chichi-pui is a Japanese platform where AI-generated illustrations, photos and gravure are posted with their prompts attached, and where members can generate new images directly in the browser. It combines a creator community, a searchable prompt archive and a free daily generation quota backed by in-site currency.
Used this tool? Rate it
Used this tool? Rate it
chichi-pui, read in Japanese as "chichipui", is a Japanese web platform built for one narrow purpose: hosting images that were generated by AI, together with the text that generated them. It is not a general art site that tolerates AI work, and it is not a standalone generator that happens to have a gallery. It is a posting board where the prompt is part of the artifact, plus a generation tool wired directly into that board.
The site's own About page dates the launch precisely. According to its official statement, chichi-pui was released on 30 October 2022 as a posting site exclusively for AI illustrations, and has since widened its genres beyond illustration to AI photography and AI gravure. That single sentence explains most of what the site is today: it started in the narrow slot that opened when NovelAI made anime-style AI illustration suddenly accessible in Japan, and it expanded outward from there rather than pivoting away.
The operator is not anonymous. The About page states that chichi-pui began as an individual's project and moved to being operated by ccpp Inc. from 2 December 2022, and that the company raised funding from East Ventures in order to run the service on a stable footing. The corporate site adds the registry details in full: the company name ccpp Inc., an address in Mishuku, Setagaya-ku, Tokyo, representative Yuki Takuma, an establishment date of 2 December 2022, and a contact address of info@chichi-pui.com. For a niche community site, that is an unusually complete disclosure, and it is worth knowing before you upload anything you care about.
There is also independent, non-vendor evidence of what the site was at birth. Japanese technology news outlet ITmedia NEWS covered the launch the day after it happened. Its report recorded that viewing illustrations is free, that free registration enables posting and unlocks the ability to view the prompt and negative prompt used to generate each posted image, and that posters must state which AI they used. Those three mechanics — free browsing, prompt disclosure gated behind a free account, and mandatory model attribution — are still the spine of the product several years later.
On most image platforms, a picture is the deliverable and the technique behind it is private. chichi-pui inverts that. The terms of service make prompt disclosure a condition of posting, not an option, and the result is that browsing the site is simultaneously browsing a working archive of what actually produced each result. The community's own slang reflects this: prompts are called jumon, "spells", and posters were originally referred to as jumonsha, "spellcasters" — a framing ITmedia noted in its launch coverage.
That design choice is why the site is useful to people who never intend to post. If you are trying to work out how a particular lighting treatment, costume detail or composition was reached, you can read the exact input rather than guess at it. The generation feature then closes the loop, because you can take what you read and run it without leaving the site.
Every image on chichi-pui is user-submitted. The operator is explicit that it takes no position on what users upload: the terms state that ccpp makes no representation or warranty regarding the ownership or licensing of intellectual property in posted content, whether it breaches any third-party contract, whether it violates any law, or whether it violates the terms themselves, and that the posting user handles all of this at their own responsibility. This is standard UGC risk allocation, but it is stated plainly rather than buried, and it sets expectations correctly: the platform curates surfaces and enforces rules, it does not vouch for provenance.
The defining feature is compulsory prompt publication. Under the terms of service, when posting, unless otherwise provided, users must enter the prompt and negative prompt used in generating that content into the designated form and consent to their publication. In practice this means each post carries a reusable recipe, and the site as a whole functions as a searchable corpus of working prompts rather than a wall of finished pictures.
Posts also carry the model or service used to generate them. The launch-day ITmedia report recorded this requirement, and today the homepage surfaces it structurally: image feeds are grouped under model-specific hashtags, so you can browse what a particular checkpoint or commercial model actually produces in the hands of many different people.
chichi-pui runs its own generator, and it is not a bolt-on. The terms state that by registering, users may use the in-site currency to generate images with image-generation AI on the service, and that generated images may be used for personal and commercial purposes. That commercial permission is written into the contract rather than left to inference, which matters if you intend to use output in anything beyond a hobby post.
