BUTTON COLOR. Panel 2: opens a cupboard labelled CSS, an avalanche of tangled wires falls out. Panel 3: inspects a small bug with a magnifier, behind it towers a dinosaur labelled LEGACY. Panel 4: now wearing a hard hat, demolishes the room, sign REBUILD EVERYTHING. Panel 5: midnight, exhausted, surrounded by twelve glowing monitors and sticky notes, tiny cat asleep on keyboard, caption THREE DAYS LATER. Panel 6: proud before the same laptop, button is exactly its original blue, small speech bubble IT WAS FINE BEFORE. Clear expressive physical comedy, detailed backgrounds that escalate, crisp ink and flat spot colours on warm white paper, readable exact short text, no extra panels.

gpt-image --model gpt-image-2.6-flare --moderation auto \
  -p "<EXACT PROMPT ABOVE>" --size 1024x1536 --quality high -f docs/anime-manga/five-minute-fix-26.png

E · Flat-pack dragon assembly · No. 169

A deadpan flat-pack furniture assembly instruction sheet for assembling a living dragon, named DRAKON. Six numbered illustrated stages, black technical line art on warm white paper, small mustard-yellow warning boxes. At the top a neat parts inventory: dragon head, two wings, coiled tail, four feet, huge ribbed body, bag of tiny screws, one comically small hex key. Step 1 a cheerful stick person sorts parts. Step 2 two people struggle to align a wing. Step 3 one extra screw remains. Step 4 assembled dragon sneezes a tiny flame onto the instruction page. Step 5 dragon eats the hex key. Step 6 dragon has curled up on the sofa, crushing it, while owner calmly phones support. Tiny alternate wrong-way wing diagram crossed out. Bottom warning reads WARNING: DO NOT FEED AFTER ASSEMBLY. Precise readable diagrams, escalating visual comedy, clean generous gutters, portrait layout. Fictional fantasy assembly only, no realistic dangerous instructions. Do not use a real company logo.

gpt-image --model gpt-image-2.6-flare --moderation auto \
  -p "<EXACT PROMPT ABOVE>" --size 1024x1536 --quality high -f docs/infographics-field-guides/flatpack-dragon-26.png

A few details missed the brief. B uses the company name as the character's name. D never shows the original button colour, so the dialogue carries the final joke. The run record covers these checks. You can also find the shared room and night-market seek-and-find attempt in the Reference Gallery.

✨ Made with GPT Image 2.5

These earlier samples use gpt-image-2.5-sunburst at 2048x2048, high: a mechanical watch with numbered parts, and a multi-storey cafe district opened up to reveal its interiors.

A · No. 113 · Prompt

Create a premium technical exploded-view illustration of a fictional mechanical wristwatch called the Meridian 8, centered on a dark slate background with fine blueprint grid accents. Show the watch components separated vertically with precise spacing: sapphire crystal, dial, hands, chapter ring, movement plates, escapement, balance wheel, mainspring barrel, case, crown, and leather strap sections. Use realistic brushed steel, brass, ruby jewel accents, and deep navy dial details. Add crisp callouts and labels with the in-image text "Meridian 8", "Exploded Assembly", "42 mm Case", "25 Jewels", and "Power Reserve 72 h". Include numbered callouts "01" through "10" with short labels like "Balance Wheel", "Mainspring Barrel", and "Sapphire Crystal". The result should be highly detailed, technically believable, sharply rendered, and suitable for an industrial design plate with clean hierarchy, exact labeling, and refined material realism.

B uses the reference from No. 54, keeping the overall street layout while reworking the interiors, floors and lighting. Reference attribution: EvoLinkAI · Source.

B · No. 54 · Edit prompt

Use the reference image as the layout anchor for a richly detailed isometric two-block cafe district at blue hour. Keep the street footprint, corner cafe, neighboring bookstore, bakery and fountain plaza recognizable. Transform it into a three-storey architectural cutaway diorama with coherent 30-degree isometric geometry.

