The honest version up front: this is low-setup, not no-setup, and not free. Images cost about 8 cents each on fal.ai, so 100 shots is roughly $8. And AI still drifts on exact product details (logo text, labels, packaging), so you feed it your real product photo as a reference and you review the batch and keep the best. Done that way, it genuinely replaces a lot of studio time.
the stackDownload a free editor: VS Code or Cursor. (Google's new Antigravity works too, since it's a VS Code cousin, but it's Google's own Gemini IDE, so treat it as optional.) Then install the Claude Code extension: open Extensions, search "Claude Code," install, and sign in with your paid Claude account. Official steps.
Make a free fal.ai account (you get some free credits), create an API key, and keep it handy. In your terminal you'll set it once:
export FAL_KEY="your-fal-key-here"Easiest path (no script at all): add fal's official MCP so Claude Code can generate images directly. fal MCP docs.
claude mcp add --transport http fal https://mcp.fal.ai/mcpDrop these four files into your project. Claude reads them and does the rest. Copy them as a starting point and tweak the words for your brand.
CLAUDE.md (the project brief):
# Brand visual generator
You generate on-brand product photos for my brand.
## How to work
1. When I give you a brand URL, read the site and write a brand profile into style-analysis.md
(colours, mood, lighting, typography feel, product list).
2. When I give you reference images + a product + a count, write image prompts that match
the brand profile and the references, then run generate-visuals.py to make them on fal.ai.
3. Always feed my real product photo as a reference for product accuracy. Review before delivering.
## Rules
- Model: Nano Banana 2 on fal.ai (fal-ai/nano-banana-2).
- Never invent product details. Match the reference photo.
- Save outputs into /visuals with clear names.style-analysis.md (Claude fills this from your site):
# Brand style profile
Brand: [name]
Colours: [hexes]
Mood / vibe: [words]
Lighting: [soft / hard / natural]
Settings: [where products appear]
Products: [list with a reference image path each]SKILL.md (ties it together as a reusable skill):
---
name: brand-visuals
description: Generate on-brand product photos from a brand URL and reference images using Nano Banana 2 on fal.ai. Use when I ask for product visuals, shots, or brand images.
---
# Brand visuals
Follow CLAUDE.md. Build the brand profile, write prompts that match the references,
then call generate-visuals.py. Feed the real product photo as a reference. Review the batch.generate-visuals.py (Claude writes and runs this; here's the shape):
# pip install fal-client , then set FAL_KEY
import fal_client
# Claude fills PROMPTS from your brand profile + references
PROMPTS = [
"teal wool beanie on a mossy rock, soft morning light, editorial",
]
for i, prompt in enumerate(PROMPTS):
out = fal_client.subscribe("fal-ai/nano-banana-2",
arguments={"prompt": prompt, "num_images": 1})
print(i, out["images"][0]["url"])In Claude Code, paste your website URL and say "read my site and build the brand profile." Claude browses the site, fills in style-analysis.md, and pulls your product images. This is what makes every visual feel like your brand, not a stock look.
Gather 3 to 5 strong reference images of the style you want (lighting, mood, setting) from Pinterest or Cosmos (both free). Drop them in, then just ask:
Give me 30 shots of the teal beanie on mossy rocks, soft morning light,
using my product photo for accuracy.Claude writes the prompts and generates the images with Nano Banana 2. Then review the batch and keep the winners.