Generate visual concepts with ak:ai-artist

Use ak:ai-artist to turn a visual concept into a refined Nano Banana prompt and one or more generated PNG assets. The Skill searches a bundled collection of 129 prompt patterns, validates the intended style, mood, and color direction, then routes rendering through the ak:ai-multimodal backend to Google or OpenRouter-backed Google image models.

Choose prompt-led image generation

Use ak:ai-artist when

• You need a product mockup, campaign visual, social asset, banner, infographic, avatar, concept art, or style exploration.
• A curated prompt can anchor a consistent visual direction.
• You want a close match, a remix of several patterns, or an experimental style transformation.
• You want to inspect the assembled prompt before provider generation.

Choose another workflow when

• You need to inspect or transform existing image, audio, video, or document inputs. Use ak:ai-multimodal.
• You need deterministic local resizing, compositing, encoding, or background removal. Use ak:media-processing.
• You need layout, typography, accessibility, and implementation guidance for a product interface. Use ak:frontend-design.
• You cannot send the concept or reference material to an external AI provider.

Prepare the brief and provider

Before you start:

• Complete Onboarding, and confirm Engineer Kit is installed for the current runtime and scope.
• Provide the subject, audience, use case, style, mood, palette, composition, aspect ratio, exact in-image text, exclusions, and output location.
• Decide whether the validation interview may be bypassed with --skip. Without that explicit option, the Skill must refine the brief before generation.
• Make Python 3 and npx available. The rendering script fetches and runs the npm-latest @mrgoonie/multix package.
• Configure GEMINI_API_KEY for direct Google or OPENROUTER_API_KEY for the OpenRouter route, then approve provider terms, network use, and budget.
• Confirm that the prompt and any brand information may leave the local machine.

Generate a reviewable variation

/ak:ai-artist "Create a 16:9 launch banner for a developer CLI: dark background, restrained steel-blue glow, terminal-inspired geometry, no logos, no people, and no text; use search mode and save a PNG"

/ak:ai-artist "Create a 16:9 launch banner for a developer CLI: dark background, restrained steel-blue glow, terminal-inspired geometry, no logos, no people, and no text; use search mode and save a PNG"

$ak:ai-artist "Create a 16:9 launch banner for a developer CLI: dark background, restrained steel-blue glow, terminal-inspired geometry, no logos, no people, and no text; use search mode and save a PNG"

Select the generation shape

Observe the creative stages

1. The Skill validates the brief. Unless --skip was explicit, it resolves material gaps in subject, style, mood, color, text, and output intent.
2. It searches local patterns. BM25 matching ranks the bundled prompt rows against the concept; search mode falls back to a basic prompt when no row scores above zero.
3. It adapts the prompt. Placeholders are replaced with the concept and negative constraints are added when the source lacks them.
4. It applies the selected mode. Creative mode combines matched hints; wild mode selects a random transformation; all mode prepares three variants.
5. It resolves the provider and model. The script chooses the direct Google or OpenRouter command from explicit options, environment, and credentials.
6. It renders the file. A subprocess sends the prompt and parameters to the provider and writes the requested PNG path, with a five-minute script timeout.
7. It returns evidence. The final prompt when requested, provider, model, output paths, failures, and visual review notes are reported.

Control privacy, rights, cost, and randomness

• npx can download and execute the npm-latest rendering package; generation uses network, provider quota, compute, and potentially billable requests.
• --mode all makes three provider calls. Failed or timed-out requests may still consume provider resources; check the provider account before retrying.
• Wild mode is random, and provider generation is not guaranteed reproducible. Preserve selected files and the prompt rather than assuming a rerun matches.
• Generated text, logos, anatomy, factual diagrams, and exact brand elements can be wrong. Inspect the pixels; prompt success is not asset approval.
• Validate usage rights, likeness consent, trademarks, and required disclosure before publishing. The Skill does not grant rights to source prompts or output.
• Generation writes files but does not publish, upload, commit, or deploy them unless a separate authorized workflow performs that action.

Verify the output

A complete result should include:

• The validated brief, selected mode, aspect ratio, size, provider, and model.
• The final prompt or a safe summary, including exact requested text and negative constraints.
• Every output path and the relationship between each file and its mode.
• Visual checks for subject, composition, text, unwanted elements, artifacts, and brand constraints.
• Provider errors, timeouts, missing credentials, cost-sensitive retries, and any rights or privacy review still needed.

Stable and beta ship identical prompt data, scripts, and ak:ai-artist source. No directly scoped release test proves provider availability, output quality, cross-runtime process parity, or deterministic pixels.

Troubleshoot and interpret limits