interface:
  display_name: "Image Gen"
  short_description: "Generate or edit images for websites, games, and more"
  icon_small: "./assets/imagegen-small.svg"
  icon_large: "./assets/imagegen.png"
  default_prompt: "Generate or edit the visual assets for this task with the built-in `image_gen` tool by default. First confirm that the task actually calls for a raster image; if the project already has SVG/vector/code-native assets and the user wants to extend or match those, do not use this skill. If the task includes reference images, treat them as references unless the user clearly wants an existing image modified. For multi-asset requests, loop built-in calls rather than treating batch as a separate top-level mode. Only use the fallback CLI if the user explicitly asks for it, and keep CLI-only controls such as `generate-batch`, `quality`, `input_fidelity`, masks, and output paths on that fallback path."

# CLI reference (`scripts/image_gen.py`)

This file is for the fallback CLI mode only. Read it only after the user explicitly asks to use `scripts/image_gen.py` instead of the built-in `image_gen` tool.

`generate-batch` is a CLI subcommand in this fallback path. It is not a top-level mode of the skill.

## What this CLI does
- `generate`: generate a new image from a prompt
- `edit`: edit one or more existing images
- `generate-batch`: run many generation jobs from a JSONL file

Real API calls require **network access** + `OPENAI_API_KEY`. `--dry-run` does not.

## Quick start (works from any repo)
Set a stable path to the skill CLI (default `CODEX_HOME` is `~/.codex`):

```
export CODEX_HOME="${CODEX_HOME:-$HOME/.codex}"
export IMAGE_GEN="$CODEX_HOME/skills/.system/imagegen/scripts/image_gen.py"
```

Install dependencies into that environment with its package manager. In uv-managed environments, `uv pip install ...` remains the preferred path.

## Quick start

Dry-run (no API call; no network required; does not require the `openai` package):

```bash
python "$IMAGE_GEN" generate \
  --prompt "Test" \
  --out output/imagegen/test.png \
  --dry-run
```

Notes:
- One-off dry-runs print the API payload and the computed output path(s).
- Repo-local finals should live under `output/imagegen/`.

Generate (requires `OPENAI_API_KEY` + network):

```bash
python "$IMAGE_GEN" generate \
  --prompt "A cozy alpine cabin at dawn" \
  --size 1024x1024 \
  --out output/imagegen/alpine-cabin.png
```

Edit:

```bash
python "$IMAGE_GEN" edit \
  --image input.png \
  --prompt "Replace only the background with a warm sunset" \
  --out output/imagegen/sunset-edit.png
```

## Guardrails
- Use the bundled CLI directly (`python "$IMAGE_GEN" ...`) after activating the correct environment.
- Do **not** create one-off runners (for example `gen_images.py`) unless the user explicitly asks for a custom wrapper.
- **Never modify** `scripts/image_gen.py`. If something is missing, ask the user before doing anything else.

## Defaults
- Model: `gpt-image-1.5`
- Supported model family for this CLI: GPT Image models (`gpt-image-*`)
- Size: `1024x1024`
- Quality: `auto`
- Output format: `png`
- Default one-off output path: `output/imagegen/output.png`
- Background: unspecified unless `--background` is set

## Quality, input fidelity, and masks (CLI fallback only)
These are explicit CLI controls. They are not built-in `image_gen` tool arguments.

- `--quality` works for `generate`, `edit`, and `generate-batch`: `low|medium|high|auto`
- `--input-fidelity` is **edit-only** and validated as `low|high`
- `--mask` is **edit-only**

Example:

```bash
python "$IMAGE_GEN" edit \
  --image input.png \
  --prompt "Change only the background" \
  --quality high \
  --input-fidelity high \
  --out output/imagegen/background-edit.png
```

Mask notes:
- For multi-image edits, pass repeated `--image` flags. Their order is meaningful, so describe each image by index and role in the prompt.
- The CLI accepts a single `--mask`.
- Use a PNG mask when possible; the script treats mask handling as best-effort and does not perform full preflight validation beyond file checks/warnings.
- In the edit prompt, repeat invariants (`change only the background; keep the subject unchanged`) to reduce drift.

## Output handling
- Use `tmp/imagegen/` for temporary JSONL inputs or scratch files.
- Use `output/imagegen/` for final outputs.
- Reruns fail if a target file already exists unless you pass `--force`.
- `--out-dir` changes one-off naming to `image_1.<ext>`, `image_2.<ext>`, and so on.
- Downscaled copies use the default suffix `-web` unless you override it.

## Common recipes

Generate with augmentation fields:

```bash
python "$IMAGE_GEN" generate \
  --prompt "A minimal hero image of a ceramic coffee mug" \
  --use-case "product-mockup" \
  --style "clean product photography" \
  --composition "wide product shot with usable negative space for page copy" \
  --constraints "no logos, no text" \
  --out output/imagegen/mug-hero.png
```

Generate + also write a downscaled copy for fast web loading:

```bash
python "$IMAGE_GEN" generate \
  --prompt "A cozy alpine cabin at dawn" \
  --size 1024x1024 \
  --downscale-max-dim 1024 \
  --out output/imagegen/alpine-cabin.png
```

Generate multiple prompts concurrently (async batch):

```bash
mkdir -p tmp/imagegen output/imagegen/batch
cat > tmp/imagegen/prompts.jsonl << 'EOF'
{"prompt":"Cavernous hangar interior with a compact shuttle parked near the center","use_case":"stylized-concept","composition":"wide-angle, low-angle","lighting":"volumetric light rays through drifting fog","constraints":"no logos or trademarks; no watermark","size":"1536x1024"}
{"prompt":"Gray wolf in profile in a snowy forest","use_case":"photorealistic-natural","composition":"eye-level","constraints":"no logos or trademarks; no watermark","size":"1024x1024"}
EOF

python "$IMAGE_GEN" generate-batch \
  --input tmp/imagegen/prompts.jsonl \
  --out-dir output/imagegen/batch \
  --concurrency 5

rm -f tmp/imagegen/prompts.jsonl
```

Notes:
- `generate-batch` requires `--out-dir`.
- generate-batch requires --out-dir.
- Use `--concurrency` to control parallelism (default `5`).
- Per-job overrides are supported in JSONL (for example `size`, `quality`, `background`, `output_format`, `output_compression`, `moderation`, `n`, `model`, `out`, and prompt-augmentation fields).
- `--n` generates multiple variants for a single prompt; `generate-batch` is for many different prompts.
- In batch mode, per-job `out` is treated as a filename under `--out-dir`.

## CLI notes
- Supported sizes: `1024x1024`, `1536x1024`, `1024x1536`, or `auto`.
- Transparent backgrounds require `output_format` to be `png` or `webp`.
- `--prompt-file`, `--output-compression`, `--moderation`, `--max-attempts`, `--fail-fast`, `--force`, and `--no-augment` are supported.
- This CLI is intended for GPT Image models. Do not assume older non-GPT image-model behavior applies here.

## See also
- API parameter quick reference for fallback CLI mode: `references/image-api.md`
- Prompt examples shared across both top-level modes: `references/sample-prompts.md`
- Network/sandbox notes for fallback CLI mode: `references/codex-network.md`

# Codex network approvals / sandbox notes

This file is for the fallback CLI mode only. Read it only after the user explicitly asks to use `scripts/image_gen.py`.

This guidance is intentionally isolated from `SKILL.md` because it can vary by environment and may become stale. Prefer the defaults in your environment when in doubt.

## Why am I asked to approve image generation calls?
The fallback CLI uses the OpenAI Image API, so it needs outbound network access. In many Codex setups, network access is disabled by default and/or the approval policy requires confirmation before networked commands run.

