Flyer Maker

This skill orchestrates three lower-level components:

1. image-prompt style anchor + composition-zone vocabulary
2. common/runners execute layer — multi-aspect batch generation via the chosen image provider
3. common/style-library/carousel/ — reuses the 24 carousel visual styles as the visual vocabulary (poster styles overlap heavily with editorial carousel styles)

Use when the user wants a finished flyer image — not just a prompt. Without --execute, returns the assembled prompts; with --execute, generates the actual PNGs.

This skill does NOT:

• Generate animated flyers (use reel-builder for motion)
• Render print-ready CMYK / 300DPI files — outputs are RGB at standard digital resolutions (1080 / 1240 px). For print: open the PNG in a real DTP tool (Affinity / InDesign / Photoshop) and convert.
• Compose multi-image collages — single image per aspect.
• Generate QR codes (deferred to v2.5.x — if you need a QR, generate separately and overlay manually).
• Search Eventbrite / Meetup / Luma for event metadata — you provide title / date / location explicitly.
• Replace a designer's judgment on layout — the composition zones are starter conventions; tweak via prompt iteration.

ROLE

Read the event details + optional photo + optional style → pick a model (text-friendly + multi-ref-capable if photo provided) → assemble per-aspect prompts with composition zones filled in → batch execute via the image provider → save PNGs + manifest → print paths.

PIPELINE

1. Resolve event details:

   • Required: --title
   • Strongly recommended: at least one of --date, --location, or --cta (a flyer without a single detail is just a styled portrait)
   • Optional: --subtitle, --time (can be merged with date)

2. Resolve photo (optional):

   • --photo <path-or-url>: passed as reference image to the chosen model.
   • Multi-ref capable models needed (Nano Banana Pro / Flux Kontext / gpt-image-2 / Ideogram 3 / Seedream 5.0).
   • If --photo provided AND --model auto: skill picks a ref-capable model.
   • If --photo provided AND --model <slug> that can't handle refs: warn + auto-substitute, or exit if user passed --strict.

3. Resolve style — see references/event-details.md:

   • --style <library-id>: load from common/style-library/carousel/<id>.md. Use the Style anchor (text-in-image mode) block (flyers always have text).
   • --style auto: pick from library based on the event type implied by the title + the chosen tone (formal, casual, indie, corporate). Algorithm same as carousel-builder.
   • --style-mod "<override>": append a tweak to the anchor.

4. Pick model — see references/model-picker.md:

   • --model auto:
     • Photo provided + identity preserve critical (named speaker) → nano-banana-pro
     • Photo provided + brand-style transfer needed → flux-2-pro or seedream-5
     • Photo provided + Latin/CJK text needed → gpt-image-2
     • No photo + text-heavy → ideogram-3-quality
     • No photo + photoreal style → nano-banana-pro or flux-2-pro
   • --model <slug>: explicit override.
   • ONE model for all aspects (consistency across the family).

5. Build per-aspect prompts — see references/composition-zones.md: Each aspect has a composition template with three zones:

   • Headline zone (top 20-25%): title in large bold typography
   • Visual zone (middle 50-55%): photo integration if provided, else style anchor visual
   • Details zone (bottom 25-30%): date + location + CTA in smaller typography

   The prompt assembly per aspect:

   <style anchor (text-in-image mode)>
   
   Composition: <aspect-specific zone template>
   
   Embed text:
     Headline: "<TITLE>"
     <Subtitle if present>
     <Date if present>
     <Location if present>
     <CTA if present>
   
   Aspect ratio: <portrait|square|story|landscape|a4>
   Size: <pixel dimensions>
   

6. Estimate cost + confirm — sum aspects × variants. Under default carousel budget ($1.50) → no prompt; over → ask once.

7. Batch execute — common.runners.batch.run_batch():

   • Parallelism default 2 (each aspect = ~5-10s API call; running 3 in parallel rarely problematic but rate-limit safety)
   • Manifest updated after each aspect
   • --resume retries only failed aspects

8. Output:

   ./generated/flyer/<event-slug>/
     portrait.png            (1080×1350)
     square.png              (1080×1080)
     story.png               (1080×1920)
     landscape.png           (1920×1080, if requested)
     a4.png                  (1240×1754, if requested)
     manifest.json
     style-used.md
     prompts.md
   

   stdout last lines:

   Flyer: ./generated/flyer/<event-slug>/  (N/M aspects succeeded)
   Files: portrait.png · square.png · story.png
   

MODES

Required

• flyer-maker --title "<text>"

Strongly recommended (provide ≥1)

• --date "<text>" — free-form date / time
• --location "<text>"
• --cta "<text>" — call to action (e.g., "Tickets: link in bio", "RSVP via DM", "Free entry")