Open the front-facing walls to reveal the cafe espresso bar and upstairs jazz lounge; bookshelves, reading nooks and a spiral staircase in the bookstore; pastry cases and a working oven in the bakery. Add a rooftop glass greenhouse, tiny terraces, copper plumbing, tiled stairs, balconies, hanging plants and warm lights visible through rain-speckled windows. At street level show wet cobbles, bicycles, the coffee cart, varied miniature pedestrians and reflections around the fountain. Every floor, doorway and staircase should connect plausibly.

Use warm amber interiors against deep teal evening shadows, tactile brick, glazed tiles, glass and brushed brass. Preserve crisp detail throughout the scene, with a clean dark navy background and room around the floating diorama. Give the scene depth through cutaway rooms and layered architecture. Use restrained, readable storefront lettering: "NIGHT OWL CAFE", "OPEN BOOKS", and "DAWN BAKERY". Keep the composition square and visually balanced.

See the sample record for settings, inputs and review notes.

🖼️ From reference image to result

@LunarXuan contributed Get Prompt from Image. Give a vision-capable agent a reference image, extract a prompt, then pass it to gpt-image or another generator. Below are the contributor's reference and generated result.

Attach an image and invoke the skill with a slash command, $get-prompt-from-image, or plain language:

/get-prompt-from-image
Extract a reusable English positive prompt and a targeted negative prompt from this image, then recreate it with gpt-image.

Positive Prompt

A highly polished semi-realistic Japanese narrative illustration rendered in a painterly digital style, using varied brush widths, a combination of hard edges and soft transitions, restrained contour lines, and controlled surface texture. The image should feel like a cold cinematic game-concept artwork. Use a wide 16:9 composition with strong depth in a snowy urban alley, where the snow-covered road narrows toward a distant vanishing point near the center. Place a large fluffy dark blue-gray wolfdog in the left foreground, shown in side profile facing right with its head raised, interacting with a hooded young woman kneeling near the center-right. She crouches in the snow facing left, gently touching the wolfdog’s muzzle or forehead with one gloved hand while the other rests near her knee for balance, creating a restrained and intimate gesture. She wears an oversized pale-gray winter hooded jacket with pointed ear-like details on top, dark gray panels, pockets, straps, and small muted red-orange accents, over black clothing, fitted black pants, and heavy dark boots. Short black or deep-brown hair falls from beneath the hood; her face is partly shadowed as she looks down at the wolfdog with a quiet, tired, yet gentle expression. Render the wolfdog’s fur with layered directional brushstrokes, making the back, neck, and tail thick and voluminous, with cool blue-gray shadows, pale highlights, and a subtle rim light along the silhouette. On the left, include metal fencing, utility boxes, and dense dark shrubs; in the distance, show tall urban buildings, street lamps, utility poles, and a blue-gray sky. On the right, include dark building facades, windows, snow-covered roof edges, evergreen branches, and foreground cardboard boxes and industrial clutter. Any environmental labels should remain blurred graphic marks with no readable text. Let the main light enter from the distant upper-left side of the alley, combining cold blue ambient shadows with warm golden reflections in the distance. Add subtle rim light to the snow, the woman, and the wolfdog, with medium-high contrast and warm orange clothing details acting as focal accents. Snow, slush, and shallow puddles in the foreground should show damp reflections. Use atmospheric perspective to soften distant buildings while keeping the woman and wolfdog clear. Establish depth through foreground, middle ground, background, occlusion, and perspective lines rather than strong blur. The mood is loneliness, trust, and a brief moment of tenderness in a frozen city. Preserve rough painterly strokes, cool-warm contrast, cinematic composition, and refined post-processing. Clearly remain a 2D semi-realistic painterly illustration, not photography, pure flat vector art, or 3D rendering.

Negative Prompt

photorealistic, 3D render, flat vector style, pure cel shading, watercolor bleed, oil painting impasto, chibi proportions, deformed anatomy, malformed hands, extra limbs, oversized wolf, sunny summer weather, cluttered composition, readable text, watermark

🚀 Quick start

Task	Use
Generate or edit an image	gpt-image
Extract a prompt from a reference	get-prompt-from-image
Work in a terminal	CLI examples below, with the selected --model

🎛️ Image 2.6 by default

Model	Best starting point
gpt-image-2.6-flare	Default: fast generation and editing
gpt-image-2.6-sunburst	Detailed visual work and precise editing
gpt-image-2.5-flare / gpt-image-2.5-sunburst	Continue existing 2.5 workflows
gpt-image-2	Retain Image 2 workflows

The CLI and Skill use gpt-image-2.6-flare unless you choose another model. The Skill passes that choice explicitly with --model and does not ask you to confirm the default. To keep an older command on Image 2, add --model gpt-image-2.