## Important note about approvals vs network
- `--ask-for-approval never` suppresses approval prompts.
- It does **not** by itself enable network access.
- In `workspace-write`, network access still depends on your Codex configuration (for example `[sandbox_workspace_write] network_access = true`).

## How do I reduce repeated approval prompts?
If you trust the repo and want fewer prompts, use a configuration or profile that both:
- enables network for the sandbox mode you plan to use
- sets an approval policy that matches your risk tolerance

Example `~/.codex/config.toml` pattern:

```toml
approval_policy = "on-request"
sandbox_mode = "workspace-write"

[sandbox_workspace_write]
network_access = true
```

If you want quieter automation after network is enabled, you can choose a stricter approval policy, but do that intentionally and with care.

## Safety note
Enabling network and reducing approvals lowers friction, but increases risk if you run untrusted code or work in an untrusted repository.

# Image API quick reference

This file is for the fallback CLI mode only. Use it only after the user explicitly asks to use `scripts/image_gen.py` instead of the built-in `image_gen` tool.

These parameters describe the Image API and bundled CLI fallback surface. Do not assume they are normal arguments on the built-in `image_gen` tool.

## Scope
- This fallback CLI is intended for GPT Image models (`gpt-image-1.5`, `gpt-image-1`, and `gpt-image-1-mini`).
- The built-in `image_gen` tool and the fallback CLI do not expose the same controls.

## Endpoints
- Generate: `POST /v1/images/generations` (`client.images.generate(...)`)
- Edit: `POST /v1/images/edits` (`client.images.edit(...)`)

## Core parameters for GPT Image models
- `prompt`: text prompt
- `model`: image model
- `n`: number of images (1-10)
- `size`: `1024x1024`, `1536x1024`, `1024x1536`, or `auto`
- `quality`: `low`, `medium`, `high`, or `auto`
- `background`: output transparency behavior (`transparent`, `opaque`, or `auto`) for generated output; this is not the same thing as the prompt's visual scene/backdrop
- `output_format`: `png` (default), `jpeg`, `webp`
- `output_compression`: 0-100 (jpeg/webp only)
- `moderation`: `auto` (default) or `low`

## Edit-specific parameters
- `image`: one or more input images. For GPT Image models, you can provide up to 16 images.
- `mask`: optional mask image
- `input_fidelity`: `low` (default) or `high`

Model-specific note for `input_fidelity`:
- `gpt-image-1` and `gpt-image-1-mini` preserve all input images, but the first image gets richer textures and finer details.
- `gpt-image-1.5` preserves the first 5 input images with higher fidelity.

## Output
- `data[]` list with `b64_json` per image
- The bundled `scripts/image_gen.py` CLI decodes `b64_json` and writes output files for you.

## Limits and notes
- Input images and masks must be under 50MB.
- Use the edits endpoint when the user requests changes to an existing image.
- Masking is prompt-guided; exact shapes are not guaranteed.
- Large sizes and high quality increase latency and cost.
- High `input_fidelity` can materially increase input token usage.
- If a request fails because a specific option is unsupported by the selected GPT Image model, retry manually without that option.

## Important boundary
- `quality`, `input_fidelity`, explicit masks, `background`, `output_format`, and related parameters are fallback-only execution controls.
- Do not assume they are built-in `image_gen` tool arguments.

# Prompting best practices

These prompting principles are shared by both top-level modes of the skill:
- built-in `image_gen` tool (default)
- explicit `scripts/image_gen.py` CLI fallback

This file is about prompt structure, specificity, and iteration. Fallback-only execution controls such as `quality`, `input_fidelity`, masks, output format, and output paths live in the fallback docs.

## Contents
- [Structure](#structure)
- [Specificity policy](#specificity-policy)
- [Allowed and disallowed augmentation](#allowed-and-disallowed-augmentation)
- [Composition and layout](#composition-and-layout)
- [Constraints and invariants](#constraints-and-invariants)
- [Text in images](#text-in-images)
- [Input images and references](#input-images-and-references)
- [Iterate deliberately](#iterate-deliberately)
- [Fallback-only execution controls](#fallback-only-execution-controls)
- [Use-case tips](#use-case-tips)
- [Where to find copy/paste recipes](#where-to-find-copypaste-recipes)

## Structure
- Use a consistent order: scene/backdrop -> subject -> key details -> constraints -> output intent.
- Include intended use (ad, UI mock, infographic) to set the level of polish.
- For complex requests, use short labeled lines instead of one long paragraph.

## Specificity policy
- If the user prompt is already specific and detailed, normalize it into a clean spec without adding creative requirements.
- If the prompt is generic, you may add tasteful detail when it materially improves the output.
- Treat examples in `sample-prompts.md` as fully-authored recipes, not as the default amount of augmentation to add to every request.

## Allowed and disallowed augmentation

Allowed augmentation for generic prompts:
- composition and framing cues
- intended-use or polish-level hints
- practical layout guidance
- reasonable scene concreteness that supports the request

Do not add:
- extra characters, props, or objects that are not implied
- brand palettes, slogans, or story beats that are not implied
- arbitrary side-specific placement unless the surrounding layout supports it

## Composition and layout
- Specify framing and viewpoint (close-up, wide, top-down) and placement only when it materially helps.
- Call out negative space if the asset clearly needs room for UI or copy.
- Avoid making left/right layout decisions unless the user or surrounding layout supports them.

## Constraints and invariants
- State what must not change (`keep background unchanged`).
- For edits, say `change only X; keep Y unchanged` and repeat invariants on every iteration to reduce drift.

## Text in images
- Put literal text in quotes or ALL CAPS and specify typography (font style, size, color, placement).
- Spell uncommon words letter-by-letter if accuracy matters.
- For in-image copy, require verbatim rendering and no extra characters.

## Input images and references
- Do not assume that every provided image is an edit target.
- Label each image by index and role (`Image 1: edit target`, `Image 2: style reference`).
- If the user provides images for style, composition, or mood guidance and does not ask to modify them, treat the request as generation with references.
- If the user asks to preserve an existing image while changing specific parts, treat the request as an edit.
- For compositing, describe how the images interact (`place the subject from Image 2 into Image 1`).

## Iterate deliberately
- Start with a clean base prompt, then make small single-change edits.
- Re-specify critical constraints when you iterate.
- Prefer one targeted follow-up at a time over rewriting the whole prompt.

## Fallback-only execution controls
- `quality`, `input_fidelity`, explicit masks, output format, and output paths are fallback-only execution controls.
- Do not assume they are built-in `image_gen` tool arguments.
- If the user explicitly chooses CLI fallback, see `references/cli.md` and `references/image-api.md` for those controls.

## Use-case tips
Generate:
- photorealistic-natural: Prompt as if a real photo is captured in the moment; use photography language (lens, lighting, framing); call for real texture; avoid over-stylized polish unless requested.
- product-mockup: Describe the product/packaging and materials; ensure clean silhouette and label clarity; if in-image text is needed, require verbatim rendering and specify typography.
- ui-mockup: Describe the target fidelity first (shippable mockup or low-fi wireframe), then focus on layout, hierarchy, and practical UI elements; avoid concept-art language.
- infographic-diagram: Define the audience and layout flow; label parts explicitly; require verbatim text.
- logo-brand: Keep it simple and scalable; ask for a strong silhouette and balanced negative space; avoid decorative flourishes unless requested.
- illustration-story: Define panels or scene beats; keep each action concrete.
- stylized-concept: Specify style cues, material finish, and rendering approach (3D, painterly, clay) without inventing new story elements.
- historical-scene: State the location/date and required period accuracy; constrain clothing, props, and environment to match the era.

