Open Pose & Costume: 5,800+ Pose, Outfit and Character References in One Workspace

When you draw a character or build an AI image, the hard part is often not the idea. It is working out where the hands should go and how a jacket actually sits on a body. Most people end up with a pile of browser tabs: one site for poses, another for outfits, and a third place to improvise the character. Open Pose & Costume is an attempt to put all three in one page.

According to the official site, Open Pose & Costume is a bilingual workspace where you can look for inspiration, combine pose and outfit references, and generate your next character. The site lives at omilic.com, and the author posts on X as @HongyeMain. In practice it is three things at once: a reference gallery, a mix-and-match studio, and an image generation feature that you have to connect to your own service. The interface comes in English and Simplified Chinese, so you can use it without reading any Chinese.

The project is young. The author shipped several versions in late October and was still adding material in early November. The code lives on GitHub under the MIT license, so expect features and numbers to keep changing, and what you see may differ from this article.

The outfit row is the one I find most telling. With 2,769 outfits, clothing makes up well over half of the library, which suggests the author sees clothing references as the biggest gap. Poses, outfits and characters are managed separately, each with its own filters, so results do not get mixed together.

How do you search and save references?

Searching and saving is the everyday workflow in Open Pose & Costume, letting you find any reference in the three libraries by name, keyword or ID in English or Chinese. Beyond search, you can filter by category, source library, character tag and visual style, and the character library adds country/region and hair length. Sorting and the displayed view are independent, so you can filter down to a pose type first and decide on the camera angle afterward.

Every pose set lets you switch between views, and you can turn on the depth map when you need it. Open a card to zoom in and download the original, while hovering or tapping a thumbnail only changes that one card, not the whole list. Custom poses that have a depth map can switch to it directly, and the rest stay as line art with a notice. For anyone who draws, seeing the same action from five angles is what makes this more useful than a plain pose gallery.

Anything you like can go into "My Favorites", so you can come back to it, keep mixing, or download it later. Share links keep your language, filters and selections, so a friend sees the same screen you do. The official site says favorites are saved in your current browser, which means switching browsers or clearing site data may wipe them, so write down the IDs of anything important. There are also keyboard shortcuts: / to search, the arrow keys to move through details, and Esc to close, which makes browsing much smoother.

What changed in the latest update?

The latest update is a large content expansion that adds anime outfits, model and dance poses, and male reference material. There are now 1,350 anime outfits, of which 1,320 come from an upstream anime gallery and 30 were made by the author. The upstream anime outfits map one-to-one onto the original realistic outfits, so the same look now has an anime version, with IDs that start with ANIME- and favorites tracked separately. Across both styles the library has 2,769 outfits, and you can filter by realistic or anime.

The anime outfits keep eight categories: spring, summer, autumn, winter, lingerie, party, swimwear and glamour looks. By default only 932 of the 1,320 are shown, and you need to switch on "Show extended categories" to see the full set, which is where lingerie, swimwear and glamour looks live. I think hiding them by default is sensible, since plenty of people open tools like this at work or in public. If you stream, teach or share your screen, check that toggle first.

On the pose side, the update adds 10 female model and dance pose sets, each with five views (front, top, bottom, side and three-quarter) plus matching depth maps, for 50 line drawings and 50 depth maps. It also adds 13 male pose sets with 65 views, four male realistic outfits with full footwear, and six male character references. These custom poses keep the white-background structural line-art style and can be searched in both the gallery and the generation picker. The depth maps are AI-estimated relative depth, and the author notes they are a visual guide, not calibrated 3D measurements.

The character library gains 22 fictional adult characters, grouped into 11 country or region sets with one man and one woman each. The author is explicit that these groupings are just catalog labels, not real nationalities or typical looks of any group. Overall, male material is still thin, and the original poses, outfits and characters were labeled female. If your character is mainly male, your choices are limited for now.

How does one-click generation work?

One-click generation is the built-in mixing feature of Open Pose & Costume, which sends your chosen pose, outfit and optional character reference to an image service you configure yourself and returns a combined character image. The key phrase is "you configure yourself": the site does not generate anything for you, and the generate button stays disabled until an API is set up. The README also links to a free-trial guide hosted on Feishu, which I have not checked in detail. Because real generation needs your own paid service, this walkthrough follows the official documentation, and results will vary with the model and your reference images.

Step 1: Configure your image service

Open "One-click generation" from the sidebar, then click "Configure image service" or the settings icon at the top right. You will enter a service URL, an API key and a model name, and without them the generate button cannot be clicked. The defaults are https://atomflow.vip and gpt-image-2, and you can change both to a service you prefer.

Step 2: Choose a pose, an outfit and a character

Next, choose your references: pose and outfit are required, and the character is optional. The pose sets the action, the outfit sets the clothing, and the character keeps the face and overall look. You can pick from the library or press "Random mix" for ideas, and character images can be uploaded, dragged in or pasted as PNG, JPG or WebP files up to 10 MB each.

Step 3 (optional): Extract your own outfit

Click "Upload outfit", choose your outfit photo on the left, press "Extract outfit" in the middle, and click "Confirm" on the right once the result looks right. This step sends one extra generation request, you can retry if it fails, and your original reference is not replaced until you confirm. I find this handy for turning your own clothing photos into references, but remember that the original photo is sent to the service you configured.

Step 4: Adjust the settings and generate

You can switch the view on a single card, pick a viewing angle and decide whether to use the depth reference. There are four visual styles: follow the reference, natural photo, studio look and anime illustration. You can also set the size and add a description, and the advanced settings let you edit the full prompt. After you press "Generate image" you can keep browsing the library, and if a run fails, just retry it.