The CLI generates and edits PNG, JPEG, WebP and GIF images. It defaults to high quality; use low for drafts. Models 2.5 and 2.6 also accept xhigh and max. SVG, complexity and markup controls are not exposed yet. Use gpt-image animate to assemble a sprite sheet locally. See the run record for the 2.6 Flare generations and reference edits we tested.

For example, generate a transparent icon with 2.6 Flare:

gpt-image --model gpt-image-2.6-flare \
  -p "An original flat leaf icon, centered with generous padding, transparent background" \
  --quality medium --background transparent --format png -f leaf.png

The model notes list compatibility and tested settings. The GPT Image 2.5 prompt templates are available to try, but we have not validated their outputs locally. Gallery images retain their original model and source credits.

After installing, copy a gallery prompt into gpt-image --model <CHOSEN_MODEL> -p "…", or ask your agent to use it: "Generate the Boston Spring poster from the skill gallery."

Text → image

gpt-image --model gpt-image-2.6-flare -p "a photorealistic convenience store at 10pm" --size 1k --quality high -f store.png

Under the hood: POST /v1/images/generations with the explicitly selected model.

Text + reference image → image (edit)

# Single-reference edit / restyle
gpt-image --model gpt-image-2.6-sunburst -p "Make it a winter evening with heavy snowfall" \
  -i chess.png --quality high -f chess-winter.png

# Multi-reference edit: the edits endpoint accepts multiple input images
gpt-image --model gpt-image-2.6-sunburst -p "Place the dog from image 2 next to the woman in image 1. Match the same lighting, composition, and background. Do not change anything else." \
  -i woman.png -i dog.png --size portrait --quality medium -f woman-with-dog.png

# Mask-based inpaint: opaque = keep, transparent = regenerate
gpt-image --model gpt-image-2.6-sunburst -p "replace sky with aurora" \
  -i photo.jpg -m sky_mask.png -f aurora.png

Under the hood: POST /v1/images/edits (multipart form). GPT Image 2, 2.5 and 2.6 models use this endpoint, with multiple -i inputs and an optional -m mask. Read results from data[].b64_json and omit response_format. See model compatibility.

📥 Install

Send this to your coding agent for a first install or an update:

Install or update GPT Image Skill for me: https://raw.githubusercontent.com/wuyoscar/GPT-Image2-Skill/main/docs/install.md

The agent checks your current installation, uses the matching updater, and verifies the result. You can also follow the commands below yourself.

Choose either skill or install both: gpt-image generates and edits images, while get-prompt-from-image extracts prompts from reference images.

Check for an existing skill or CLI before installing. Preserve existing skill folders and API-key files. Use your runtime's skill list/status command when available, and ask before installing into a global or shared directory.

command -v gpt-image || true
command -v uv >/dev/null && uv tool list | grep -E '^gpt-image-cli([[:space:]]|$)' || true
test -n "${OPENAI_API_KEY:-}" && echo "OPENAI_API_KEY is already set (value hidden)"

/plugin marketplace add wuyoscar/gpt_image_2_skill
/plugin install gpt-image@wuyoscar-skills

For a managed install that records the remote source and supports skills update:

npx --yes skills@latest add wuyoscar/gpt_image_2_skill --skill gpt-image get-prompt-from-image --agent codex --copy

Recent Codex versions can also install the native plugin; see the full steps. The skill-installer and manual alternatives follow.