The operator's own instructional site publishes the generator's specification. It lists the page URL as the site's generator path, supported devices as PC and smartphone browsers, pricing as free for basic use with up to 10 images per day and the ability to exceed that by purchasing in-site currency, and commercial use of generated images as permitted. That is a complete access model stated in one table: no install, no GPU, no environment setup.
The generator is not a one-box toy. According to the operator's instructional site, users choose a model, set a seed, pick from square 512 x 512, portrait 512 x 768 and landscape 768 x 512 sizes, and can tick an option to double the resolution; advanced settings expose the sampler with DPM++ 2M Karras and DPM++ SDE Karras among the choices, a step count specifiable between 1 and 70, and a scale value specifiable between 1 and 15. Anyone who has used a local Stable Diffusion interface will recognise the vocabulary immediately, and anyone who has not can leave the defaults alone.
Negative prompts are pre-populated with a general-purpose default rather than left blank, which is a small but meaningful onboarding decision: a beginner's first generation is much less likely to be ruined by anatomical artefacts they do not yet know how to suppress.
Output does not dead-end in a download folder. The instructional site explains that the generation history displays 80 images in order of newest first, that clicking an image lets you view its detailed information, and that from the history you can reuse the prompt or post the image to chichi-pui. That is the loop the whole product is built around: read a prompt from the feed, run it, iterate on it, publish it with your version of the prompt attached, and hand the next person a better starting point.
The model line-up is not static. The homepage announcement panel shows a running cadence of limited-time model releases and extensions — for example a market feature released on 13 August, and an announcement that a new model called HONEY was opened to the generation feature for a limited period while COFFEE and LEMON were extended. Model names on this platform are house-branded rather than upstream names, so the practical takeaway is that the roster changes and that a model you like may be temporary.
Around the archive sits a full community layer. The terms provide for like functionality, follow functionality between users, content sharing functionality and other features that the operator may add, adjust or remove. The homepage exposes recommendation feeds, trend feeds, a manga section, user-curated collections, and — distinctively — a "user-hosted posting projects" section, where the events and themed contests are proposed and run by members rather than staff.
There is an off-site layer too. The About page lists an X account and a Discord server, the latter used for previewing work in progress, casual conversation and irregular voice broadcasts by the operators.
The platform has grown a commercial layer for creators. An official membership subsystem lets individual creators sell access to member-only content priced in in-site coins on a monthly basis; one creator's public description states that they publish R18 images as membership-only content within the range that does not violate chichi-pui's terms. A marketplace was also announced on the homepage in August 2026. Treat details here as fast-moving — this is the part of the product that has changed most recently.
The highest-value use is also the simplest: browse, and read the prompt under anything that looks like what you want. Because prompt disclosure is contractual rather than voluntary, coverage is close to total, so you rarely hit the frustrating "no prompt provided" dead end common on general art sites. For someone learning what actually moves the output — token order, weighting syntax, negative-prompt hygiene — a large corpus of real examples with visible results beats any tutorial.
The second use case is for people who have no capable GPU and no appetite for installing a local pipeline. Because generation runs in the browser on PC or smartphone with a free daily quota, the distance between curiosity and a first generated image is a free account and a few sentences of English keywords. The instructional site's own guidance reflects this beginner orientation, walking through what each slider does and telling readers to leave most of them at defaults.
The About page is candid about the site's founding motive: after NovelAI's release in October 2022 produced a flood of AI illustrations, uploading them to existing illustration sites and social networks caused various problems, and chichi-pui was built so that generated work could be published freely and easily. That remains a real use case. On a platform where every single post is AI-generated by definition, the provenance argument simply does not occur.
The user-hosted projects section makes the site usable as event infrastructure. Members propose themed calls, other members submit to them, and the results collect under a shared banner. The operator has also run first-party campaigns and offline events. For a creator who wants an audience with a reason to look, a themed project is a far better vehicle than posting into a general feed.
