Skip to content

notque/vexjoy-agent / skills/content/image-gen

image-gen

Verified

AI image generation: Gemini and Nano Banana backends; single/series/batch workflows with prompt-to-disk.

image-gen

Backend-agnostic image generation: single images, series with anchor-chain consistency, batch pipelines. Two backends: Gemini (API) and Nano Banana (local scripts with post-processing).

Deep References

SignalLoadContent
Gemini backend selectedreferences/backends/gemini.mdModels, env vars, flag table, examples, error codes
Nano Banana needed (post-processing, series, JSON batch, style-match)references/backends/nano-banana.mdSubcommands, flags, aspect ratios, prompt patterns by asset type

Phase 1: Detect Backend

python3 skills/content/image-gen/scripts/detect-backend.py
  • gemini -- load references/backends/gemini.md
  • ask -- no key found; ask the user to set GEMINI_API_KEY

Gate: backend confirmed.

Phase 2: Write Prompt Files

Write all prompts to disk before any API call. Prompt files are the generation record and anchor-chain input for series.

File naming: single prompts/YYYY-MM-DD-{slug}.md, series prompts/{series-name}-01.md through -NN.md.

---
model: gemini-3-pro-image-preview
aspect-ratio: 1:1
flags: []
---

Full prompt text. Be explicit about subject, style, background, constraints.
mkdir -p prompts

For series: write ALL prompt files before calling any generation script.

Gate: all prompt files written and reviewed.

Phase 3: Select Script

Use caseScriptSubcommand
Single image, Geminiscripts/generate_image.py--prompt
Batch from text file, Geminiscripts/generate_image.py--batch
Single with post-processingscripts/nano-banana-generate.pygenerate
Style match from referencescripts/nano-banana-generate.pywith-reference
Batch from JSON manifestscripts/nano-banana-generate.pybatch
Series (anchor chain)scripts/nano-banana-generate.pygenerate then with-reference
Post-processing onlyscripts/nano-banana-process.pycrop / remove-bg / pipeline

Model selection:

ScenarioModel
Draft, testing, batch, cost-sensitivegemini-2.5-flash-image (2-5s)
Final asset, character art, typographygemini-3-pro-image-preview (~30s)

Aspect ratio by use case:

Asset typeRatio
Sprites, characters, icons1:1
Card art, landscape16:9
Vertical maps, portrait bg9:16
Portrait cards3:4
Wide banners21:9

Generate at the target ratio. Generating 1:1 and cropping to 16:9 loses 56% of pixels.

Gate: script and subcommand identified.

Phase 4: Generate

Use absolute paths for output files. Show full script output.

Anchor-Chain Algorithm (Series)

Character drift occurs when images are generated independently. Prevent it by passing the previous output as --reference:

image-01.png (no ref) -> image-02.png (ref=01) -> image-03.png (ref=02) -> ...
  1. Generate image 1 with no reference.
  2. Use output of image 1 as --reference for image 2.
  3. Continue: each image N references image N-1.

Save originals with --save-original for any batch or expensive generation. Re-processing a saved original is free; re-generating costs quota and may break the chain.

Gate: script exits 0.

Phase 5: Verify and Report

Read the generated image to verify:

  • Subject matches prompt
  • No unwanted watermarks or artifacts
  • Aspect ratio and framing correct
  • No excessive padding or dark borders

If inspection fails: regenerate with adjusted prompt. Report the issue before retrying.

Report: output file path (absolute), dimensions, model, post-processing applied, verification result. Report only what was requested.

Error Handling

ErrorCauseResolution
GEMINI_API_KEY not setMissing env varexport GEMINI_API_KEY=your_key
No image in responseSafety filter or text-only responseAdjust prompt; remove policy-adjacent content
Missing dependency: google-genaiPackage absentpip install google-genai pillow
Rate limit exceeded (429)Too many requestsIncrease --delay; script retries automatically
Content policy violation (400)Restricted contentRephrase using neutral language
Model not foundWrong model stringUse exact strings: gemini-2.5-flash-image or gemini-3-pro-image-preview
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 467kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 319kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 319kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 319kUpdated Claude CodeCodexVerifiedInstall