Codex ships with built-in skill helpers such as $skill-installer and $skill-creator. Open Codex and invoke the built-in installer with the GitHub skill-folder URL for each skill you want:

# gpt-image
$skill-installer
Install this skill from GitHub:
https://github.com/wuyoscar/gpt_image_2_skill/tree/main/skills/gpt-image

# get-prompt-from-image
$skill-installer
Install this skill from GitHub:
https://github.com/wuyoscar/gpt_image_2_skill/tree/main/skills/get-prompt-from-image

The installer downloads each GitHub folder and places it under your Codex skills directory, usually:

~/.codex/skills/gpt-image
~/.codex/skills/get-prompt-from-image

Restart Codex after installation so the new skills are loaded.

If you prefer to install both manually, copy their skill folders into Codex's skills directory:

git clone https://github.com/wuyoscar/gpt_image_2_skill.git
cd gpt_image_2_skill

mkdir -p "${CODEX_HOME:-$HOME/.codex}/skills"
for skill in gpt-image get-prompt-from-image; do
  test -e "${CODEX_HOME:-$HOME/.codex}/skills/$skill" && echo "$skill already exists; stop before overwriting" && exit 1
  cp -R "skills/$skill" "${CODEX_HOME:-$HOME/.codex}/skills/"
done

For runtimes supported by the cross-agent skills installer, select either skill or install both together from GitHub:

# Change --agent to claude-code, codex, opencode, openclaw, or another supported runtime.
npx --yes skills@latest add wuyoscar/gpt_image_2_skill \
  --skill gpt-image \
  --skill get-prompt-from-image \
  --agent codex --copy

These examples intentionally avoid --global. Add --global only when you explicitly want this skill installed into that runtime's global/shared skills directory.

Other runtimes can use the manual agent-skill installation below.

Set AGENT_SKILLS_DIR to the skills directory used by your agent runtime, then symlink one or both skill folders into it.

git clone https://github.com/wuyoscar/gpt_image_2_skill.git
cd gpt_image_2_skill

# Choose the skill directory for your runtime.
# Examples:
#   Codex:      ~/.codex/skills
#   Claude Code / OpenClaw / Hermes Agent / other runtimes: use that runtime's documented skills directory.
export AGENT_SKILLS_DIR="/path/to/your/agent/skills"

mkdir -p "$AGENT_SKILLS_DIR"
for skill in gpt-image get-prompt-from-image; do
  test -e "$AGENT_SKILLS_DIR/$skill" && echo "$skill already exists; stop before overwriting" && exit 1
  ln -s "$PWD/skills/$skill" "$AGENT_SKILLS_DIR/$skill"
done

uvx --from git+https://github.com/wuyoscar/gpt_image_2_skill gpt-image --model gpt-image-2.6-flare -p "a cat astronaut"

# or install to PATH if not already installed
command -v gpt-image >/dev/null || uv tool install git+https://github.com/wuyoscar/gpt_image_2_skill
gpt-image --model gpt-image-2.6-flare -p "a cat astronaut"

Use the updater that matches your installation, then start a new agent session.

Installed as	Update command
Claude Code plugin	claude plugin update gpt-image@wuyoscar-skills
Project skills for Codex / other agents via skills	npx --yes skills@latest update gpt-image get-prompt-from-image --project
Global skills via skills	Replace --project above with --global
Codex native plugin	codex plugin marketplace upgrade wuyoscar-skills && codex plugin add gpt-image@wuyoscar-skills
Standalone Python CLI	uv tool upgrade gpt-image-cli
Skills symlinked to a Git checkout	git -C /path/to/gpt_image_2_skill pull --ff-only

Claude Code can auto-update wuyoscar-skills from the Marketplaces tab in /plugin. After refreshing a Codex native plugin, check its installed version in /plugins in a new session. Check the CLI with gpt-image --version.

Skills, plugins and a standalone CLI are separate installs. Updating one may not update another, and hand-copied folders may not be managed by skills. See install and update for scope, setup and official sources.

API keys: The CLI reads OPENAI_API_KEY from the process environment, then .env, then ~/.env. It keeps any value already set. To avoid using a local key, check all three places; unset OPENAI_API_KEY clears only the process variable. Codex users can choose the built-in image tool for platform-managed generation.

