What does this skill do, and when should you use it?
This is a focused image-generation and editing skill for GPT Image 2 and OpenAI-compatible image APIs. It supports Garden-local generation, delegation to a host image tool, and prompt-only advisor mode. Its templates cover posters, UI mockups, product visuals, infographics, academic figures, technical diagrams, comics, avatars, storyboards, and branding visuals. The skill separates generation and editing workflows and organizes prompts by visual category.
It first runs check-mode.js to determine the runtime mode, then selects either an image-generation or image-editing workflow and reads a relevant template from references/. In Mode A, generate.js calls an OpenAI-compatible image endpoint and edit.js handles image and mask-based edits; in Mode B, the rendered prompt is passed to the host image tool; in Mode C, the skill produces and saves a reusable prompt. Prompts default to garden-gpt-image-2/prompt/, while Mode A images default to garden-gpt-image-2/image/.
- A designer creating campaign posters, banners, editorial covers, or packaging visuals can use the matching structured template.
- A product team creating chat-interface, livestream-commerce, or landing-page hero mockups can use the UI mockup templates.
- A researcher preparing neural-network diagrams, method pipelines, graphical abstracts, or publication charts can use the academic-figures templates.
- An engineer producing PNG system architectures, flowcharts, sequence diagrams, state machines, or ER diagrams can use the technical-diagrams templates.
- A user with an existing product, portrait, or scene image can use the editing workflows for background replacement, object removal, or retouching.
- An agent without an image tool can use Advisor mode to produce a high-quality prompt for execution in another image system.
How do you install this skill?
- Do not send user images or prompts to OpenAI or a third-party compatible gateway without confirmation; OPENAI_BASE_URL can target arbitrary gateways and their data practices are unspecified.
- The supplied evidence omits the actual scripts and package.json, so dependencies, API behavior, filesystem writes, and error handling remain unverified.
- Technical-diagram templates produce PNG rather than editable SVG; generated text, data, and academic content require human review.
- Mainland-China network compatibility is not documented, and Mode A may depend on reachable external APIs.
- Shell / CLI
- Network access
- Local filesystem
Node.js
Install only this skill with the skills CLI: npx skills add ConardLi/garden-skills -s gpt-image-2 --global. Alternatively, clone the repository and copy skills/gpt-image-2 into a directory scanned by the target agent, such as .claude/skills/, .agents/skills/, or .codex/skills/. The collection is licensed under MIT.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/ConardLi/garden-skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/gpt-image-2" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- A cute baby sea otter
- Replace the background with a clean studio scene
Start by running node skills/gpt-image-2/scripts/check-mode.js. In Mode A, generate an image with node skills/gpt-image-2/scripts/generate.js --prompt "A cute baby sea otter" --size 1024x1024 --quality high, or edit one with node skills/gpt-image-2/scripts/edit.js --image assets/source.png --prompt "Replace the background with a clean studio scene". Configure ENABLE_GARDEN_IMAGEGEN; Mode A also requires OPENAI_API_KEY. Without Garden generation enabled, the skill selects Mode B or Mode C based on whether the host provides an image tool.
What are this skill's strengths and limitations?
- Covers image generation, image editing, and prompt-only advisory use cases.
- Provides structured, scenario-based templates for UI, product, academic, technical, comic, and branding visuals.
- Works with OpenAI-compatible endpoints and supports an alternate OPENAI_BASE_URL.
- Defines consistent prompt and image saving, naming, and reuse rules.
- Mode A requires Node.js, network access, and OPENAI_API_KEY; API, network, or quota failures can stop generation.
- Mode B depends on an image tool supplied by the host agent, while Mode C cannot render images.
- Technical-diagram templates produce PNG bitmaps rather than editable SVG files.
- The README says there are 79 templates, while SKILL.md says 80+, so the documented count is inconsistent.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| GPT Image 2 Creative Workflow this page | 53 · Use with care | ★ 13k | 16d ago | MIT |
| Codex Image Generation Skill (imagegen) ✓ OpenAI · Official | 56 · Use with care | ★ 129k | 12d ago | Apache-2.0 |
| Banana Claude — Gemini Image Creative Director | 59 · Recommended | ★ 1.1k | 9d ago | MIT |
| Codex Image Studio ✓ OpenAI · Official | 55 · Use with care | ★ 28k | 4mo ago | — |
| Blog Image Creative Director | 53 · Use with care | ★ 2.4k | 13d ago | MIT |
Compared with a host's native image tool, this skill adds mode detection, scenario-specific templates, prompt persistence, and image-editing scripts; in Mode B, it still relies on the host tool for rendering. Compared with a prompt-only advisor, Mode A can also call an OpenAI-compatible endpoint and save the resulting image.
How did FollowSkills review this skill?
The documentation separates local, host-delegated, and advisor modes, requires the ENABLE_GARDEN_IMAGEGEN gate, and discloses use of OPENAI_API_KEY, environment files, and a configurable third-party OPENAI_BASE_URL. It does not require confirmation before calls, explain sensitive-data handling, define network-transfer boundaries, or provide rollback and cleanup procedures, so points are deducted.
The workflow, mode branches, command examples, and intended error-handling direction are reasonably clear. However, the scripts are not included in the supplied evidence, so dependency availability, argument validation, reproducibility of key paths, and diagnostic failure behavior cannot be verified. The “80+” claim also conflicts with the README’s “79,” warranting further deduction.
The skill clearly targets image generation, editing, and prompt advising, with A/B/C modes, missing-input questions, and template non-fit guidance. Host-tool detection relies on vague tool-name signals, failure boundaries are not fully specified, and mainland-China network reachability is not addressed, so points are deducted.
The SKILL.md, bilingual READMEs, layered references, manifest version, naming rules, environment variables, examples, and MIT license provide a usable structure. There is no changelog, explicit maintenance owner, update path, installation/dependency guidance, or systematic troubleshooting documentation, so points are deducted.
The files describe a broad template library, graceful mode degradation, and reusable prompt archival, which could cover many image tasks. Because this is a static review and no generation, editing, or output-file usability was verified, the score is capped at the static maximum of 7.
The supplied files provide detailed workflows, parameters, and templates that are partially auditable. They do not include the referenced script implementations, committed tests, CI coverage, or independent reproduction evidence; the gallery is presented as external case material rather than verifiable execution evidence, so coverage remains limited.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →