Grafiki i miniatury za grosze? Claude Code to potrafi (poradnik dla nietechnicznych)

Main idea / problem addressed

• Creating graphics (thumbnails, infographics, posters) is often expensive when outsourced (typically ~100–200 PLN per thumbnail; infographic designer costs ~200–400 PLN per piece).
• The video demonstrates how to build an AI-driven “agent” using Claude Code that can generate visuals cheaply by combining Claude Code with image-generation models.

Proposed solution: Claude Code + image models

Build an AI agent in Claude Code that:

1. Sends requests to an image/graphics model (primarily Google Nano Banana; optionally OpenAI Image 2 / GPT).
2. Learns how to prompt the model effectively—using structured tool requests rather than only generic prompting.
3. Applies a consistent style for outputs (matching colors/style across infographics).

What the video shows you can generate

• YouTube thumbnails, including ones based on the creator’s face/photo.
  • Claim: newer models handle faces better, reducing “face mismatch” issues common in some image generators.
• Infographics derived from lesson text
  • Upload text → get an infographic in a consistent style.
  • Iterate: adjust prompt/text if something doesn’t fit → regenerate a corrected result.

Pricing / cost analysis (claims from subtitles)

• Very low per-image cost:
  • Example stated: about 50 groszy up to roughly 1 PLN per generated unit.
  • Later example: ~16 groszy per image (calculated from a “0.04 cents for pro” / per-image estimate).
• Emphasis: still far cheaper than outsourcing a graphic designer.

Workflow tutorial (step-by-step concept)

1. Choose the model(s)
   • The speaker suggests Gemini alone may be enough (though multiple Nano Banana variants are discussed later).
2. Use official Nano Banana documentation as the configuration source
   • Copy Nano Banana documentation from Google API docs into Claude Code.
   • Claude Code uses it to create a skill for generating images/infographics.
3. Create project files/folders in Claude Code
   • Create a new folder like graphics in the repository.
   • Add an API key file (e.g., an .env-style file) so Claude Code can call the model.
4. Obtain and store API key
   • Go to Google AI Studio → API Keys → create key.
   • Treat the key as secret (don’t share it).
5. Run in “planning mode” / safe automation mode
   • The video describes auto/planning behavior where Claude Code asks for approval only when needed, using a “classifier” to decide whether approval is required.
   • The agent can run long tasks (minutes) while the user does other work.
6. Connect billing/card
   • Error example: “no card connected to this API”.
   • Fix: add a payment method; generation costs will be charged during requests.

Quality / styling and prompt control

• The agent can:
  • Edit/rewrite prompts.
  • Generate outputs with specific style descriptors (e.g., “highly detailed digital illustration, sci‑fi movie poster style”).
  • Support different sizes/resolutions (example: 16:9 mentioned).
• Tip: better resolution may improve quality; subtitles claim price may be similar between certain sizes (1K vs 2K) for Nano Banana.

Boosting prompt quality with a large template repository

• Before generating more (to avoid wasting money), the video recommends installing an additional skill from a repo:
  • ~15,000 prompt templates (repo called something like “osome nano banana pro prompts”).
• Claude Code can automatically select the best template based on your description, simplifying prompting.
• Combine:
  • the template-skill
  • with your custom graphics skill
• Then ask Claude Code to improve the skill so it uses templates.

Use cases / who benefits

• Primarily creators for:
  • YouTube, Instagram, LinkedIn, and business marketing materials.
• Also helpful for businesses that need many visuals:
  • Example claim: generating offer visualizations/graphics can save ~1000–3000 PLN vs normal outsourcing.
• Even if you have a graphic designer:
  • AI-generated visuals/prototypes can serve as a starting point for a human to refine.

Main speaker / sources

• Robert Szewczyk (main presenter/partner)
• Narrator/host of the “Claude Code Guys” series (speaks alongside Robert)

Sources mentioned

• Google Nano Banana documentation (used to configure the skill)
• Google AI Studio (API keys)
• Repo of Nano Banana prompt templates (“~15,000 prompt templates”)
• operatorzyai.pl community (mentioned as a place to join/exchange Claude AI assistant skills)