Edit:
- text-localization: Change only the text; preserve layout, typography, spacing, and hierarchy; no extra words or reflow unless needed.
- identity-preserve: Lock identity (face, body, pose, hair, expression); change only the specified elements; match lighting and shadows.
- precise-object-edit: Specify exactly what to remove/replace; preserve surrounding texture and lighting; keep everything else unchanged.
- lighting-weather: Change only environmental conditions (light, shadows, atmosphere, precipitation); keep geometry, framing, and subject identity.
- background-extraction: Request a clean cutout; crisp silhouette; no halos; preserve label text exactly; no restyling.
- style-transfer: Specify style cues to preserve (palette, texture, brushwork) and what must change; add `no extra elements` to prevent drift.
- compositing: Reference inputs by index; specify what moves where; match lighting, perspective, and scale; keep the base framing unchanged.
- sketch-to-render: Preserve layout, proportions, and perspective; choose materials and lighting that support the supplied sketch without adding new elements.

## Where to find copy/paste recipes
For copy/paste prompt specs (examples only), see `references/sample-prompts.md`. This file focuses on principles, specificity, and iteration patterns.

# Sample prompts (copy/paste)

These prompt recipes are shared across both top-level modes of the skill:
- built-in `image_gen` tool (default)
- explicit `scripts/image_gen.py` CLI fallback

Use these as starting points. They are intentionally complete prompt recipes, not the default amount of augmentation to add to every user request.

When adapting a user's prompt:
- keep user-provided requirements
- only add detail according to the specificity policy in `SKILL.md`
- do not treat every example below as permission to invent extra story elements

The labeled lines are prompt scaffolding, not a closed schema. `Asset type` and `Input images` are prompt-only scaffolding; the CLI does not expose them as dedicated flags.

Execution details such as explicit CLI flags, `quality`, `input_fidelity`, masks, output formats, and local output paths depend on mode. Use the built-in tool by default; only apply CLI-specific controls after the user explicitly opts into fallback mode.

For prompting principles (structure, specificity, invariants, iteration), see `references/prompting.md`.

## Generate

### photorealistic-natural
```
Use case: photorealistic-natural
Primary request: candid photo of an elderly sailor on a small fishing boat adjusting a net
Scene/backdrop: coastal water with soft haze
Subject: weathered skin with wrinkles and sun texture
Style/medium: photorealistic candid photo
Composition/framing: medium close-up, eye-level
Lighting/mood: soft coastal daylight, shallow depth of field, subtle film grain
Materials/textures: real skin texture, worn fabric, salt-worn wood
Constraints: natural color balance; no heavy retouching; no glamorization; no watermark
Avoid: studio polish; staged look
```

### product-mockup
```
Use case: product-mockup
Primary request: premium product photo of a matte black shampoo bottle with a minimal label
Scene/backdrop: clean studio gradient from light gray to white
Subject: single bottle centered with subtle reflection
Style/medium: premium product photography
Composition/framing: centered, slight three-quarter angle, generous padding
Lighting/mood: softbox lighting, clean highlights, controlled shadows
Materials/textures: matte plastic, crisp label printing
Constraints: no logos or trademarks; no watermark
```

### ui-mockup
```
Use case: ui-mockup
Primary request: mobile app home screen for a local farmers market with vendors and daily specials
Asset type: mobile app screen
Style/medium: realistic product UI, not concept art
Composition/framing: clean vertical mobile layout with clear hierarchy
Constraints: practical layout, clear typography, no logos or trademarks, no watermark
```

### infographic-diagram
```
Use case: infographic-diagram
Primary request: detailed infographic of an automatic coffee machine flow
Scene/backdrop: clean, light neutral background
Subject: bean hopper -> grinder -> brew group -> boiler -> water tank -> drip tray
Style/medium: clean vector-like infographic with clear callouts and arrows
Composition/framing: vertical poster layout, top-to-bottom flow
Text (verbatim): "Bean Hopper", "Grinder", "Brew Group", "Boiler", "Water Tank", "Drip Tray"
Constraints: clear labels, strong contrast, no logos or trademarks, no watermark
```

### logo-brand
```
Use case: logo-brand
Primary request: original logo for "Field & Flour", a local bakery
Style/medium: vector logo mark; flat colors; minimal
Composition/framing: single centered logo on a plain background with generous padding
Constraints: strong silhouette, balanced negative space; original design only; no gradients unless essential; no trademarks; no watermark
```

### illustration-story
```
Use case: illustration-story
Primary request: 4-panel comic about a pet left alone at home
Scene/backdrop: cozy living room across panels
Subject: pet reacting to the owner leaving, then relaxing, then returning to a composed pose
Style/medium: comic illustration with clear panels
Composition/framing: 4 equal-sized vertical panels, readable actions per panel
Constraints: no text; no logos or trademarks; no watermark
```

### stylized-concept
```
Use case: stylized-concept
Primary request: cavernous hangar interior with tall support beams and drifting fog
Scene/backdrop: industrial hangar interior, deep scale, light haze
Subject: compact shuttle parked near the center
Style/medium: cinematic concept art, industrial realism
Composition/framing: wide-angle, low-angle
Lighting/mood: volumetric light rays cutting through fog
Constraints: no logos or trademarks; no watermark
```

### historical-scene
```
Use case: historical-scene
Primary request: outdoor crowd scene in Bethel, New York on August 16, 1969
Scene/backdrop: open field with period-appropriate staging
Subject: crowd in period-accurate clothing, authentic environment
Style/medium: photorealistic photo
Composition/framing: wide shot, eye-level
Constraints: period-accurate details; no modern objects; no logos or trademarks; no watermark
```

## Asset type templates (taxonomy-aligned)

### Website assets template
```
Use case: <photorealistic-natural|stylized-concept|product-mockup|infographic-diagram|ui-mockup>
Asset type: <hero image / section illustration / blog header>
Primary request: <short description>
Scene/backdrop: <environment or abstract backdrop>
Subject: <main subject>
Style/medium: <photo/illustration/3D>
Composition/framing: <wide/centered; note usable negative space only if needed>
Lighting/mood: <soft/bright/neutral>
Color palette: <brand colors or neutral>
Constraints: <no text; no logos; no watermark; leave room for UI if needed>
```

### Website assets example: minimal hero background
```
Use case: stylized-concept
Asset type: landing page hero background
Primary request: minimal abstract background with a soft gradient and subtle texture
Style/medium: matte illustration / soft-rendered abstract background
Composition/framing: wide composition with usable negative space for page copy
Lighting/mood: gentle studio glow
Color palette: restrained neutral palette
Constraints: no text; no logos; no watermark
```

### Website assets example: feature section illustration
```
Use case: stylized-concept
Asset type: feature section illustration
Primary request: simple abstract shapes suggesting connection and flow
Scene/backdrop: subtle light-gray backdrop with faint texture
Style/medium: flat illustration; soft shadows; restrained contrast
Composition/framing: centered cluster; open margins for UI
Color palette: muted neutral palette
Constraints: no text; no logos; no watermark
```

### Website assets example: blog header image
```
Use case: photorealistic-natural
Asset type: blog header image
Primary request: overhead desk scene with notebook, pen, and coffee cup
Scene/backdrop: warm wooden tabletop
Style/medium: photorealistic photo
Composition/framing: wide crop with clean room for page copy
Lighting/mood: soft morning light
Constraints: no text; no logos; no watermark
```

