How to Turn Any Photo Into a Cartoon Character You Can Reuse Everywhere

Ask an AI tool for a cartoon version of yourself twice and you get two different faces. This free Claude Code skill fixes that by locking your character once, so every image after uses the same face, hair and colours.

The short answer: download the free ip_illustration_for_yourself skill from GitHub, add it to Claude Code, and upload one clear photo. Claude locks your character's face, hair and colours, then writes a full image prompt for whatever scene you describe. Paste that prompt into any image tool, GPT Image 2, Google Flow or similar, to get the picture. No image credits are spent inside Claude Code itself.

Skill link: https://github.com/EverettFish/ip_illustration_for_yourself/commit/ca146e31127a52be6a2a71f097237fd3442f0545

A character that stays the same is worth more than one cute picture. Brands reuse a single mascot for years, and creators build entire pages around one recognisable face. Doing this by hand means briefing an illustrator for every new pose. Doing it with a plain AI prompt usually means a new face each time you ask.

What this skill actually builds

This is not a built-in Claude feature. It is a free, independently built skill shared on GitHub, and you add it to Claude Code yourself. A skill is just a folder with one required file, SKILL.md, that gives Claude a fixed set of instructions for one job, here building a small illustrated character and keeping it identical across every image you ask for afterwards.

On GitHub it goes by the name ip_illustration_for_yourself. IP here just means your own little intellectual property, basically a personal mascot. The look it aims for is a specific one: a mini, hand-drawn character with a deliberately rough, slightly wobbly pen outline, simple dot eyes and a lot of white space around it, closer to a doodle in a notebook margin than a polished vector logo.

The skill was built to work across Claude Code, Codex and a few other AI agents, so the exact file names inside it can shift as the developer updates it. What stays constant is the job: lock one character, then generate scene after scene without the face quietly changing.

Add the skill to Claude Code

Setting this up takes a few minutes and no coding knowledge.

1. Download the skill. Go to the ip_illustration_for_yourself repository on GitHub and download the files, either as a ZIP or with git clone.

2. Move the folder into your skills folder. On Mac or Linux, that is ~/.claude/skills/. On Windows, it is C:\Users\<your name>\.claude\skills\. Make sure SKILL.md sits directly inside the skill's own folder, not nested one level too deep.

3. Restart Claude Code. Skills are only picked up at the start of a session, so close and reopen Claude Code after adding the folder.

4. Check that it loaded. Ask Claude what skills are available, or type / to see it in the list. If it does not show up, double check the folder location and that the file is named exactly SKILL.md.

Build your character from one photo

Open a Claude Code session and upload one clear photo, your own face, a pet, or any character you want turned into an illustrated version. Ask Claude to build the character using the skill.

Claude studies the photo and locks down what it calls the identity: hairstyle, face shape, eye colour, outfit, accessories and palette, whatever it can actually see. This locked description is what keeps the character from drifting in later images. Skip this step and ask for finished scenes right away, and each new picture will quietly reinvent the face.

Get the image without spending API credits

Claude Code cannot draw the picture on its own; it has no built-in image generator. When you describe a scene, the skill's own instructions kick in and it writes out the complete image prompt instead: character description, scene, style rules and all, ready to hand over.

That is the whole point of asking Claude to write the prompt rather than the image: it takes a few seconds inside Claude Code and uses no image generation credits at all. Copy that finished prompt and paste it into whichever image tool you already use. The skill defaults to GPT Image 2, but the same prompt works just as well with Google Flow's Nano Banana Pro, Higgsfield or any other tool that takes a reference photo and a text prompt together.

Where to use the character

WhatsApp sticker packs. Ask for a set of expressions, happy, laughing, confused, tired, and you get the same character reacting differently across every sticker instead of a new face each time.

A recurring mascot for videos. Explainer videos and reels read as more polished when one character shows up scene after scene, rather than a new AI generated face in every clip.

Consistent illustrations across a series. Carousel posts, blog headers or a course's slide deck all look like they belong together once one character is doing the illustrating throughout.

Common mistakes when building an AI mascot in 2026

Switching reference photos midway. Uploading a different photo every time you regenerate slowly drifts the character away from the original. Stick to the same accepted reference image once the character is locked.

Skipping straight to scenes. Asking for a finished illustration before the character anchor is built usually gets you a decent looking but unlocked face, one that will not match the next image.

Accepting a too smooth outline. This particular style is meant to look roughly hand drawn. A slick, glossy or perfectly clean line is considered off style by the skill's own rules and worth asking Claude to redraw.

Forgetting to say what the picture is for. A sticker, an article illustration and a folder icon all use different proportions and framing. Mention the end use upfront so the prompt Claude writes actually matches it.

FAQ

Is this an official Claude feature?

No. It is a free, independently built skill published on GitHub, and SKILL.md is plain text, so you can read exactly what it tells Claude to do before you install it.

Does using this skill cost anything?

The skill itself is free. Whatever you spend happens outside Claude Code, in whichever image tool you paste the finished prompt into.

Can I use a photo of someone other than myself?

Yes, any clear reference photo works. If you are turning a friend or a client into a public character or sticker pack, get their go ahead first.

What if the picture comes out too smooth or polished?

Tell Claude the outline is too smooth and ask it to redraw with a rougher, more hand drawn line. That repair is actually built into the skill's own instructions.

Does this only work in Claude Code?

The same skill format also works in a few other AI coding assistants. Check the GitHub repository for the current list, since it gets updated as new agents add support.

In short

One photo, a free skill and a Claude Code session are enough to lock an illustrated character once and keep reusing it. Claude never draws the picture itself, it hands you a finished prompt, and that prompt is what keeps the character consistent from your first sticker to your fiftieth reel.