Character registry · cross-platform prompts
An on-chain ID and a history anyone can verify — so your character can be checked, transferred, or sold. Prompts for every platform come with it: use your character anywhere.
Google or GitHub · free credits included · no card
What this is
Most characters are not made here. They are made in Midjourney, in a 3D tool, in someone's own model — and then they sit in a folder. This is where one gets registered: an ID that does not change, a history of every version, and the wording that makes it come out the same on each platform you take it to.
Nothing gets rebuilt. Upload the images you already have.
How it works
More angles read more accurately. They are checked against each other first, so a stray image from a different character gets flagged before it becomes part of the spec. Face matching does not apply to non-human or heavily stylised characters — that is reported as not applicable, not as a low score.
The specification is extracted from the images rather than typed in. Where the images disagree — a shaved side that is on the left in one and the right in another — that is put in front of you as a question instead of being quietly averaged away. Those disagreements come first in the list, because they are the only ones nobody else can answer.
The ID is issued the moment the spec is confirmed; there is nothing else to apply for. Per-platform prompts come with it, and every later revision is registered as a new version rather than overwriting the old one.
The ID
A real record — the fingerprint links to the public check, no account needed. It covers the specification, not the images: adding a reference image does not create a new version; changing what the character is does.
Three separate claims sit behind that card, and they are not worth the same amount. The site keeps them apart on purpose.
| Layer | What it means | Who you have to trust |
|---|---|---|
| Registered | This specification was recorded here, with a timestamp | Us |
| Anchored | Its fingerprint was included in a batch written to a public chain | Us |
| Verified on chain | The transaction is read back and the proof path recomputed — the fingerprint really is inside that batch | Nobody |
Only the third line is worth anything against a platform that might be wrong, or gone. It is the one the public check performs, and it needs no account — any fingerprint can be looked up directly:
app.avatar-lab.app/api/verify/<fingerprint>
The precedent is a professional register, not a token. Taiwan's Ministry of Justice has published lawyers' practising certificates this way since January 2021 — each certificate's fingerprint anchored to a public chain, each lawyer given a code the public can check. No wallet, no coin, and the certificate cannot be traded. A character ID is the same shape.
What it proves, and what it does not. It proves this specification existed here, unchanged, from the date registered. It does not prove the holder authored it, that it infringes nobody, or that they are entitled to sell it. Copyright arises on creation; a timestamp is evidence, not title.
Why this reads correctly
Extracting a specification is the easy half. The hard half is the sentence that makes another platform reproduce it — and that sentence is different on every platform, in ways that are not documented anywhere and are frequently the opposite of what the settings suggest.
So the prompts a registered character travels with are not written from intuition. They are written from a measurement run, and these are four of its results.
| What you'd reasonably assume | What we measured |
|---|---|
| Turn reference strength downto get more scene freedom | Identity and scene adherence fall together. At the low end the output has no measurable relationship to the scene description at all. There is no trade to make. |
| Use a full-body referenceto get a full-body shot | The framing got tighter. Composition is driven by how much facial detail sits in the prompt, not by the reference image — and widening it costs 67% of face identity for 7% more scene adherence. |
| Train the character into the platformfor the strongest lock | For a stylised character, the trained version kept 0% of its declared markers — hair colour, face mark, collar palette all wrong. Plain reference images kept 100%. Verified on two platforms. |
| Higher similarity score is betterso pick the top of the table | The highest-scoring platform in our run had the lowest identity score. It reproduced the reference framing five times and ignored three of the five scenes. One number is not enough to tell those apart. |
None of this is in any vendor's documentation. Two of the four contradicted what we ourselves believed four days earlier.
What it is
Why portability has to come from outside the platforms. A platform has no reason to help a character leave. Trained character objects have no export field — that is not an oversight. The layer that carries a character between platforms cannot be built by any of them.
Evidence
The main run: one character, seven arms across five independent base models, the same five scenes each, five metrics. Three further rounds — Midjourney, Kling, video — ran under their own conditions and are reported separately rather than pooled. Eight independent base models in total, with the limitations stated before the results and three of our own hypotheses recorded as refuted.
Identity similarity alone is easy to game: a model that ignores the prompt and repeats the reference framing scores extremely well on it. So every result is reported on axes that can disagree — face identity, whole-subject similarity, prompt adherence, and how rigid the output set became.
Open format
Ten rules for writing a character that survives, each stated with the experiment that produced it. Apply them by hand; no tool required. The one that matters most costs nothing to adopt — an identity marker you cannot answer with yes or no is not an identity marker.
Open on purpose. A character that only works inside one product has the same problem as a character that only works inside one platform.
Status
The tooling that automates this — reading the spec out of your images, generating the reference set, producing per-platform prompts, scoring the output — is open to anyone. Sign in with Google or GitHub and it is yours; you start with free credits and no card is asked for.
It will not stay free forever, and saying otherwise would be its own kind of guessing. Reading a spec out of a set of images, producing prompts for each platform, and anchoring a registration all cost real money to run. What is not settled is the price — nobody has paid for this yet, so nobody knows what it is worth. When that changes it will be stated here before it takes effect, not discovered at a paywall.
Two limits worth stating up front, because finding them out later is worse. Credits are capped per account and across the whole site, so a heavy day can run out. And the measurements on this site cover seven platforms — a character bound for somewhere else may work, but nobody has checked.