Why is the content policy the decisive document for 2D work?

Because the style filter is where this branch is actually constrained. A vendor that publishes what its filter refuses lets you find out before paying whether your use case is supported; one that publishes nothing has reserved the right to refuse anything, and in practice refuses more than buyers expect.

Do vendors name the models they run?

The ones confident in their stack usually do, because the model line-up is what this audience compares. A service that describes its stack as proprietary has made itself unverifiable and has also reserved the right to change the product underneath you without notice.

Is 2D cheaper to work in?

Per usable result, yes, and the reason is that flat colour hides the structural errors every generator makes. That is a statement about the medium rather than about any product, and it is the one generalisation in this category that holds across vendors.

Does anyone publish what is blocked?

Stylised output is the cheapest branch of this category to work in and the most heavily moderated. The document that decides whether a subscription is usable is the content policy, and most of them are silent on the part that matters.

Category checkNo product was usedRevised 2026-10-19

The question

Buyers in this branch hit refusals more often than anywhere else in generation, and almost nobody can predict which prompt will trigger one. That unpredictability is not a technical limit; it is an undocumented policy. A vendor could publish the categories its filter refuses. Some do.

What we checked

Content policy

What is prohibited, and is likeness transfer named?

Satisfied by

A published list of prohibited categories, with material involving real people's likenesses explicitly among them.

Usually found

A policy that bans the obvious and stays quiet about likeness, which is the one that decides whether the operator survives the year.

At stake

An operator vague about likeness is an operator one enforcement action away from taking your purchased credits down with it.
Pricing page

Are the limits stated in units a buyer can count?

Satisfied by

Clip length in seconds, resolution in pixels, generations per day as a number, queue priority described in words that mean something.

Usually found

A credit balance with no stated conversion, and the word unlimited somewhere above a fair-use clause that contradicts it.

At stake

This is the difference between a subscription that finishes work and one that runs out on a Tuesday.
Terms of service

Who holds rights in what the tool produces?

Satisfied by

An explicit grant of output rights to the user, and a narrow, stated licence back to the vendor for operating the service.

Usually found

Silence about the user's rights and a broad perpetual licence to the vendor covering promotion and model training.

At stake

Publishing anything made with the tool depends entirely on this paragraph, and it is never on the pricing page.
Support channel

Can a question be asked before any money moves?

Satisfied by

A channel that answers a pre-sales question — anything from a ticket form to a staffed chat — with a human reply.

Usually found

A form that routes to nothing, or a help centre whose articles were written for a different product.

At stake

The channel that will not answer before payment is the channel that will not answer about a renewal.

What filter transparency looks like

The useful version of a content policy in this branch names three things: the categories refused outright, the basis on which output is judged, and what happens to an account after a refusal. All three are publishable without giving anyone a route around the filter — the categories are the same everywhere, because payment processors set them.

None of this weakens a filter. It makes the subscription predictable, which is what is being sold.
Published	What the buyer can then do
The refused categories	Decide before paying whether the work is supported
Judged on rendered output, not prompt text	Stop writing descriptions that trigger refusals pointlessly
Consequence of a refusal	Know whether it is a declined generation or a closed account
Appeal route	Recover an account closed in error, which happens

One category is refused by every operator in this branch and is worth stating plainly: output coded to look younger than adult. That is enforced on appearance rather than on any age a prompt asserts, and it is the most common reason an account is closed here. Our own position is on the teen category.

Verdict

Pay for predictability, not permissiveness

Between two vendors with similar output, the one that publishes its refusal categories is the better purchase, because the other one's silence resolves into refusals you cannot plan around. Add a named model line-up and this branch becomes the most economical place to work in the whole category.

Paid placement

SpicyAI — Discover AI characters with unique personalities. Chat, share stories, or create a companion of your own — with text, voice, images, and video.

Contradicted elsewhere Pricing page. The only vendor read here that publishes its limits and its prices without a sign-in: $6.99 a week, $9.90 a month, $59.88 for twelve months, with 100 images and 10 videos a month on Premium. The same page then states the free allowance twice, as five chat turns on the plan card and ten turns per character in the table, and the Premium allowance twice, as unlimited on the card and 1,000 turns in total in the table.

Published Terms of service. Effective 2 December 2025, with the operator named and a contact address in the document. The user owns what they generate, and the licence running back is perpetual, irrevocable, sublicensable and transferable — and extends beyond operating the service to promotion and advertising.

Published Privacy policy. No retention period: the policy gives the factors it weighs instead, which is the same answer six of the eight catalogued vendors give. Chats and user content are used to train the models and are monitored for compliance, and the policy says so plainly rather than burying it.

Contradicted elsewhere Content policy. The community guidelines are the strongest of any vendor read here: eighteen and over, no character presented as younger than that and no aged-up rewrite of one, no sexualised depictions of real people without consent, human and automated review. The public tag list then carries jk, a Japanese abbreviation that denotes exactly the age bracket those guidelines rule out.

Published Checkout and cancellation. Card only, with other methods described as coming. Cancellation any time, with access running to the end of the billing cycle and no further charge — stated on the page where the subscription is sold rather than in a separate document.

Contradicted elsewhere Support channel. A support address is published, which several vendors here do not manage. Two of them, in fact: the terms give one address and the community guidelines give the same mailbox at a different domain.

Go to SpicyAI Go to the Telegram bot

Questions

Why is the content policy the decisive document for 2D work?: Because the style filter is where this branch is actually constrained. A vendor that publishes what its filter refuses lets you find out before paying whether your use case is supported; one that publishes nothing has reserved the right to refuse anything, and in practice refuses more than buyers expect.
Do vendors name the models they run?: The ones confident in their stack usually do, because the model line-up is what this audience compares. A service that describes its stack as proprietary has made itself unverifiable and has also reserved the right to change the product underneath you without notice.
Is 2D cheaper to work in?: Per usable result, yes, and the reason is that flat colour hides the structural errors every generator makes. That is a statement about the medium rather than about any product, and it is the one generalisation in this category that holds across vendors.