fal-nano-banana-2-image-gen
fal-nano-banana-2-image-gen
Required inputs
--prompt "..."(required)workers.jsoncwithfal.key(required)
Optional inputs
- Output sizes / aspect ratios:
--image-size <preset>(repeatable) or--sizes a,b,c(comma-separated presets)--size <WxH>(repeatable pixel-size hint; sent as{ width, height }when supported)
--negative-prompt,--seed,--num-images, and other model parameters:--param key=value(repeatable; supports dot-paths for nesting)--body-json path/to/body.json(merge arbitrary JSON into the request body)
Workflow
- Prepare inputs
- Write a prompt that is specific and testable:
- Subject: what the image is of.
- Composition: angle, crop, framing, background.
- Lighting + color: time of day, soft/hard light, mood.
- Style: photoreal, studio product shot, cinematic, etc.
- If you need to avoid certain elements (logos, text, hands, extra objects), use
--negative-prompt.
- Run
- Use the bundled script to call fal.ai.
- Final deliverables go to
artifacts/; request/response logs and intermediate files stay inlogs/.- Basic:
node skills/car-ads-designer/fal-nano-banana-2-image-gen/scripts/generate.mjs --prompt "..." --image-size square_hd
- Multiple sizes (one call, multiple requests):
node skills/car-ads-designer/fal-nano-banana-2-image-gen/scripts/generate.mjs --prompt "..." --image-size square_hd --image-size landscape_16_9
- Pass extra model parameters (from the model's
llms.txt/ docs):node skills/car-ads-designer/fal-nano-banana-2-image-gen/scripts/generate.mjs --prompt "..." --param guidance_scale=4.5 --param steps=30
- Basic:
- Stop and request review (required when subjective/high impact)
- After generation completes, do not take any further actions (no auto-retouching, no upscaling, no additional variants) until a human or designated reviewer confirms the outputs are acceptable.
- Review outputs
- Inspect the final images under
artifacts/...first. - If something looks off, use the working folder in
logs/...to debug (it includes request/response payloads).
Outputs
- Working folder under
logs/fal-nano-banana-2-image-gen/<run-id>/containing inputs and full request/response logs. - Final folder under
artifacts/fal-nano-banana-2-image-gen/<run-id>/containing:README.md(what was generated + where the working logs are)- One or more final images (and per-size subfolders when multiple sizes are requested)
Definition of done
- The script completes without errors and writes at least one image file under
artifacts/. - The output matches the prompt intent.
- A human (or designated agent reviewer) has reviewed the outputs and explicitly approved any next steps.
Safety / quality checklist
- Do not commit secrets.
- Do not paste/print keys from
workers.jsonc; only confirm whether values are present. - Avoid generating copyrighted logos/marks you don’t have rights to use.
- Treat rendered text as unreliable; avoid high-stakes text (prices, legal claims) unless you will manually verify/correct.
- Don’t include sensitive personal data in prompts unless you have permission and a clear need.
More from lingkaix/smartworkers
smart-skill-maker
Create or improve SmartWorkers-style skills from a workspace-local `skills/` source tree, then apply them to Codex with `npx skills`. Use when you want one guided tool for new skill creation, skill upgrades, or SmartWorkers skill-maintenance work while keeping repo conventions aligned.
22ads-info-gatherer
Use when the user needs a dealer car-ad source pack built from live web sources or a provided worker-pack folder, with VIN-anchored retail truth, source assets, and provenance preserved for downstream generation and review.
7ads-suite-pipeline
Run the full dealer car-ad production pipeline from source pack or approved anchor design to final approved deliverables: generate clean images, review/regenerate, build SVG copy overlays, and finish consistent multi-size or multi-model suites. Use whenever the user asks to produce a full ad suite, continue from an approved ad, adapt one campaign across sizes/models, review and improve generations, or keep retail offer copy exact while moving from source pack to deliverable assets.
7