A skill that makes Claude act as the character lead on a set of AI images and clips, so one person, mascot or product character stays recognisably the same in every image. It starts with what the character is for and who has the right to its face. It writes a character sheet of fixed traits with numbers (age range, hair parting, skin and outfit colours in hex) and a paragraph that goes into every prompt word for word. Then it plans one master image and a frozen set of reference views, sets up the character feature in your tool, and checks every result you send back.
Inside are five reference files. A tool file covers the character features of Midjourney, GPT Image, Nano Banana, FLUX, Ideogram, Firefly, Runway, Kling, Veo and Flow as checked on 18 October 2026, with their reference limits and sources. A rights file sets out what a real person's consent must cover, the resemblance check for invented faces and the US disclosure rules, including New York's law on AI-generated performers in ads. A guide to the sheet, a drift-check guide with face ratios and colour tolerances, and a worked example for a fictional coffee brand complete the set. Claude will not use a real person's likeness without written consent, flip an approved image, or reuse a drifted image as a reference.
Claude cannot make images or watch video. You run every prompt in your own tool and send back the downloaded files. For clips, it asks for still frames at set points and checks those. Generators still drift, especially on freckles, small logos and faces in wide shots, so the skill plans for repairs instead of promising perfection. Turn on code execution so Claude can sample colours, measure face ratios and build a lineup of the whole set. Start with soft light and moderate expressions, and push the character further once it holds. The rights notes are practical guidance for a creative team, not legal advice.
Claude skill · version 1.0
AI character consistency
Checking your access.
In the download · 7 files
- SKILL.mdThe method Claude follows, in six stages: brief and rights, character sheet, master and references, tool setup, prompts and continuity, drift check.
- README.mdHow to install it in claude.ai or Claude Code, how to start, and the licence.
- references/worked-example.mdAn invented roaster for a fictional coffee brand taken through all six stages in Nano Banana Pro and Veo 3.1, with real faults and fixes, plus a short mascot example.
- references/consent-and-rights.mdThe four character types, what a real person's consent must cover, the resemblance check, children, US likeness laws, disclosure rules, platform labels and ownership, with sources and dates.
- references/character-sheet.mdHow to write locks and flex, signatures, the anchor paragraph and the wardrobe list, and how to make the master and the frozen reference set, with a blank sheet.
- references/tool-features.mdCharacter reference features of Midjourney, GPT Image, Nano Banana, FLUX, Ideogram, Firefly, Runway, Kling, Veo, Omni Flash and Flow as checked on 2026-10-18, with training and licence notes and sources.
Install it and begin
01Install it in Claude CodeTerminal mkdir -p ~/.claude/skillsunzip ~/Downloads/ai-character-consistency.zip -d ~/.claude/skills/# Start it with /ai-character-consistency. If this is your first skill, type /reload-skills once.On claude.ai there is nothing to unzip: upload the zip as it is. Every step is on Install and use a Claude skill.
02Your first message to ClaudeText Use the ai character consistency skill. My character is [WHO OR WHAT THE CHARACTER IS]. I need [HOW MANY IMAGES AND CLIPS] for [WHERE THEY WILL RUN]. I use [YOUR TOOLS AND PLANS]. The character is [INVENTED, OR A REAL PERSON WHO SIGNED CONSENT]. Start with the brief.Attach any image or drawing of the character you already like, your brand colours, and real product photos if a product appears. For a real person, have their signed consent to hand before you upload a single photo.
Try this
Hold one character across six images
Invent one character, a person or a mascot, and take it through the skill to an approved master, a four-view reference set and six scene images made in your own tool. Then put all six beside the master and run the drift check and the lineup.
Time: About two hours. You need: An image tool that takes reference images, such as Nano Banana, Midjourney, Ideogram or FLUX, a one-paragraph idea for an invented character, not based on any real person, Code execution switched on in Claude, for the colour checks and the lineup sheet.
Look for
- The anchor paragraph pasted word for word into every prompt, with only the scene lines changing.
- The scar, parting or pocket on the same side of the body in all six images.