### Game assets template
```
Use case: stylized-concept
Asset type: <game environment concept art / game character concept / game UI icon / tileable game texture>
Primary request: <biome/scene/character/icon/material>
Scene/backdrop: <location + set dressing> (if applicable)
Subject: <main focal element(s)>
Style/medium: <realistic/stylized>; <concept art / character render / UI icon / texture>
Composition/framing: <wide/establishing/top-down>; <camera angle>; <focal point placement>
Lighting/mood: <time of day>; <mood>; <volumetric/fog/etc>
Constraints: no logos or trademarks; no watermark
```

### Game assets example: environment concept art
```
Use case: stylized-concept
Asset type: game environment concept art
Primary request: cavernous hangar interior with tall support beams and drifting fog
Scene/backdrop: industrial hangar interior, deep scale, light haze
Subject: compact shuttle parked near the center
Style/medium: cinematic concept art, industrial realism
Composition/framing: wide-angle, low-angle
Lighting/mood: volumetric light rays cutting through fog
Constraints: no logos or trademarks; no watermark
```

### Game assets example: character concept
```
Use case: stylized-concept
Asset type: game character concept
Primary request: desert scout character with layered travel gear
Subject: long coat, satchel, practical travel clothing
Style/medium: character render; stylized realism
Composition/framing: neutral hero pose on a simple backdrop
Constraints: no logos or trademarks; no watermark
```

### Game assets example: UI icon
```
Use case: stylized-concept
Asset type: game UI icon
Primary request: round shield icon with a subtle rune pattern
Style/medium: painted game UI icon
Composition/framing: centered icon; generous padding; clear silhouette
Constraints: no text; no background scene elements; no logos or trademarks; no watermark
```

### Game assets example: tileable texture
```
Use case: stylized-concept
Asset type: tileable game texture
Primary request: worn sandstone blocks
Style/medium: seamless tileable texture; PBR-ish look
Scene/backdrop: neutral lighting reference only
Constraints: seamless edges; no obvious focal elements; no text; no logos or trademarks; no watermark
```

### Wireframe template
```
Use case: ui-mockup
Asset type: website wireframe
Primary request: <page or flow to sketch>
Style/medium: low-fi grayscale wireframe
Composition/framing: <landscape or portrait to match expected device>
Subject: <sections in order; grid/columns; key labels>
Constraints: no color; no logos; no real photos; no watermark
```

### Wireframe example: homepage (desktop)
```
Use case: ui-mockup
Asset type: website wireframe
Primary request: SaaS homepage layout with clear hierarchy
Style/medium: low-fi grayscale wireframe
Subject: top nav; hero with headline and CTA; three feature cards; testimonial strip; pricing preview; footer
Composition/framing: landscape desktop layout
Constraints: label major blocks; no color; no logos; no real photos; no watermark
```

### Wireframe example: pricing page
```
Use case: ui-mockup
Asset type: website wireframe
Primary request: pricing page layout with comparison table
Style/medium: low-fi grayscale wireframe
Subject: header; plan toggle; 3 pricing cards; comparison table; FAQ accordion; footer
Composition/framing: desktop or tablet layout
Constraints: label key areas; no color; no logos; no real photos; no watermark
```

### Wireframe example: mobile onboarding flow
```
Use case: ui-mockup
Asset type: mobile onboarding wireframe
Primary request: three-screen mobile onboarding flow
Style/medium: low-fi grayscale wireframe
Subject: screen 1 headline and CTA; screen 2 feature bullets; screen 3 form fields and CTA
Composition/framing: portrait mobile layout
Constraints: label screens and blocks; no color; no logos; no real photos; no watermark
```

### Logo template
```
Use case: logo-brand
Asset type: logo concept
Primary request: <brand idea or symbol concept>
Style/medium: vector logo mark; flat colors; minimal
Composition/framing: centered mark; clear silhouette; generous margin
Color palette: <1-2 colors; high contrast>
Text (verbatim): "<exact name>" (only if needed)
Constraints: no gradients; no mockups; no 3D; no watermark
```

### Logo example: abstract symbol mark
```
Use case: logo-brand
Asset type: logo concept
Primary request: geometric leaf symbol suggesting sustainability and growth
Style/medium: vector logo mark; flat colors; minimal
Composition/framing: centered mark; clear silhouette
Color palette: deep green and off-white
Constraints: no text unless requested; no gradients; no mockups; no 3D; no watermark
```

### Logo example: monogram mark
```
Use case: logo-brand
Asset type: logo concept
Primary request: interlocking monogram of the letters "AV"
Style/medium: vector logo mark; flat colors; minimal
Composition/framing: centered mark; balanced spacing
Color palette: black on white
Constraints: no gradients; no mockups; no 3D; no watermark
```

### Logo example: wordmark
```
Use case: logo-brand
Asset type: logo concept
Primary request: clean wordmark for a modern studio
Style/medium: vector wordmark; flat colors; minimal
Text (verbatim): "Studio North"
Composition/framing: centered text; even letter spacing
Constraints: no gradients; no mockups; no 3D; no watermark
```

## Edit

### text-localization
```
Use case: text-localization
Input images: Image 1: original infographic
Primary request: replace "Bean Hopper", "Grinder", "Brew Group", "Boiler", "Water Tank", and "Drip Tray" with "Tolva", "Molino", "Grupo de infusión", "Caldera", "Depósito de agua", and "Bandeja de goteo"
Constraints: change only the text; preserve layout, typography, spacing, and hierarchy; no extra words; do not alter logos or imagery
```

### identity-preserve
```
Use case: identity-preserve
Input images: Image 1: person photo; Image 2..N: clothing references
Primary request: replace only the clothing with the provided garments
Constraints: preserve face, body shape, pose, hair, expression, and identity; match lighting and shadows; keep the background unchanged; no accessories or text
```

### precise-object-edit
```
Use case: precise-object-edit
Input images: Image 1: room photo
Primary request: replace only the white chairs with wooden chairs
Constraints: preserve camera angle, room lighting, floor shadows, and surrounding objects; keep all other aspects unchanged
```

### lighting-weather
```
Use case: lighting-weather
Input images: Image 1: original photo
Primary request: make it look like a winter evening with gentle snowfall
Constraints: preserve subject identity, geometry, camera angle, and composition; change only lighting, atmosphere, and weather
```

### background-extraction
```
Use case: background-extraction
Input images: Image 1: product photo
Primary request: isolate the product on a clean transparent background
Constraints: crisp silhouette; no halos or fringing; preserve label text exactly; no restyling
```

### style-transfer
```
Use case: style-transfer
Input images: Image 1: style reference
Primary request: apply Image 1's visual style to a man riding a motorcycle on a plain white backdrop
Constraints: preserve palette, texture, and brushwork; no extra elements
```

### compositing
```
Use case: compositing
Input images: Image 1: base scene; Image 2: subject to insert
Primary request: place the subject from Image 2 next to the person in Image 1
Constraints: match lighting, perspective, and scale; keep the base framing unchanged; no extra elements
```

### sketch-to-render
```
Use case: sketch-to-render
Input images: Image 1: drawing
Primary request: turn the drawing into a photorealistic image
Constraints: preserve layout, proportions, and perspective; choose realistic materials and lighting; do not add new elements or text
```

#!/usr/bin/env python3
"""Fallback CLI for explicit image generation or editing with GPT Image models.

Used only when the user explicitly opts into CLI fallback mode.

