Is there a Claude Code skill that keeps my images on brand?

Yes, Wireflow's brand-consistent-images skill turns your logo, product photo and palette into on-brand images inside Claude Code. It builds a short brand brief, writes each image request so the logo and product stay as given, shows the credit cost and waits for your yes, then renders square images with Nano Banana Lite and saves them to ./wireflow-outputs/. The same brief goes into every run of a set, which is what keeps image five looking like image one.

Claude does the planning and checks the results, and a published Wireflow app renders each image, reading up to three of your brand images as references. Claude reaches that app through the Wireflow MCP connector. Without the connector, the skill still writes your brand brief and three to six image requests that you can paste into any image tool that accepts reference images. The skill's files, including its ready-made requests, are in the brand-consistent-images folder on GitHub.

How do I install the brand image skill in Claude Code?

Install the Wireflow plugin, which adds every Wireflow skill and the Wireflow connector:

/plugin marketplace add wireflowINC/skills
/plugin install wireflow-skills@wireflow

Claude Code shows the plugin's details first, so you can review it and choose its scope. Then run /mcp, choose wireflow and sign in with your Wireflow account. The skill shows up as /wireflow-skills:brand-consistent-images, and Claude also picks it up from plain requests.

To add only this skill, use the skills CLI, or copy the skills/brand-consistent-images folder into ~/.claude/skills/ (all projects) or .claude/skills/ (one project). Then connect Wireflow yourself and run /mcp to sign in. The connector uses OAuth, so there is no API key to copy.

npx skills add wireflowINC/skills --skill brand-consistent-images
claude mcp add --transport http wireflow https://www.wireflow.ai/api/mcp

To check the setup, ask for an image. The skill first looks for the Wireflow tools (run_app, get_execution, get_credit_balance, whoami, create_upload_link and get_upload_link). If they are missing, it says so and shows you how to connect.

What do I give Claude to make on-brand images?

Give Claude your logo, plus a product photo and one style reference if you have them, and say what each image is for. You can ask in plain words for on-brand images, a consistent set for social posts, ads or a campaign, or your logo on a product shot:

Make four on-brand Instagram posts for my tea brand. Our logo and a photo of
our tin are on my laptop. Show the tin on a kitchen counter in soft morning
light, with the words "Steep slowly" on a small card next to it.

Claude turns that into the inputs of the skill's Wireflow app:

Input	What goes in
brand_image_1	Your logo, or the single most important brand image. It goes on every run
brand_image_2	Optional: a product photo, packaging, or a second logo version
brand_image_3	Optional: a palette card, a brand guideline page, or a past image whose style should carry over
brand_brief	Palette, light, surfaces, people rule, mood and what to avoid, up to 2,000 characters. The same text goes on every run of a set
image_request	This image only: use, scene, where the logo appears and any exact words in quotes, up to 2,000 characters

Every run needs the logo and a request. Without either, the app uses its own saved demo, an invented candle brand called Tavrin, and you pay for someone else's picture. Without a brief, it defaults to "match the references". The two optional image slots can stay empty: the skill's test showed an empty slot runs cleanly.

For the logo, the skill asks for a clean PNG or JPG on a plain background, as large as you have. For a product photo, it asks for one product, whole and in focus, on a plain background in even light. A blank label is fine, because the request can put the logo on it. Three images is the app's limit, so pick the single most useful style reference.

Each brand image has to reach Wireflow as an https:// URL. For a file that is already online, give the direct image link, one that opens as an image rather than a web page. For a file on your computer or phone, Claude creates an upload link through the connector: upload the file there (no login needed), tell Claude, and it gets the file's URL back. With no upload tools, Claude asks you for a public direct link.

From what you tell it, or from your website or a guideline image, Claude writes the brand brief once: palette as names plus hex codes, light, surfaces and materials, the people rule, mood words and an avoid list. It keeps the brief under about 600 characters and shows it back for a quick yes or edit.