Optional content

• --subtitle "<text>" — secondary line under the title
• --photo <path-or-url> — reference image to embed (speaker face, venue shot, theme image)
• --lang en|ru — language of embedded text (default: matches the language of --title)

Visual

• --style auto|<library-id> — visual style (uses carousel style library; default auto)
• --style-mod "<override>" — append a tweak to the chosen anchor
• --aspects portrait,square,story,landscape,a4 — comma list (default: portrait,square,story)
• --variants N — N takes per aspect (default 1)
• --model auto|<slug> — image provider (default auto)

Execution

• --execute — actually generate (else returns plan + prompts)
• --output <dir> — custom output (default ./generated/flyer/<event-slug>/)
• --parallelism N — concurrent API calls (default 2, max 4)
• --yes — skip cost confirmation
• --resume — retry only failed aspects
• --strict — fail (don't auto-substitute) if --model <slug> can't handle --photo

Inspection

• --prompts-only — print plan + per-aspect prompts, exit (no API call)
• --cost-only — print total cost, exit
• --check — validate env vars + library files exist

STRUCTURED LAYOUT (Ideogram 4 only)

When --model is an ideogram-4* tier, emit the layout as structure instead of flattening it into a sentence: add json_prompt to each plan item's kwargs, following common/references/json-prompt.md. This skill already knows where every element goes; sending that directly is what stops the arrangement drifting between variants.

On any v3 tier the provider refuses a json_prompt rather than ignoring it, so keep the prose prompt as written and send nothing extra.

REFERENCES (load on demand)

EXAMPLES

See examples/before-after.md — 3 calibration runs: solo founder workshop with embedded speaker photo (Nano Banana Pro), conference poster without photo using bold typographic Swiss style (Ideogram 3 Quality), three-aspect launch event with Kinfolk-minimal aesthetic + RU+EN bilingual variant (gpt-image-2).

CONSTRAINTS

• Embedded text on image is the differentiator. Without text rendering, this is just a styled image — use image-prompt directly. The skill defaults to text-friendly models because the title + details MUST land readable.

• Headlines ≤8 words. Past 8 words, even the best text-in-image models distort. The validator warns; the prompt assembly trims if longer.

• One model for all aspects. Mixing providers across the portrait / square / story batch breaks visual cohesion. Lock per-flyer.

• One style for all aspects. Same anchor across the batch.

• Photo identity preserve uses Nano Banana Pro by default. For speaker portraits in a flyer, NBP is the safest pick. Override with --model gpt-image-2 if you need text + photo blended in a stylized layout.

• Aspect compositions differ. A 1080×1350 portrait and a 1080×1920 story have different headline/details zones — see references/composition-zones.md. Don't assume the prompt is identical across aspects.

• Cost confirm ONCE per batch. Sum across all aspects + variants.

• --prompts-only is the safety dry-run. Always recommend it for first iteration — flyers iterate on copy + composition more than on style.

• Manifest updates after every aspect. --resume retries only failed ones.

• No QR codes in v1. If you need a QR (ticketing link, venue map), generate separately and overlay in a real image editor.

• No print-ready output in v1. Outputs are RGB PNG at digital resolutions. For print: open in a DTP tool and convert.

• Event-slug derived from title. Default: kebab-case of title, max 40 chars + date suffix. Override via --output <dir>.

• --photo accepts local paths AND URLs. Provider routes accordingly.

• Never print API keys. Mask in errors.

• Output dir is ./generated/flyer/<event-slug>/ by default.

INVOCATION HINTS

When the user says any of:

• "make a flyer for X", "event poster for Y", "workshop flyer"
• "meetup poster", "concert poster", "exhibition flyer"
• "use this photo on a flyer", "put my face on a workshop poster"
• "flyer with embedded date and location"
• "промо для мероприятия", "афиша", "плакат", "флайер"
• "сделай афишу для воркшопа / встречи / митапа"
• "афиша с моей фоткой"

Defaults: --aspects portrait,square,story --style auto --model auto. Without --execute, returns prompts + plan; with --execute, generates files.

If the user mentions a specific platform (Instagram → emphasize 4:5 portrait + 9:16 story; LinkedIn → emphasize 1:1 square + 1.91:1 landscape; print → suggest A4 aspect), bias --aspects accordingly.

This skill is distinct from:

• image-prompt — that writes prompts for any image; this orchestrates a multi-aspect batch for flyers specifically
• carousel-builder — that's N-slide narrative; this is 1-image-per-aspect event posters
• reel-builder — that's video; this is static images

For album / book / podcast covers: use image-prompt --execute with a simpler one-shot prompt (no event-details structure). A dedicated cover-maker skill is planned (see ROADMAP).

For YouTube thumbnails: image-prompt --execute with 16:9 size + face + bold title. A dedicated thumbnail-maker skill is planned.