🛠️ CLI reference

Parameters

Flag	Values	Default	Applies to	Notes
-p, --prompt	str	required	both	Full prompt text.
--model	exact model ID above	gpt-image-2.6-flare	both	The Skill passes the resolved model explicitly, including the default.
-f, --file	path	./fig/YYYY-MM-DD-HH-MM-SS-<slug>.png	both	Explicit output path.
-i, --image	path (repeatable)	omitted	edits	Presence routes through /v1/images/edits.
-m, --mask	path (PNG, alpha)	omitted	edits	Opaque = preserved, transparent = regenerated. Requires -i.
--input-fidelity	low · high	omitted	edits	Image 2 cannot set it, so the CLI omits it. For 2.5/2.6, explicit values pass through but behavior remains unverified; see model notes.
--size	1k · 2k · 4k · portrait · landscape · square · wide · tall · literal 1024x1024 etc.	1024x1024	both	Image 2/2.5: 16-px multiples, max edge 3840, 3:1 cap, 655k-8.3M pixels. 2.6 custom sizes are validated by the API; see model notes.
--quality	auto · low · medium · high · xhigh · max	high	both	xhigh / max require a 2.5 or 2.6 model; use low drafts to assess quality, latency, and cost first.
-n, --n	1-10	1	both	Batch generation. n>1 suffixes filenames _0, _1, …
--background	auto · opaque · transparent	API default	both	Transparency requires PNG/WebP; Image 2 transparency remains in preview.
--moderation	auto · low	low	generations	low is the default here for broader prompt exploration; switch to auto if you want the stricter API-side default.
--format	png · jpeg · webp · gif	png	both	Response encoding.
--frames / --frame-duration	2-16 / 20-2000 ms	API: 9 / 200	GIF	Frame count and delay; GIF size defaults to auto.
--compression	0-100	omitted	both	JPEG/WebP only.

Budget / quality guide

Use --quality to manage the generation budget.

• low = cheap draft / collect / many variants
• medium = normal exploration / style probing
• high = final posters, Chinese text, diagrams, paper figures, banners

If you are generating dozens of candidates, start at low and only rerun finalists at high. On 2.5, compare xhigh / max for detail-critical assets. Agree on additional outputs and costs before comparing settings.

From gallery prompt → CLI / SDK

Each entry includes a prompt and metadata ("size" · "quality" · source). Use the shared CLI/SDK pattern below. This example uses "portrait" and "high":

# CLI
gpt-image --model gpt-image-2.6-flare -p "<PROMPT FROM ENTRY>" --size portrait --quality high -f out.png

# OpenAI SDK — `size` is the literal pixels; the CLI shortcut maps to `1024x1536` for portrait
from openai import OpenAI
client = OpenAI()
result = client.images.generate(
    model="gpt-image-2.6-flare",
    prompt="<PROMPT FROM ENTRY>",
    size="1024x1536",
    quality="high",
)

For reference-based edits, add -i ref.png (repeatable) and optionally -m mask.png on the CLI, or call client.images.edit(...) with image=[open(p, "rb") for p in refs]. Shared output options stay the same. --moderation applies to generation; edit calls omit it.

Exit codes: 0 success · 1 API/refusal error (full response body echoed to stderr) · 2 bad args or missing OPENAI_API_KEY.

📚 Guides and reading

[!CAUTION] Use generated images as references, workflow sketches or style targets for research figures. Prepare and verify the final figure separately before including it in a paper; directly publishing a raw generation can mislead readers.

📖 Prompting fundamentals

These notes draw on the Image 2 Cookbook. The 2.5 reference index links to official guidance by task, with migration notes in a separate file. For Image 2, start with a gallery example. For 2.5, a clear brief and a selected model are enough to begin; consult a task guide when you need help. Check model notes for API parameters. Community templates keep their own source credits.