- A face that still reads inside its age range, not drifting younger or smoother.
- Wide shots where the face holds, or a second pass that repaired it.
- Outfit colours within tolerance in neutral light, and in the right relationship in scene light.
- A lineup where image six matches the master, not only image five.
What to do it with
- ClaudeFree tier, paid above it
A language model that writes, edits, reads long documents and runs tools, including agents that work inside a codebase.
Bad at It is agreeable by default, so it will improve a bad idea instead of telling you it is bad. Ask it to argue against you and the quality of the answer changes.
- Claude CodeSubscription
An agent that reads a whole repository, edits files, runs the tests and explains what it changed.
Bad at It is fast enough to make a large mistake quickly. Work on a branch, read the diff, and never let it push. Watch the starting mode: on Pro, Max and Team plans it opens in auto mode, which Anthropic's own list says may read a .env file and send the key in it to the API that key belongs to. Switch it off.
- MidjourneySubscription
Generates images from a written description, with the strongest default aesthetic of anything in the category.
Bad at That default aesthetic is the problem as often as it is the draw. It has a house look, it pulls everything towards it, and getting a plain honest photograph of a real product out of it is harder than it should be. Text inside an image is unreliable.
- GPT ImagePay per use
Generates and edits images with unusually good instruction-following, and can hold text inside a picture better than most.
Bad at Faces drift between generations unless you feed a reference every time. Skin can come out plasticky at default settings, and a heavy fidelity setting costs many times more and takes many times longer.
- ChatGPTFree tier, paid above it
General assistant with image generation, data analysis and web access in one window.
Bad at Its default voice has become recognisable enough that readers now spot it, which is a brand problem if you ship it unedited.
- GeminiFree tier, paid above it
Google's assistant, with a very long context window and tight links to their other products.
Bad at Behaviour changes between the app, the API and the versions inside other Google products.
- Google AI StudioFree tier, paid above it
Where Gemini API keys, projects, billing and spend caps are made, with a playground for the models.
Bad at Free-tier prompts are used to improve Google's products, and the billing screens changed several times in 2026.
- FLUXOpen weights or open source
A family of image models, several with open weights, known for prompt adherence and clean type.
Bad at The open versions need a machine and a person who is comfortable with one. Licence terms differ between the variants, so check which one you are running before it goes anywhere near a client.
- IdeogramFree tier, paid above it
An image model built around getting words right inside the picture.
Bad at Strong at type, weaker at photographic realism. A poster comes out well, a product on a plinth less so.
- Adobe FireflyFree tier, paid above it
Adobe's generative models, trained on licensed and public-domain material and wired into Photoshop and Illustrator.
Bad at Raw image quality trails the leaders. What you are buying is the indemnity and the fact that it sits inside files you already have.
- RunwayFree tier, paid above it
A video generation and editing suite with motion brush, camera control and per-frame steering.
Bad at Credit burn is fast and the good controls sit on the higher tiers. Output length still suits cuts rather than scenes.
- KlingFree tier, paid above it
Image-to-video with unusually stable human motion.
Bad at Queue times vary by a lot, and the interface assumes you already know what its settings mean.
- VeoPay per use
Text or image to video, with sound, at a length and coherence nothing had two years ago.
Bad at Cost per second is high enough to change how you storyboard. Long shots still drift, and hands and logos wander. Through the Gemini API on 25 September 2026 it refused a last frame on every tier, though the docs describe one.
- Google FlowFree tier, paid above it
A web studio for making and extending clips with Veo 3.1 and Gemini Omni, paid for in credits from a Google AI plan.
Bad at Credits do not roll over, features differ from the API (Quality takes no references), and Google's own help pages disagree on who can use it.
Why it matters
Ask a chat to keep a character consistent and it adds "same woman as before" to the prompt, then compares each new image with the last one. Over twenty images the face drifts younger, the scar changes sides and the jacket changes colour. This skill makes Claude fix the character in writing and in frozen references first, check every result against the master instead of the previous image, and refuse any face nobody has the right to use.
- Type
- technique
- Period
- ongoing