Defaults to gpt-image-1.5 and a structured prompt augmentation workflow.
"""

from __future__ import annotations

import argparse
import asyncio
import base64
import json
import os
from pathlib import Path
import re
import sys
import time
from typing import Any, Dict, Iterable, List, Optional, Tuple

from io import BytesIO

DEFAULT_MODEL = "gpt-image-1.5"
DEFAULT_SIZE = "1024x1024"
DEFAULT_QUALITY = "auto"
DEFAULT_OUTPUT_FORMAT = "png"
DEFAULT_CONCURRENCY = 5
DEFAULT_DOWNSCALE_SUFFIX = "-web"
DEFAULT_OUTPUT_PATH = "output/imagegen/output.png"
GPT_IMAGE_MODEL_PREFIX = "gpt-image-"

ALLOWED_SIZES = {"1024x1024", "1536x1024", "1024x1536", "auto"}
ALLOWED_QUALITIES = {"low", "medium", "high", "auto"}
ALLOWED_BACKGROUNDS = {"transparent", "opaque", "auto", None}
ALLOWED_INPUT_FIDELITIES = {"low", "high", None}

MAX_IMAGE_BYTES = 50 * 1024 * 1024
MAX_BATCH_JOBS = 500


def _die(message: str, code: int = 1) -> None:
    print(f"Error: {message}", file=sys.stderr)
    raise SystemExit(code)


def _warn(message: str) -> None:
    print(f"Warning: {message}", file=sys.stderr)


def _dependency_hint(package: str, *, upgrade: bool = False) -> str:
    command = f"uv pip install {'-U ' if upgrade else ''}{package}"
    return (
        "Activate the repo-selected environment first, then install it with "
        f"`{command}`. If this repo uses a local virtualenv, start with "
        "`source .venv/bin/activate`; otherwise use this repo's configured shared fallback "
        "environment. If your project declares dependencies, prefer that project's normal "
        "`uv sync` flow."
    )


def _ensure_api_key(dry_run: bool) -> None:
    if os.getenv("OPENAI_API_KEY"):
        print("OPENAI_API_KEY is set.", file=sys.stderr)
        return
    if dry_run:
        _warn("OPENAI_API_KEY is not set; dry-run only.")
        return
    _die("OPENAI_API_KEY is not set. Export it before running.")


def _read_prompt(prompt: Optional[str], prompt_file: Optional[str]) -> str:
    if prompt and prompt_file:
        _die("Use --prompt or --prompt-file, not both.")
    if prompt_file:
        path = Path(prompt_file)
        if not path.exists():
            _die(f"Prompt file not found: {path}")
        return path.read_text(encoding="utf-8").strip()
    if prompt:
        return prompt.strip()
    _die("Missing prompt. Use --prompt or --prompt-file.")
    return ""  # unreachable


def _check_image_paths(paths: Iterable[str]) -> List[Path]:
    resolved: List[Path] = []
    for raw in paths:
        path = Path(raw)
        if not path.exists():
            _die(f"Image file not found: {path}")
        if path.stat().st_size > MAX_IMAGE_BYTES:
            _warn(f"Image exceeds 50MB limit: {path}")
        resolved.append(path)
    return resolved


def _normalize_output_format(fmt: Optional[str]) -> str:
    if not fmt:
        return DEFAULT_OUTPUT_FORMAT
    fmt = fmt.lower()
    if fmt not in {"png", "jpeg", "jpg", "webp"}:
        _die("output-format must be png, jpeg, jpg, or webp.")
    return "jpeg" if fmt == "jpg" else fmt


def _validate_size(size: str) -> None:
    if size not in ALLOWED_SIZES:
        _die(
            "size must be one of 1024x1024, 1536x1024, 1024x1536, or auto for GPT image models."
        )


def _validate_quality(quality: str) -> None:
    if quality not in ALLOWED_QUALITIES:
        _die("quality must be one of low, medium, high, or auto.")


def _validate_background(background: Optional[str]) -> None:
    if background not in ALLOWED_BACKGROUNDS:
        _die("background must be one of transparent, opaque, or auto.")


def _validate_input_fidelity(input_fidelity: Optional[str]) -> None:
    if input_fidelity not in ALLOWED_INPUT_FIDELITIES:
        _die("input-fidelity must be one of low or high.")


def _validate_model(model: str) -> None:
    if not model.startswith(GPT_IMAGE_MODEL_PREFIX):
        _die(
            "model must be a GPT Image model (for example gpt-image-1.5, gpt-image-1, or gpt-image-1-mini)."
        )


def _validate_transparency(background: Optional[str], output_format: str) -> None:
    if background == "transparent" and output_format not in {"png", "webp"}:
        _die("transparent background requires output-format png or webp.")


def _validate_generate_payload(payload: Dict[str, Any]) -> None:
    _validate_model(str(payload.get("model", DEFAULT_MODEL)))
    n = int(payload.get("n", 1))
    if n < 1 or n > 10:
        _die("n must be between 1 and 10")
    size = str(payload.get("size", DEFAULT_SIZE))
    quality = str(payload.get("quality", DEFAULT_QUALITY))
    background = payload.get("background")
    _validate_size(size)
    _validate_quality(quality)
    _validate_background(background)
    oc = payload.get("output_compression")
    if oc is not None and not (0 <= int(oc) <= 100):
        _die("output_compression must be between 0 and 100")


def _build_output_paths(
    out: str,
    output_format: str,
    count: int,
    out_dir: Optional[str],
) -> List[Path]:
    ext = "." + output_format

    if out_dir:
        out_base = Path(out_dir)
        out_base.mkdir(parents=True, exist_ok=True)
        return [out_base / f"image_{i}{ext}" for i in range(1, count + 1)]

    out_path = Path(out)
    if out_path.exists() and out_path.is_dir():
        out_path.mkdir(parents=True, exist_ok=True)
        return [out_path / f"image_{i}{ext}" for i in range(1, count + 1)]

    if out_path.suffix == "":
        out_path = out_path.with_suffix(ext)
    elif output_format and out_path.suffix.lstrip(".").lower() != output_format:
        _warn(
            f"Output extension {out_path.suffix} does not match output-format {output_format}."
        )

    if count == 1:
        return [out_path]

    return [
        out_path.with_name(f"{out_path.stem}-{i}{out_path.suffix}")
        for i in range(1, count + 1)
    ]


def _augment_prompt(args: argparse.Namespace, prompt: str) -> str:
    fields = _fields_from_args(args)
    return _augment_prompt_fields(args.augment, prompt, fields)


def _augment_prompt_fields(augment: bool, prompt: str, fields: Dict[str, Optional[str]]) -> str:
    if not augment:
        return prompt

    sections: List[str] = []
    if fields.get("use_case"):
        sections.append(f"Use case: {fields['use_case']}")
    sections.append(f"Primary request: {prompt}")
    if fields.get("scene"):
        sections.append(f"Scene/background: {fields['scene']}")
    if fields.get("subject"):
        sections.append(f"Subject: {fields['subject']}")
    if fields.get("style"):
        sections.append(f"Style/medium: {fields['style']}")
    if fields.get("composition"):
        sections.append(f"Composition/framing: {fields['composition']}")
    if fields.get("lighting"):
        sections.append(f"Lighting/mood: {fields['lighting']}")
    if fields.get("palette"):
        sections.append(f"Color palette: {fields['palette']}")
    if fields.get("materials"):
        sections.append(f"Materials/textures: {fields['materials']}")
    if fields.get("text"):
        sections.append(f"Text (verbatim): \"{fields['text']}\"")
    if fields.get("constraints"):
        sections.append(f"Constraints: {fields['constraints']}")
    if fields.get("negative"):
        sections.append(f"Avoid: {fields['negative']}")

    return "\n".join(sections)