Step 5: Compare the result and keep creating

Results can be viewed as "Result only", "Two-column preview" or "Overlay preview", and the overlay mode lets you drag a slider to compare the generated image against the pose original. If the figure size or position does not line up, "Adjust alignment" lets you scale and drag each image, and this only affects the preview, never the downloaded file. When you like a result, "Continue with this character" keeps generating with the same person, and the history drawer keeps your 20 most recent successful results in your browser.

The official documentation has a section on data handling. Your API key is kept only for the current browser session by default, and it is saved locally only if you tick "Remember key", which you can clear in settings. When you generate, your reference images and description are sent to the service you configured, and fees follow that service. The author says the Worker used for forwarding does not store keys or images. My advice: if you are on a computer that is not yours, do not tick "Remember key".

What is the companion Skill?

The companion Skill is open-pose-costume-skill, a separate project by the same author that lets you combine a character, a pose and an outfit into one image through natural language in Codex. The split is clear: the website is for browsing and picking references, and the Skill does the actual generation. It accepts images, local files, reachable image links or clear text descriptions, and it can select references by ID, such as P047 or SPR-0050. The first public version, v0.1.0, was released on October 29.

The README adds that if the image service is unavailable or lacks image generation, the Skill says what is blocking it instead of passing off a prompt as a finished picture. I like that, because at least nobody is left thinking it worked. It is Codex-only for now, so if you do not use Codex you can safely skip it.

What are the licensing limits?

The code and the images in Open Pose & Costume are licensed separately, with the code under MIT and the images following different rules depending on where each one comes from. According to the official GitHub documentation, the images are not covered by the MIT license. The table below sorts the common material types into what you can and cannot do, but read the official terms again before you rely on it.

If you work on commercial projects, the main thing is to know which kind of material you used. Original poses and outfits can be used commercially under the original site's terms, but custom material, and any image generated from it, is limited to personal learning and non-commercial use. I would note down the IDs of everything you use as you generate, so you can trace the source of each asset when it is time to deliver or publish.

The 1,407 realistic outfits match the size of the COSTUME-DB-01 gallery from a Japanese creator exactly, so they very likely come from the same source. The official pages do not name the author outright, so treat that as my inference from the numbers, and rely on the project's SOURCES file for the actual terms. The README also thanks the original pose and outfit creators.

Who is Open Pose & Costume for?

Open Pose & Costume is best for illustrators, character designers and AI artists who need a lot of pose and clothing references. If you normally hop between several galleries, having poses, outfits and characters together, plus favorites and IDs, saves real switching time. For AI artists, the depth maps and one-click generation add another layer of use.

A few things are worth thinking through first. Generation needs your own service and budget, the library still leans heavily toward female material, and the project is new enough that numbers and features keep changing. If you only need an occasional sitting pose, plain browsing is enough and you never have to set up generation.

I would start with the gallery and get comfortable with IDs and filters, then decide whether to connect an image service. That way you can judge whether the material suits your style before you spend money on generation. On the other hand, if you already have an image service, the mixing studio and comparison previews are easy to pick up.

What else do readers ask about Open Pose & Costume?

This section answers the six questions readers ask most, covering cost, commercial use, how it differs from other galleries, and availability.

Does Open Pose & Costume cost anything?

I found nothing in the official material saying that browsing the library requires payment. One-click generation does need an image service you set up yourself, and fees depend on that service. The README also links a free-trial guide, so check the official documentation for details.

Can I use the material commercially?

It depends on the type of material. Original poses and outfits can be used as reference, edited and used commercially under the source site's terms, but the images cannot be reposted or redistributed. Custom material and images derived from it are for personal learning and non-commercial use only, and the character library is labeled CC0 at its source, which also asks people not to repost the image data.

How is it different from a plain pose or outfit gallery?

Many galleries focus on a single kind of reference. Open Pose & Costume puts poses, outfits and characters in one place, with search in English or Chinese, multi-view and depth maps, favorites, one-click generation and a companion Skill. The trade-off is that generation needs your own service, and the material is still weighted toward female references.

Do I have to use the default atomflow.vip service?

No. The service URL, API key and model can all be changed, and the interface uses an OpenAI-compatible POST /v1/images/edits endpoint. If you use another service, it has to allow browser cross-origin (CORS) access, or the connection may fail.

Is it available in my country, and does it work on phones?

The official site and GitHub mention no regional restrictions. The site is a static web page with a responsive layout, so it can be browsed on a phone. For generation, the documentation says mobile generation on the live site goes through a restricted Cloudflare Worker that forwards requests to the default service. Whether your chosen image service works in your country, or runs smoothly on a phone, depends on that service.

Can I run it on my own computer?

Yes. You need Node.js 22.12 or newer, run npm ci, then npm run dev, and open http://127.0.0.1:5199 in your browser. If that port is already in use, the program exits, and you can pick another with npm run dev -- --port 5200. Note that running locally only runs the website code, and the image licensing rules do not change.

Is Open Pose & Costume worth bookmarking?

Open Pose & Costume is a creative tool that squeezes the job of finding references into one page. What stood out is not the sheer number of images but that poses, outfits and characters share one way to search, save and mix. Generation needs your own service and the licensing needs a close read, so I treat it as "a very good reference library with an advanced option on the side". For creators, the time saved is not in one-click generation but in no longer hunting for references across ten tabs. Until you have read the licensing for each asset, even the handiest mixing studio is only for practice and inspiration.