1. Structure, then goal. Use a consistent order: background/scene → subject → key details → constraints, and state the intended use (ad, UI mock, infographic) so the model picks the right mode and polish level.
2. Any format works; consistency matters more. Minimal prompts, descriptive paragraphs, JSON-style structures, instruction-style prompts, and tag-based prompts all work. For production, choose a format you can easily read and maintain.
3. Specificity + quality cues. Be concrete about materials, shapes, textures, and medium (photo, watercolor, 3D render). Add targeted levers only when they matter: film grain, textured brushstrokes, macro detail. For photorealism, say "photorealistic" directly; "real photograph", "taken on a real camera", and "iPhone photo" also help.
4. Quote the exact text. Put required slogans, prices and labels in straight quotes. Copy the wording verbatim.
5. Choose aspect ratio early. Decide 1:1 / 3:4 / 4:3 / 9:16 / 16:9 / 3:1 before writing the prompt. Keep the prompt's aspect ratio consistent with --size.
6. One hero, supporting cast. Complex scenes work best when one subject is clearly primary and the rest is framed as supporting detail.
7. Evaluate quality against the actual task. For 2.5 text, diagrams and multi-panel layouts, compare approved settings against legibility and layout requirements; high remains the CLI default. Judge each setting by the resulting text and layout.

What you need	Reference
Gallery index	gallery.md: choose a category, then read its prompts.
Concrete examples	Load a relevant gallery-*.md file, such as products, UI or paper figures.
Prompt craft	craft.md: text, composition, data relationships and edit constraints.
GPT Image 2.6	Model defaults, API compatibility and validation status.
GPT Image 2.5	Reference index: separate generation, layout/text, editing and migration notes.
Historical Image 2 guidance	OpenAI Cookbook, with original source and license. Use current model notes and the user's settings for API calls.

📚 GPT Image 2.5: understand the workflow

English-language official sources and established publications first, developer guides next; Chinese-language reading supplements them. Read these links when useful. OpenAI's documentation is the reference for API details.

🌐 English-first · 8 sources

Source	Read	What to take away
🏛️ 1 · OpenAI	Official launch	Reference-led editing, ChatGPT's new creative tools, and the Flare/Sunburst lineup.
📖 2 · OpenAI Docs	Image prompting: GPT Image 2.5	Official model selection, reference roles, focused edits and output checks.
📰 3 · Axios	Exclusive hands-on	A reporter's experience with likeness preservation and visually targeted edits.
🧪 4 · TechRadar	24-hour hands-on	Everyday editing controls, sketch-based creation and task templates.
✏️ 5 · The Verge	Sketch walkthrough †	Sketch coverage; full-text retrieval pending.
🛠️ 6 · Apidog	API walkthrough · Model comparison	Third-party implementation and migration perspectives; verify API details against official docs.
⚖️ 7 · OrcaRouter	Flare vs Sunburst for builders ‡	Model-selection advice by workload; performance claims need independent checking.
🧩 8 · FindSkill	Sketch step by step	Use a drawing for layout and text for additional instructions.

🗂️ Chinese-language supplements · 2 sources

Source	Read	What to take away
🧪 9 · 人人都是产品经理	Reader-supplied article †	Supplied as a 2 vs 2.5 hands-on comparison; content remains unverified.
🗞️ 10 · 軟體玩家	Feature roundup by 阿正老師	A feature roundup based on official announcements and media reports.

Source-access status, checked 2026-09-09: † Full text pending for The Verge and 人人都是产品经理. ‡ Indexed content only for OrcaRouter.

Using ChatGPT sketches with the CLI: Sketch, Templates and Comments operate inside ChatGPT. Export a sketch as an image and attach it with -i. For task-specific local notes, choose one slice from the optional 2.5 reference index.

🎨 Gallery

Open a category to view its images and prompts. The complete 171-prompt collection is indexed in gallery.md, with full entries in the linked gallery-*.md files.

Source labels. Curated means a repo-curated or substantially reworked prompt/image; outside-source items keep visible author/source links.

---

🎌 Anime & Manga

Anime fashion portrait triptych

📝 Prompts for all three panels

Prompt A: Elegant cafe fashion

Create a tasteful portrait-oriented anime fashion illustration of an adult woman, age 24, with a cute playful expression, looking at the camera in a cozy European cafe at golden hour. She wears a cream blouse, charcoal pleated skirt, tailored cropped jacket, sheer black stockings, loafers, and a small ribbon hair clip; she is seated sideways at a small marble table with latte art, a sketchbook, and warm window light. Composition: three-quarter fashion portrait, elegant legs visible but relaxed and non-explicit, wholesome editorial mood, no nudity, no lingerie, no school uniform, no explicit pose, adult character only. Use polished modern anime rendering, crisp line art, luminous eyes, soft cel shading, subtle fabric texture, gentle blush, background bokeh, and a refined magazine-cover color palette.