def _fields_from_args(args: argparse.Namespace) -> Dict[str, Optional[str]]:
    return {
        "use_case": getattr(args, "use_case", None),
        "scene": getattr(args, "scene", None),
        "subject": getattr(args, "subject", None),
        "style": getattr(args, "style", None),
        "composition": getattr(args, "composition", None),
        "lighting": getattr(args, "lighting", None),
        "palette": getattr(args, "palette", None),
        "materials": getattr(args, "materials", None),
        "text": getattr(args, "text", None),
        "constraints": getattr(args, "constraints", None),
        "negative": getattr(args, "negative", None),
    }


def _print_request(payload: dict) -> None:
    print(json.dumps(payload, indent=2, sort_keys=True))


def _decode_and_write(images: List[str], outputs: List[Path], force: bool) -> None:
    for idx, image_b64 in enumerate(images):
        if idx >= len(outputs):
            break
        out_path = outputs[idx]
        if out_path.exists() and not force:
            _die(f"Output already exists: {out_path} (use --force to overwrite)")
        out_path.parent.mkdir(parents=True, exist_ok=True)
        out_path.write_bytes(base64.b64decode(image_b64))
        print(f"Wrote {out_path}")


def _derive_downscale_path(path: Path, suffix: str) -> Path:
    if suffix and not suffix.startswith("-") and not suffix.startswith("_"):
        suffix = "-" + suffix
    return path.with_name(f"{path.stem}{suffix}{path.suffix}")


def _downscale_image_bytes(image_bytes: bytes, *, max_dim: int, output_format: str) -> bytes:
    try:
        from PIL import Image
    except Exception:
        _die(f"Downscaling requires Pillow. {_dependency_hint('pillow')}")

    if max_dim < 1:
        _die("--downscale-max-dim must be >= 1")

    with Image.open(BytesIO(image_bytes)) as img:
        img.load()
        w, h = img.size
        scale = min(1.0, float(max_dim) / float(max(w, h)))
        target = (max(1, int(round(w * scale))), max(1, int(round(h * scale))))

        resized = img if target == (w, h) else img.resize(target, Image.Resampling.LANCZOS)

        fmt = output_format.lower()
        if fmt == "jpg":
            fmt = "jpeg"

        if fmt == "jpeg":
            if resized.mode in ("RGBA", "LA") or ("transparency" in getattr(resized, "info", {})):
                bg = Image.new("RGB", resized.size, (255, 255, 255))
                bg.paste(resized.convert("RGBA"), mask=resized.convert("RGBA").split()[-1])
                resized = bg
            else:
                resized = resized.convert("RGB")

        out = BytesIO()
        resized.save(out, format=fmt.upper())
        return out.getvalue()


def _decode_write_and_downscale(
    images: List[str],
    outputs: List[Path],
    *,
    force: bool,
    downscale_max_dim: Optional[int],
    downscale_suffix: str,
    output_format: str,
) -> None:
    for idx, image_b64 in enumerate(images):
        if idx >= len(outputs):
            break
        out_path = outputs[idx]
        if out_path.exists() and not force:
            _die(f"Output already exists: {out_path} (use --force to overwrite)")
        out_path.parent.mkdir(parents=True, exist_ok=True)

        raw = base64.b64decode(image_b64)
        out_path.write_bytes(raw)
        print(f"Wrote {out_path}")

        if downscale_max_dim is None:
            continue

        derived = _derive_downscale_path(out_path, downscale_suffix)
        if derived.exists() and not force:
            _die(f"Output already exists: {derived} (use --force to overwrite)")
        derived.parent.mkdir(parents=True, exist_ok=True)
        resized = _downscale_image_bytes(raw, max_dim=downscale_max_dim, output_format=output_format)
        derived.write_bytes(resized)
        print(f"Wrote {derived}")


def _create_client():
    try:
        from openai import OpenAI
    except ImportError:
        _die(f"openai SDK not installed in the active environment. {_dependency_hint('openai')}")
    return OpenAI()


def _create_async_client():
    try:
        from openai import AsyncOpenAI
    except ImportError:
        try:
            import openai as _openai  # noqa: F401
        except ImportError:
            _die(
                f"openai SDK not installed in the active environment. {_dependency_hint('openai')}"
            )
        _die(
            "AsyncOpenAI not available in this openai SDK version. "
            f"{_dependency_hint('openai', upgrade=True)}"
        )
    return AsyncOpenAI()


def _slugify(value: str) -> str:
    value = value.strip().lower()
    value = re.sub(r"[^a-z0-9]+", "-", value)
    value = re.sub(r"-{2,}", "-", value).strip("-")
    return value[:60] if value else "job"


def _normalize_job(job: Any, idx: int) -> Dict[str, Any]:
    if isinstance(job, str):
        prompt = job.strip()
        if not prompt:
            _die(f"Empty prompt at job {idx}")
        return {"prompt": prompt}
    if isinstance(job, dict):
        if "prompt" not in job or not str(job["prompt"]).strip():
            _die(f"Missing prompt for job {idx}")
        return job
    _die(f"Invalid job at index {idx}: expected string or object.")
    return {}  # unreachable


def _read_jobs_jsonl(path: str) -> List[Dict[str, Any]]:
    p = Path(path)
    if not p.exists():
        _die(f"Input file not found: {p}")
    jobs: List[Dict[str, Any]] = []
    for line_no, raw in enumerate(p.read_text(encoding="utf-8").splitlines(), start=1):
        line = raw.strip()
        if not line or line.startswith("#"):
            continue
        try:
            item: Any
            if line.startswith("{"):
                item = json.loads(line)
            else:
                item = line
            jobs.append(_normalize_job(item, idx=line_no))
        except json.JSONDecodeError as exc:
            _die(f"Invalid JSON on line {line_no}: {exc}")
    if not jobs:
        _die("No jobs found in input file.")
    if len(jobs) > MAX_BATCH_JOBS:
        _die(f"Too many jobs ({len(jobs)}). Max is {MAX_BATCH_JOBS}.")
    return jobs


def _merge_non_null(dst: Dict[str, Any], src: Dict[str, Any]) -> Dict[str, Any]:
    merged = dict(dst)
    for k, v in src.items():
        if v is not None:
            merged[k] = v
    return merged


def _job_output_paths(
    *,
    out_dir: Path,
    output_format: str,
    idx: int,
    prompt: str,
    n: int,
    explicit_out: Optional[str],
) -> List[Path]:
    out_dir.mkdir(parents=True, exist_ok=True)
    ext = "." + output_format

    if explicit_out:
        base = Path(explicit_out)
        if base.suffix == "":
            base = base.with_suffix(ext)
        elif base.suffix.lstrip(".").lower() != output_format:
            _warn(
                f"Job {idx}: output extension {base.suffix} does not match output-format {output_format}."
            )
        base = out_dir / base.name
    else:
        slug = _slugify(prompt[:80])
        base = out_dir / f"{idx:03d}-{slug}{ext}"

    if n == 1:
        return [base]
    return [
        base.with_name(f"{base.stem}-{i}{base.suffix}")
        for i in range(1, n + 1)
    ]


def _extract_retry_after_seconds(exc: Exception) -> Optional[float]:
    # Best-effort: openai SDK errors vary by version. Prefer a conservative fallback.
    for attr in ("retry_after", "retry_after_seconds"):
        val = getattr(exc, attr, None)
        if isinstance(val, (int, float)) and val >= 0:
            return float(val)
    msg = str(exc)
    m = re.search(r"retry[- ]after[:= ]+([0-9]+(?:\\.[0-9]+)?)", msg, re.IGNORECASE)
    if m:
        try:
            return float(m.group(1))
        except Exception:
            return None
    return None