A pattern visible throughout the feeds is creators maintaining a recurring original character across many posts, with consistent naming and tags. The site supports this well: tags, collections, follows and rankings all reinforce continuity, and the terms explicitly permit multiple accounts where the purpose is to separate artistic styles or pen names, provided other clauses are not breached. That clause is unusually creator-aware, and it is the mechanism by which someone can keep an all-ages line and a separate line cleanly apart.
For creators who have built a following, the membership subsystem and marketplace turn attention into revenue without leaving for a general doujin platform. This is a genuinely differentiated capability — most AI image communities have no native monetisation at all — but it is also the newest, so verify current mechanics on the site itself.
Viewing is free and requires no account, as ITmedia recorded at launch. Spend time in the trend and recommendation feeds and in the model hashtags first. This is the cheapest way to find out whether the site's aesthetic centre of gravity matches what you want to make, because it is quite specific.
Registration is done through account linkage. The terms list registration by Twitter account linkage, Google account linkage, GitHub account linkage, and other methods separately designated by the operator. There is no classic email-and-password path in that list, so decide which identity you want attached before you start. Note that at launch only Twitter linkage existed and it requested read-only permission; the current list is broader.
Registration is what reveals the prompt and negative prompt on posted work. Once you can see them, read them in bulk rather than one at a time. Patterns emerge quickly: which quality tags a community actually relies on, what goes in negative prompts by default, how a particular look is consistently achieved.
Open the generator, pick a model, and write the prompt in English keywords. The operator's guide gives a concrete example of this convention — describing a blonde, blue-eyed girl in a sailor uniform as a comma-separated keyword list rather than a sentence. Leave the negative prompt at its default at first. Choose portrait or landscape rather than square if you are unsure; the official guide recommends exactly that.
To refine rather than reroll, fix the seed. The operator's guidance is that fixing the value lets you keep the composition while changing the illustration slightly. This is the single most useful technique for moving from "close" to "right", and it is the difference between using the tool and gambling with it.
When you have something worth sharing, post it from the generation history, which retains 80 recent images and offers both prompt reuse and direct posting. Fill in the prompt fields honestly — this is a contractual obligation, not a courtesy — and, critically, set the age rating correctly before you publish.
This is the single most consequential habit on this platform. The R-18 guideline requires that when posting an R-18 work you must select the age restriction item R-18 for sexually themed cases or R-18G for violent or criminal themes. Getting this wrong is not a cosmetic error; the operator can and does re-categorise, restrict access to, or delete work, and the R-15 guideline states that the operator may determine a work's category and change tags at its discretion regardless of whether the poster applied the tag.
If you post explicit work, the requirement is quantified rather than vague. The R-18 guideline specifies mosaic processing using square rectangular mosaics whose sides are at least 1/100 of the image's long edge — with a minimum of 4 pixels — covering the entirety of the target area including outlines, and requires that the mosaic layer and the image layer be the same layer, in a state where the original image cannot technically be restored. It also warns that Gaussian blur may not achieve sufficient concealment and should be replaced by mosaic. Flattening matters: a "censored" layer you can toggle off is not censored.
The in-site currency has an expiry. The terms state that the validity period of coins purchased for a fee is 90 days from the date of purchase. Free daily grants and purchased balances are therefore not equivalent, and buying a large balance you cannot consume within the window is simply a loss. Buy in the smallest increment that covers your actual near-term usage.
The consequences reach your wallet. The terms provide that where the operator suspends or deletes an account because the user violated the terms or guidelines, it may accordingly invalidate or suspend the in-site currency that user holds. Unspent balance is not ring-fenced from moderation outcomes.
The terms require compliance with the licences of the model you used, naming Stable Diffusion, Waifu Diffusion, NovelAI, Trinart, Midjourney, DALL·E 2 and others among the licensors, and requiring that where a licensor demands that content be taken down, the user must promptly comply and immediately notify the operator. Posting here does not launder an upstream licence problem.