Prompt B: Neon arcade fashion

Create a portrait-oriented anime fashion illustration of an adult woman, age 25, in a neon arcade district at night. She has a cute confident smile and looks directly at the viewer while standing beside glowing claw machines and retro game cabinets. Outfit: black turtleneck, red satin bomber jacket, high-waisted skirt, patterned dark stockings, platform shoes, small crossbody bag, star earrings. Composition: full-body fashion portrait with strong silhouette, neon reflections on wet pavement, vending machines, sticker-covered walls, colorful signage, and cinematic rim light. Keep the pose playful but non-explicit, no nudity, no lingerie, no fetish framing, adult character only. Use high-end anime key visual rendering, crisp line art, saturated magenta-cyan lighting, clean readable background details, and glossy cyber-pop atmosphere.

Prompt C: Roadside mirror selfie

Create a portrait-oriented anime fashion illustration of an adult woman, age 24, taking a playful roadside mirror selfie in the reflection of a parked scooter mirror on a quiet Tokyo side street. She looks into the mirror with a bright mischievous smile, one hand making a small peace sign near her cheek, the other holding a phone with a cute sticker case. Outfit: soft ivory knit cardigan, navy pleated skirt, sheer black stockings, loafers, small shoulder bag, ribbon hair clip, tasteful everyday street fashion. Composition: the mirror reflection is the main frame, with blurred street signs, vending machine glow, crosswalk stripes, and spring evening light around the mirror edge. Keep the pose cute, stylish, and non-explicit; no nudity, no lingerie, no fetish framing, adult character only. Use polished modern anime rendering, crisp line art, luminous eyes, soft cel shading, warm reflections, natural street-photo energy, and a charming slice-of-life mood.

---

MAPPA-style anime action still (Jujutsu-Kaisen aesthetic)

📝 Prompt

An anime action still in the visual style of MAPPA's Jujutsu Kaisen (2020 TV anime). Landscape 16:9.

A silver-white-haired young man in a dark navy school-uniform jacket, a blue blindfold across his eyes, in a mid-fight stance — one palm extended outward releasing a swirling dense-blue energy sphere with lightning-like crackles around its edge. Opposite him, a demonic shadow creature made of liquid black mass with multiple eyes lunges from the right.

Backdrop: ruined urban street at dusk, shattered asphalt, cracked neon kanji sign "呪術" in split red LED, destroyed vehicles, rubble suspended mid-air by the shockwave, rain particles caught mid-flight.

Art direction: MAPPA-style digital 2D animation — heavy cel shading, crisp line-art, rim-light on both figures, motion-blur streaks around the energy sphere. Palette of deep navy, electric cyan, crimson splashes. Kinetic-impact composition in the tradition of JJK's Shibuya arc.

---

Shōnen battle key-visual (Naruto-Shippuden aesthetic)

📝 Prompt

A shōnen anime battle key-visual in the visual style of Studio Pierrot's Naruto Shippuden. Landscape 16:9.

Two ninja figures clash mid-air at the exact instant their signature jutsu collide — a glowing blue spiral of swirling chakra on the left fighter's right palm, a crackling white lightning blade on the right fighter's right palm. The collision point sends a circular shockwave outward.

Both fighters wear hitai-ate forehead protectors, jounin-style tactical vests with scroll pouches, ninja sandals. Left: spiky blond hair, whisker cheek marks, focused snarl, blue eyes. Right: dark hair, one red sharingan-like eye with three tomoe, calm expression.

Backdrop: nighttime valley, cracked earth, giant uprooted trees mid-crash, moonlit clouds parting, sakura petals caught in the shockwave.

Art direction: Studio Pierrot Naruto-Shippuden aesthetic — dynamic perspective, strong speed lines radiating from the collision, anime-action key-frame quality, digital 2D cel shading, saturated but not neon, visible genga-quality line-art, dramatic backlight.

---

Manga / anime 1×2 panel