def _is_rate_limit_error(exc: Exception) -> bool:
    name = exc.__class__.__name__.lower()
    if "ratelimit" in name or "rate_limit" in name:
        return True
    msg = str(exc).lower()
    return "429" in msg or "rate limit" in msg or "too many requests" in msg


def _is_transient_error(exc: Exception) -> bool:
    if _is_rate_limit_error(exc):
        return True
    name = exc.__class__.__name__.lower()
    if "timeout" in name or "timedout" in name or "tempor" in name:
        return True
    msg = str(exc).lower()
    return "timeout" in msg or "timed out" in msg or "connection reset" in msg


async def _generate_one_with_retries(
    client: Any,
    payload: Dict[str, Any],
    *,
    attempts: int,
    job_label: str,
) -> Any:
    last_exc: Optional[Exception] = None
    for attempt in range(1, attempts + 1):
        try:
            return await client.images.generate(**payload)
        except Exception as exc:
            last_exc = exc
            if not _is_transient_error(exc):
                raise
            if attempt == attempts:
                raise
            sleep_s = _extract_retry_after_seconds(exc)
            if sleep_s is None:
                sleep_s = min(60.0, 2.0**attempt)
            print(
                f"{job_label} attempt {attempt}/{attempts} failed ({exc.__class__.__name__}); retrying in {sleep_s:.1f}s",
                file=sys.stderr,
            )
            await asyncio.sleep(sleep_s)
    raise last_exc or RuntimeError("unknown error")


async def _run_generate_batch(args: argparse.Namespace) -> int:
    jobs = _read_jobs_jsonl(args.input)
    out_dir = Path(args.out_dir)

    base_fields = _fields_from_args(args)
    base_payload = {
        "model": args.model,
        "n": args.n,
        "size": args.size,
        "quality": args.quality,
        "background": args.background,
        "output_format": args.output_format,
        "output_compression": args.output_compression,
        "moderation": args.moderation,
    }

    if args.dry_run:
        for i, job in enumerate(jobs, start=1):
            prompt = str(job["prompt"]).strip()
            fields = _merge_non_null(base_fields, job.get("fields", {}))
            # Allow flat job keys as well (use_case, scene, etc.)
            fields = _merge_non_null(fields, {k: job.get(k) for k in base_fields.keys()})
            augmented = _augment_prompt_fields(args.augment, prompt, fields)

            job_payload = dict(base_payload)
            job_payload["prompt"] = augmented
            job_payload = _merge_non_null(job_payload, {k: job.get(k) for k in base_payload.keys()})
            job_payload = {k: v for k, v in job_payload.items() if v is not None}

            _validate_generate_payload(job_payload)
            effective_output_format = _normalize_output_format(job_payload.get("output_format"))
            _validate_transparency(job_payload.get("background"), effective_output_format)
            job_payload["output_format"] = effective_output_format

            n = int(job_payload.get("n", 1))
            outputs = _job_output_paths(
                out_dir=out_dir,
                output_format=effective_output_format,
                idx=i,
                prompt=prompt,
                n=n,
                explicit_out=job.get("out"),
            )
            downscaled = None
            if args.downscale_max_dim is not None:
                downscaled = [
                    str(_derive_downscale_path(p, args.downscale_suffix)) for p in outputs
                ]
            _print_request(
                {
                    "endpoint": "/v1/images/generations",
                    "job": i,
                    "outputs": [str(p) for p in outputs],
                    "outputs_downscaled": downscaled,
                    **job_payload,
                }
            )
        return 0

    client = _create_async_client()
    sem = asyncio.Semaphore(args.concurrency)

    any_failed = False

    async def run_job(i: int, job: Dict[str, Any]) -> Tuple[int, Optional[str]]:
        nonlocal any_failed
        prompt = str(job["prompt"]).strip()
        job_label = f"[job {i}/{len(jobs)}]"

        fields = _merge_non_null(base_fields, job.get("fields", {}))
        fields = _merge_non_null(fields, {k: job.get(k) for k in base_fields.keys()})
        augmented = _augment_prompt_fields(args.augment, prompt, fields)

        payload = dict(base_payload)
        payload["prompt"] = augmented
        payload = _merge_non_null(payload, {k: job.get(k) for k in base_payload.keys()})
        payload = {k: v for k, v in payload.items() if v is not None}

        n = int(payload.get("n", 1))
        _validate_generate_payload(payload)
        effective_output_format = _normalize_output_format(payload.get("output_format"))
        _validate_transparency(payload.get("background"), effective_output_format)
        payload["output_format"] = effective_output_format
        outputs = _job_output_paths(
            out_dir=out_dir,
            output_format=effective_output_format,
            idx=i,
            prompt=prompt,
            n=n,
            explicit_out=job.get("out"),
        )
        try:
            async with sem:
                print(f"{job_label} starting", file=sys.stderr)
                started = time.time()
                result = await _generate_one_with_retries(
                    client,
                    payload,
                    attempts=args.max_attempts,
                    job_label=job_label,
                )
                elapsed = time.time() - started
                print(f"{job_label} completed in {elapsed:.1f}s", file=sys.stderr)
            images = [item.b64_json for item in result.data]
            _decode_write_and_downscale(
                images,
                outputs,
                force=args.force,
                downscale_max_dim=args.downscale_max_dim,
                downscale_suffix=args.downscale_suffix,
                output_format=effective_output_format,
            )
            return i, None
        except Exception as exc:
            any_failed = True
            print(f"{job_label} failed: {exc}", file=sys.stderr)
            if args.fail_fast:
                raise
            return i, str(exc)

    tasks = [asyncio.create_task(run_job(i, job)) for i, job in enumerate(jobs, start=1)]

    try:
        await asyncio.gather(*tasks)
    except Exception:
        for t in tasks:
            if not t.done():
                t.cancel()
        raise

    return 1 if any_failed else 0


def _generate_batch(args: argparse.Namespace) -> None:
    exit_code = asyncio.run(_run_generate_batch(args))
    if exit_code:
        raise SystemExit(exit_code)


def _generate(args: argparse.Namespace) -> None:
    prompt = _read_prompt(args.prompt, args.prompt_file)
    prompt = _augment_prompt(args, prompt)

    payload = {
        "model": args.model,
        "prompt": prompt,
        "n": args.n,
        "size": args.size,
        "quality": args.quality,
        "background": args.background,
        "output_format": args.output_format,
        "output_compression": args.output_compression,
        "moderation": args.moderation,
    }
    payload = {k: v for k, v in payload.items() if v is not None}

    output_format = _normalize_output_format(args.output_format)
    _validate_transparency(args.background, output_format)
    payload["output_format"] = output_format
    output_paths = _build_output_paths(args.out, output_format, args.n, args.out_dir)
    downscaled = None
    if args.downscale_max_dim is not None:
        downscaled = [str(_derive_downscale_path(p, args.downscale_suffix)) for p in output_paths]

    if args.dry_run:
        _print_request(
            {
                "endpoint": "/v1/images/generations",
                "outputs": [str(p) for p in output_paths],
                "outputs_downscaled": downscaled,
                **payload,
            }
        )
        return

    print(
        "Calling Image API (generation). This can take up to a couple of minutes.",
        file=sys.stderr,
    )
    started = time.time()
    client = _create_client()
    result = client.images.generate(**payload)
    elapsed = time.time() - started
    print(f"Generation completed in {elapsed:.1f}s.", file=sys.stderr)

    images = [item.b64_json for item in result.data]
    _decode_write_and_downscale(
        images,
        output_paths,
        force=args.force,
        downscale_max_dim=args.downscale_max_dim,
        downscale_suffix=args.downscale_suffix,
        output_format=output_format,
    )