If you produce both all-ages and restricted work, the multiple-account allowance exists for exactly this. Keeping them apart protects followers who only want one of the two and makes your own tagging discipline much easier to maintain.
Because the prompt is published with the work, a well-organised prompt is part of your contribution. Grouping tokens logically and keeping the negative prompt meaningful rather than a copied blob makes your post more useful to the next person, which is the currency that actually earns follows on a site organised around technique.
The clearest fit. The browser-based generator with a free daily quota removes the hardware requirement, the cost requirement and the setup requirement simultaneously, and the surrounding archive supplies the knowledge requirement. Someone who has never generated an image can go from arrival to output in minutes and then improve by copying what visibly works.
People who care about technique rather than only results get more from this site than from a general gallery, because the input is always visible. If your workflow involves studying how others reach an effect, mandatory prompt disclosure is a structural advantage no amount of community goodwill can replicate elsewhere.
The site's centre of gravity is Japanese-language, anime-adjacent and gravure-adjacent. Creators working in that space will find an audience that understands the reference points, plus event infrastructure and monetisation routes tuned to it.
The membership layer and marketplace make this platform unusual among AI image communities: attention can convert to revenue on-platform. For a creator whose output suits the audience, that is a real reason to consolidate here rather than posting everywhere.
Non-Japanese readers who want a general-purpose generator will find both friction and mismatch: the interface and community are Japanese, and the model roster is house-branded and rotating rather than a stable published list of upstream models. Teams needing enterprise controls, API access or business support will find nothing here — this is a consumer community, not a vendor. And anyone who must avoid adult material entirely should read the Limitations section before browsing, because restricted content is a first-class, formally administered category on this site rather than an edge case.
chichi-pui is a web service with no dedicated client. The terms make the boundary explicit by putting the environment on the user: users must themselves prepare the information communication terminal, operating system, browser, internet connection and other usage and communication environment necessary to connect to the service. There is no mention of a native application anywhere in the contractual surface.
The operator's instructional site confirms the practical reach, listing supported devices as PC and smartphone browsers. Smartphone support is not an afterthought — the guide gives phone-specific advice, telling mobile users to adjust the step and scale values with the seek bar.
Access is federated rather than self-contained: Twitter, Google and GitHub account linkage, plus any other method the operator designates. If you keep creative identities separate from your main accounts, plan which provider you link before registering, since that linkage is your login.
The service spans several official subdomains rather than one page: the main community and generator, an instructional library site publishing prompt collections and technique articles, the membership subsystem for creator monetisation, and the corporate site for ccpp Inc. Being aware of this layout is useful, because a fact you find on one property may not be true of another.
Japanese, throughout. There is no evidence of an official English interface. Machine translation handles browsing adequately, but prompt discussion, guideline nuance and community etiquette are all conducted in Japanese, and the guidelines in particular are legal-register text where translation quality matters.
Basic use is free. The operator's instructional site states pricing as free for basic use with up to 10 images generatable per day, and the ability to generate more than 10 by purchasing in-site currency. Browsing has always been free without an account. There is no subscription tier for the core service in any official material reviewed.
Paid usage runs on internal currency rather than a plan. The terms establish that the operator may issue in-site currency usable on the service, and that users may purchase it via the operator's designated payment methods, with the explicit exception that minors under 18 cannot purchase it. That age gate is a hard rule stated in the contract, and it is worth flagging to younger users before they hit it.
Consumption is variable rather than per-image-flat. Third-party Japanese guides consistently describe consumption scaling with model and resolution, and the official guide's note that doubling resolution is an option implies the same. Because the exact table is behind the logged-in purchase page and rotating sales are frequent, verify current rates on the official points page rather than trusting any secondary source, including this one.
Two contractual facts should drive purchasing behaviour. First, the validity period of coins purchased for a fee is 90 days from the date of purchase. Second, purchased in-site currency cannot be refunded, returned, exchanged for other in-site currency, combined with, shared with or transferred to other accounts, except where the operator provides otherwise in writing or by electromagnetic record. Buy what you will use.