Then Claude writes one request per image. Each names what the image is for (an Instagram post, ad, email header, website tile or product page), the scene (setting, surface, two or three props at most, light and camera angle), where the logo sits, and any exact words in double quotes, where short lines of two to four words work best. Claude checks the physics before anything runs, for example that a lit candle needs its lid off, and asks you when unsure. Ready-made requests cover common uses: social post, flat lay, ad backdrop, seasonal, packaging close-up and lifestyle in use.

Use only a brand you own or work for. Claude confirms the brand is yours or a client's, and keeps other companies' logos, products and trademarks out of the requests.

For plain product scenes with no logo or brand references, use the product-photography skill instead. For images from text alone, use ai-image-generation.

Does it keep my logo and product as they are?

It is built to: a fixed rule block inside the skill's app tells the model to copy the logo exactly, keep any product as photographed and add no words beyond the ones your request quotes. Results can still drift, so Claude looks at each image next to your brand images before calling it done:

• Logo: the same mark, the same letters in the same order, the same colors. Small taglines under a wordmark can come out soft.
• Product: the same shape, proportions, material and label area.
• Palette and light: in line with the brief, with no stray color creeping in.
• Quoted words: spelled exactly, with nothing extra added.

Claude names any drift plainly, for example "the tagline under the logo is blurry" or "the jar is a little slimmer than yours". A redo costs 16 credits and needs its own yes.

The logo is not a watermark: the model paints it into the scene, so each request says where it sits physically, such as on the product label, a sign, a bag or a card in the scene. Requests also call your product "the product from the product photo" rather than giving it a new name. The skill's guardrails rule out invented claims, awards, prices, reviews or certifications in any image, and real people's faces taken from elsewhere.

What do I get back from a run?

Each run gives you one square image, 1024x1024, in about 15 seconds. Claude saves it in a dated folder under ./wireflow-outputs/ as brand- plus a short name, keeping the file extension from the image's CDN link. It finishes with the file paths, the CDN links and the credits used, worked out as your balance before minus your balance after.

For each image you approved, Claude calls run_app, then polls get_execution every 5 seconds while the run is PENDING, RUNNING or IN_PROGRESS. If a run ends FAILED, CANCELLED or TIMEOUT, Claude reports the exact error for that image and carries on with the others only if you agree.

The ratio is fixed at 1:1 because Nano Banana Lite's ratio is a setting the app cannot expose. If you need a 9:16 story or a 16:9 banner, say so up front: you can crop the square, or open the board in Wireflow from the link the skill offers and change the ratio there. The board is also where you can add an upscaler or swap the model. For ad variations in 1:1, 4:5 and 9:16 at once, with words spelled exactly, see the ai-ad-creative-generator skill.

When it reports back, Claude also offers more images with the same brief and brand images, with one quote for the batch.

Images generated through Wireflow are yours to use for personal and commercial purposes, subject to the underlying AI model provider's terms, as the Terms of Service say. The Privacy Policy says Wireflow does not use your prompts, images or generated content to train AI models.

How many credits does one image cost?

One image costs 16 credits, according to the skill's test run on 15 October 2026 and the Wireflow price list.

What	Credits
One image, square, 1024x1024	16 credits
A redo of one image	16 credits, only after your yes
A set of 4 images	64 credits (4 images x 16 credits)

Before anything runs, Claude checks your account and balance with whoami and get_credit_balance, then shows the list of images and the total in one line, for example "4 images x 16 credits = 64 credits. Your balance: 1,240." If your balance is short, it stops there. Nothing runs until you say yes to that list at that price.

Generating needs a Wireflow account with credits: see pricing for what each plan includes.

Does the brand image skill work in Cursor, Codex or Claude Desktop?

Yes: the skill is made for Claude Code and Claude, and the skills CLI also installs it in Cursor, Codex, Gemini CLI and other agents that read SKILL.md. Each agent must be connected to the Wireflow MCP server before the skill can render anything, and Wireflow has not tested its skills in other agents itself.

In Claude.ai or Claude Desktop, add https://www.wireflow.ai/api/mcp as a custom connector. Generation needs that connector and a Wireflow account with credits, and saving files needs a shell.

Connection steps for Cursor, Codex and every other client are in Use Wireflow from your agent. Wireflow's other skills are on the agent skills page, and to call Wireflow from your own code instead of an agent, see the run API reference.