def _edit(args: argparse.Namespace) -> None:
    prompt = _read_prompt(args.prompt, args.prompt_file)
    prompt = _augment_prompt(args, prompt)

    image_paths = _check_image_paths(args.image)
    mask_path = Path(args.mask) if args.mask else None
    if mask_path:
        if not mask_path.exists():
            _die(f"Mask file not found: {mask_path}")
        if mask_path.suffix.lower() != ".png":
            _warn(f"Mask should be a PNG with an alpha channel: {mask_path}")
        if mask_path.stat().st_size > MAX_IMAGE_BYTES:
            _warn(f"Mask exceeds 50MB limit: {mask_path}")

    payload = {
        "model": args.model,
        "prompt": prompt,
        "n": args.n,
        "size": args.size,
        "quality": args.quality,
        "background": args.background,
        "output_format": args.output_format,
        "output_compression": args.output_compression,
        "input_fidelity": args.input_fidelity,
        "moderation": args.moderation,
    }
    payload = {k: v for k, v in payload.items() if v is not None}

    output_format = _normalize_output_format(args.output_format)
    _validate_transparency(args.background, output_format)
    payload["output_format"] = output_format
    _validate_input_fidelity(args.input_fidelity)
    output_paths = _build_output_paths(args.out, output_format, args.n, args.out_dir)
    downscaled = None
    if args.downscale_max_dim is not None:
        downscaled = [str(_derive_downscale_path(p, args.downscale_suffix)) for p in output_paths]

    if args.dry_run:
        payload_preview = dict(payload)
        payload_preview["image"] = [str(p) for p in image_paths]
        if mask_path:
            payload_preview["mask"] = str(mask_path)
        _print_request(
            {
                "endpoint": "/v1/images/edits",
                "outputs": [str(p) for p in output_paths],
                "outputs_downscaled": downscaled,
                **payload_preview,
            }
        )
        return

    print(
        f"Calling Image API (edit) with {len(image_paths)} image(s).",
        file=sys.stderr,
    )
    started = time.time()
    client = _create_client()

    with _open_files(image_paths) as image_files, _open_mask(mask_path) as mask_file:
        request = dict(payload)
        request["image"] = image_files if len(image_files) > 1 else image_files[0]
        if mask_file is not None:
            request["mask"] = mask_file
        result = client.images.edit(**request)

    elapsed = time.time() - started
    print(f"Edit completed in {elapsed:.1f}s.", file=sys.stderr)
    images = [item.b64_json for item in result.data]
    _decode_write_and_downscale(
        images,
        output_paths,
        force=args.force,
        downscale_max_dim=args.downscale_max_dim,
        downscale_suffix=args.downscale_suffix,
        output_format=output_format,
    )


def _open_files(paths: List[Path]):
    return _FileBundle(paths)


def _open_mask(mask_path: Optional[Path]):
    if mask_path is None:
        return _NullContext()
    return _SingleFile(mask_path)


class _NullContext:
    def __enter__(self):
        return None

    def __exit__(self, exc_type, exc, tb):
        return False


class _SingleFile:
    def __init__(self, path: Path):
        self._path = path
        self._handle = None

    def __enter__(self):
        self._handle = self._path.open("rb")
        return self._handle

    def __exit__(self, exc_type, exc, tb):
        if self._handle:
            try:
                self._handle.close()
            except Exception:
                pass
        return False


class _FileBundle:
    def __init__(self, paths: List[Path]):
        self._paths = paths
        self._handles: List[object] = []

    def __enter__(self):
        self._handles = [p.open("rb") for p in self._paths]
        return self._handles

    def __exit__(self, exc_type, exc, tb):
        for handle in self._handles:
            try:
                handle.close()
            except Exception:
                pass
        return False


def _add_shared_args(parser: argparse.ArgumentParser) -> None:
    parser.add_argument("--model", default=DEFAULT_MODEL)
    parser.add_argument("--prompt")
    parser.add_argument("--prompt-file")
    parser.add_argument("--n", type=int, default=1)
    parser.add_argument("--size", default=DEFAULT_SIZE)
    parser.add_argument("--quality", default=DEFAULT_QUALITY)
    parser.add_argument("--background")
    parser.add_argument("--output-format")
    parser.add_argument("--output-compression", type=int)
    parser.add_argument("--moderation")
    parser.add_argument("--out", default=DEFAULT_OUTPUT_PATH)
    parser.add_argument("--out-dir")
    parser.add_argument("--force", action="store_true")
    parser.add_argument("--dry-run", action="store_true")
    parser.add_argument("--augment", dest="augment", action="store_true")
    parser.add_argument("--no-augment", dest="augment", action="store_false")
    parser.set_defaults(augment=True)

    # Prompt augmentation hints
    parser.add_argument("--use-case")
    parser.add_argument("--scene")
    parser.add_argument("--subject")
    parser.add_argument("--style")
    parser.add_argument("--composition")
    parser.add_argument("--lighting")
    parser.add_argument("--palette")
    parser.add_argument("--materials")
    parser.add_argument("--text")
    parser.add_argument("--constraints")
    parser.add_argument("--negative")

    # Post-processing (optional): generate an additional downscaled copy for fast web loading.
    parser.add_argument("--downscale-max-dim", type=int)
    parser.add_argument("--downscale-suffix", default=DEFAULT_DOWNSCALE_SUFFIX)


def main() -> int:
    parser = argparse.ArgumentParser(
        description="Fallback CLI for explicit image generation or editing via GPT Image models"
    )
    subparsers = parser.add_subparsers(dest="command", required=True)

    gen_parser = subparsers.add_parser("generate", help="Create a new image")
    _add_shared_args(gen_parser)
    gen_parser.set_defaults(func=_generate)

    batch_parser = subparsers.add_parser(
        "generate-batch",
        help="Generate multiple prompts concurrently (JSONL input)",
    )
    _add_shared_args(batch_parser)
    batch_parser.add_argument("--input", required=True, help="Path to JSONL file (one job per line)")
    batch_parser.add_argument("--concurrency", type=int, default=DEFAULT_CONCURRENCY)
    batch_parser.add_argument("--max-attempts", type=int, default=3)
    batch_parser.add_argument("--fail-fast", action="store_true")
    batch_parser.set_defaults(func=_generate_batch)

    edit_parser = subparsers.add_parser("edit", help="Edit an existing image")
    _add_shared_args(edit_parser)
    edit_parser.add_argument("--image", action="append", required=True)
    edit_parser.add_argument("--mask")
    edit_parser.add_argument("--input-fidelity")
    edit_parser.set_defaults(func=_edit)

    args = parser.parse_args()
    if args.n < 1 or args.n > 10:
        _die("--n must be between 1 and 10")
    if getattr(args, "concurrency", 1) < 1 or getattr(args, "concurrency", 1) > 25:
        _die("--concurrency must be between 1 and 25")
    if getattr(args, "max_attempts", 3) < 1 or getattr(args, "max_attempts", 3) > 10:
        _die("--max-attempts must be between 1 and 10")
    if args.output_compression is not None and not (0 <= args.output_compression <= 100):
        _die("--output-compression must be between 0 and 100")
    if args.command == "generate-batch" and not args.out_dir:
        _die("generate-batch requires --out-dir")
    if getattr(args, "downscale_max_dim", None) is not None and args.downscale_max_dim < 1:
        _die("--downscale-max-dim must be >= 1")

    _validate_size(args.size)
    _validate_quality(args.quality)
    _validate_background(args.background)
    _validate_model(args.model)
    _ensure_api_key(args.dry_run)

    args.func(args)
    return 0


if __name__ == "__main__":
    raise SystemExit(main())