Output rights are generous and explicit: the terms state that generated images may be used for personal and commercial purposes. That said, the operator disclaims involvement in disputes between users and third parties over generated content, so commercial use is permitted but not indemnified.
The membership subsystem prices access in coins per month at tiers set by individual creators, so there is no single platform price for consuming member content. This is creator-set, not platform-set, and varies widely.
The obvious comparison is pixiv, and the relationship is historical rather than abstract. ASCII.jp's coverage recorded that after NovelAI's arrival, pixiv separated AI-generated work from regular work in its rankings and added features to explicitly mark posts as AI-generated and to filter them in search. The trade-off is clear: a general platform offers a far larger audience but treats AI work as a category to be segregated, whereas chichi-pui treats it as the entire premise.
chichi-pui is not alone in its niche. The same ASCII.jp article named AICON, aivy and AI Pictors as comparable AI-generated-illustration specialist posting sites emerging in the same period, and characterised the whole cohort as low-capital or individually operated services riding a new-genre boom. If chichi-pui's culture or content mix does not suit you, near neighbours exist.
If you want generation without a community, hosted generators and local pipelines both do it better in raw capability terms: more models, more control, no house-branded rotation. What you give up is the archive. The reason to use chichi-pui's generator specifically is that it sits inside the corpus of prompts, not that it out-generates dedicated tools.
English-language AI art communities exist and are larger in aggregate, but they generally lack the mandatory prompt disclosure that defines this site, and their aesthetic centre is different. For creators working in Japanese anime and gravure idioms, that mismatch is the deciding factor rather than a detail.
Pick chichi-pui if the prompt archive and the Japanese-language creator community are what you want, and you are comfortable with a rotating house model roster and a browser-only workflow. Pick something else if you need model transparency, English support, API access, or an environment with no adult content in it.
This is the most important thing to understand before recommending the site to anyone. chichi-pui does not merely tolerate restricted content; it administers it. The R-18 guideline defines an R-18 work as one containing sexual expression, violent expression, or expression that risks inducing suicide or crime, which by its provocativeness, explicitness or directness risks impairing the healthy development of young people under 18, and requires posters to select the corresponding age restriction. A separate R-15 guideline defines a middle tier with its own multi-part test covering clothing, pose and composition, and subject matter. Combined with an official membership subsystem where creators sell R18 content, the reasonable conclusion is that this is a mixed-audience platform with a substantial restricted segment, not an all-ages site with an occasional problem.
The permissiveness has firm edges. The R-18 guideline lists categories that cannot be posted at all regardless of tagging, including depictions exposing or emphasising genitals, non-consensual intercourse, incest, bestiality, necrophilia, sexual acts accompanied by criminal conduct, and discriminatory or attacking depictions targeting social status or beliefs. Most significantly, it prohibits child pornography absolutely: for deepfakes and so-called AI photos and other content that may be mistaken for live action, sexual depictions involving children or persons whose appearance may be mistaken for a child are prohibited regardless of the actual existence of the person and regardless of the poster's subjective intent. That last clause closes the "it's not a real person" and "that wasn't my intent" defences explicitly.
The operator's disclaimers are broad and clearly written. It makes no representation or warranty regarding the intellectual property ownership or licensing of posted content, third-party contract breaches, legal violations or terms violations, leaving these to the posting user's own responsibility. It also disclaims responsibility for compensating losses in disputes between users and third parties. If you post work whose upstream rights are unclear, that exposure is yours alone.
The terms are equally direct about reliability: the operator does not represent or warrant that the service is free from defects, malfunctions, errors, bugs or specification flaws, nor as to its reliability, accuracy, completeness, validity or availability. The terms also enumerate broad grounds for temporarily suspending the service. Do not build a deadline-bound workflow on the generator without a fallback.
Three separate clauses compound here: the 90-day expiry on purchased coins, the blanket no-refund and no-transfer rule, and the provision allowing invalidation of held currency when an account is suspended or deleted for rule violations. Any one is ordinary; together they argue strongly for buying small.
Models are announced under house names on a rotating limited-time basis. Third-party directories sometimes claim the site offers named upstream models such as Midjourney or DALL·E, but this conflates two different things — the models available in the site's own generator, and the external tools whose output users are permitted to post under the licence-compliance clause. If knowing exactly which model weights generate your image matters to you, this platform does not provide that transparency.
Everything is in Japanese, and the guidelines that carry the most consequence are the most legally worded documents on the site. Non-Japanese speakers can browse comfortably and generate adequately, but should not assume they have understood the posting rules from a machine translation.
Membership milestones circulated in Japanese media originate in company press releases distributed through PR TIMES and republished by portals such as ZDNet Japan and CNET Japan. The 100,000-member figure from May 2023 and the offline event attendance figures come from that channel: they are vendor-reported and have not been independently audited. Genuinely independent editorial coverage exists — ITmedia NEWS and ASCII.jp both reported on the site — but it is concentrated in the 2022–2023 launch period.
Yes for the core experience. Browsing has been free without an account since launch, and free registration enables posting. The operator's own instructional site describes generation pricing as free for basic use with up to 10 images per day, with the ability to exceed that by purchasing in-site currency. There is no subscription plan for the core service.
Yes, and this is contractual rather than implied. The terms state that generated images may be used for personal and commercial purposes. Two caveats: the operator disclaims involvement in disputes with third parties over generated content, and if you post work generated elsewhere you must comply with that upstream model's licence.
Yes. The terms require that when posting, unless otherwise provided, you enter the prompt and negative prompt used in generating the content into the designated form and consent to their publication. It is a condition of posting rather than an optional field, and it is the mechanism that makes the site's archive worth reading.
Yes, and it is formally managed rather than incidental. The site publishes separate R-18 and R-15 guidelines, requires posters to select an age restriction of R-18 or R-18G where applicable, specifies quantified mosaic requirements, and operates an official membership subsystem where creators sell restricted content. Anyone who needs to avoid adult material should treat this as a mixed-audience platform.
Several categories are prohibited outright regardless of tagging, including genital exposure, non-consensual and criminal sexual scenarios, and discriminatory depictions. The strictest rule concerns minors: for deepfakes, AI photos and other content that may be mistaken for live action, sexual depictions of children or of persons who may be mistaken for children are prohibited regardless of the person's actual existence and regardless of the poster's subjective intent.
Through external account linkage. The terms list Twitter account linkage, Google account linkage and GitHub account linkage, plus other methods the operator may designate. There is no standalone email-and-password registration in that list, so choose the identity you want tied to your creative account before you register.
Purchased coins expire 90 days from the date of purchase, cannot be refunded, returned, exchanged, combined, shared or transferred to another account except where the operator provides otherwise in writing, and can be invalidated if your account is suspended or deleted for violating the rules. Buy in small increments matched to near-term use.
The privacy policy states that the service may automatically acquire the terminal's global IP address and port number, its OS, browser, hardware and language settings, monitor screen information such as resolution and size, and other information acquired automatically for communication protocols and JavaScript. It also states that sensitive categories such as race, creed, social status, medical history and criminal record are not acquired without prior consent.
Yes. The privacy policy provides for requests to stop using, restrict the use of, or discard your personal information, alongside disclosure and correction requests, with the operator to investigate promptly and act where the request is well-founded. The stated contact point is info@chichi-pui.com. Note the operator is Japanese and the policy is framed around Japan's Act on the Protection of Personal Information.
Yes, within limits. The terms say users may in principle register only one account per person, but that where the purpose is to distinguish artistic styles or pen names, and no other clause is breached, they may register and simultaneously use multiple accounts to the extent necessary. This is the clean way to separate all-ages and restricted output.
