If you have ever drawn a person, you probably know this feeling: you can picture someone mid-run, but the drawing looks off the moment it hits the page. Which leg goes forward, and how should the arms swing? Finding a reference photo with the right angle is hit or miss, and posing for your own camera is awkward. This time I want to introduce Pose Gacha (“ポーズガチャ” in Japanese), a free browser tool built to take that headache away.
According to the official site, it is a browser tool that builds human poses from numerical values, applies them to a VRoid character or a 3D mannequin, and lets you “pull” several options at once, gacha style. The tool is at https://tools.oyasumi-gamedev.com/posegacha/, and the author's usage guide is at https://oyasumi-gamedev.com/tools/posegacha/. It is free and needs no sign-up, though it is currently labeled as in development, so features are still being added.
In this article I will walk through what it can do, how to use it, where the exported results can be used, and whether it can be paired with ComfyUI. One note up front: the ComfyUI section is my own extension based on common workflows, and the official site does not mention it. I will mark what comes from the official site and what is only my guess.
What is Pose Gacha?
Pose Gacha is a free web tool by Japanese creator MIYADAIKU that builds character poses from adjustable numbers and lets you pull multiple options at once, gacha style. “Gacha” comes from Japan's capsule-toy vending machines and now means any random draw, which is exactly how the tool presents its poses. It lives in the self-made tools section of the author's site, “おやすみゲーム開発部” (Oyasumi Game Development Dept.), and the official site files it under manga-making tools. Right now it has 23 actions, such as walking, running, punching, sitting, looking at a phone and crying, plus 42 common preset poses like crossing arms, scratching your head and bowing.
For me, the most interesting part is that this is not a bone-by-bone editor. It hands you a pile of poses that already look plausible and lets you pick. That difference matters in practice: dragging bones assumes you already know how a pose holds together, while pulling poses lets you find ideas in the results. For anyone who has a motion in mind but no reference, the second approach has a much lower barrier.
The author has also built a series of tools for manga creators, including Name Gacha (“ネームガチャ,” a tool for rough storyboards, which manga artists call “names”), “コミカライザー” (Comicalizer) and Background Gacha (“背景ガチャ”). According to the official site, the “Pose” tab inside Name Gacha is Pose Gacha itself, so the tools work together.
How are Pose Gacha's poses made, and is it AI?
Pose Gacha's poses are not AI-generated; they come from hand-written motion curve tables combined with a set of human-movement rules. Take running as an example: each moment of a stride has a table of angles for the thighs, knees, arms, chest and head, and values such as speed and stride length reshape that table. On top of that, the author layers rules that make poses look human, such as arms swinging opposite to the legs, the waist and chest twisting in opposite directions, and the head staying pointed forward.
The same rules act as a filter when you pull. According to the official site, poses with clipping, crossed legs, impossible joints or a near-fall are discarded automatically. In my view, that is the biggest difference from random shuffling: the fun is in pulling, but the time saved is in the throwing away.
One small twist is worth mentioning. The official site says the tool's code was written in discussion with the AI agent Claude Code. So the more accurate description is that pose generation uses no AI, while development did. If you plan to introduce the tool to others, it helps to keep that distinction clear so nobody gets the wrong idea.
How do you use Pose Gacha?
Using Pose Gacha is simple: pick an action, adjust the numbers or pull the gacha, choose a character and expression, then export. According to the official site, it works in both desktop and mobile browsers, but since you want the preview and the number panel on screen together, a larger desktop screen is recommended. Here are the five steps as I understand them from the official guide.
Step 1: Open the tool and pick an action
Open the tool and choose an action at the top right, such as running, punching, looking at a phone or crying. The character appears in the large preview on the left. The page also has a language menu at the top that lists nine languages, English included, so you can switch if the Japanese labels slow you down. When you first arrive, it helps to click through a few actions to get a feel for it.
Step 2: Adjust the numbers with sliders
In the numbers panel, pick a style first. Running, for example, offers options such as all-out sprint, light jog, “girl run,” ninja run and cartoon-style exaggerated run. Then drag the sliders for timing, speed, stride and so on, and the preview updates instantly. A checkbox to the left of each value locks it so the gacha will not change it, and the “reset to default” button returns to the action's starting values.
At the bottom of the page there is a “Flow” (“流れ”) section that lines up 8 poses where only one value changes, which is timing for walking and running. If you are deciding which instant of a stride to draw, this is very handy. Click one and that value is applied to your pose.
Step 3: Press a gacha button
If you want more options, press a gacha button at the top right. I explain the three modes in a later section. The pulled poses line up in the large area on the left, and clicking one applies it to the character. After you close the results, the “Gacha Results” (“ガチャの結果”) button at the top right opens them again.
Step 4: Choose a character, a look and an expression
Under the preview on the left, choose a character: a bundled one, your own VRM file, or a 3D mannequin. You can also switch between four looks and several viewing directions. If you picked a VRoid character, you can choose an expression as well. Expressions are stored separately from poses, so changing the action or pulling again does not wipe them out.
Step 5: Set the camera and export
You can rotate the camera by dragging, pan with a right-click drag and zoom with the scroll wheel, the same controls as the author's Background Gacha. When you are happy, use the “Export” (“書き出し”) panel on the right to save a PNG, a VRMA file or a share link. The official site reminds you to keep your own copies of images and pose files, because after an update the same numbers may produce slightly different poses.
What are the 23 actions and 42 preset poses?
Pose Gacha sorts its actions into four groups (movement, action, daily life and emotion) for 23 in total, plus 42 common preset poses. The table below is organized from the official guide, and each action has several named styles to choose from.
| Category | Actions included | Count |
|---|---|---|
| Movement | Standing, walking, skipping, running, jumping, turning around, sitting | 7 |
| Action | Punching, kicking, throwing, falling | 4 |
| Daily life | Looking at a phone, phone call, eating, drinking, writing, reading, stretching, lying down | 8 |
| Emotion | Crying, getting angry, feeling shy, holding head in hands | 4 |
| Preset poses | Standing, thinking, talking gestures, emotional gestures, low postures | 42 poses |
Each action has specifically named styles. Sitting includes sitting on a chair, crossing legs, kneeling on the heels (seiza), sitting cross-legged, hugging the knees, sitting sideways and squatting. Drinking includes using a cup, holding a teacup with both hands, swigging from a bottle and downing a drink with a hand on the hip, while punching offers a straight, a jab, a hook, an uppercut and a big swing. Just reading the list shows how much effort went into the idea that one action can be played many ways.
The emotion group is the part I find especially clever. According to the official site, crying, getting angry, feeling shy and holding the head in the hands are expressed through the curve of the shoulders and back, the tilt of the head and the position of the hands. That means even a faceless 3D mannequin can read as upset or embarrassed. For manga artists, it is like having someone work out how the body should act before you start drawing.
Daily-life actions also come with props, such as a phone, a cup, a book, a newspaper, a chair and a desk, which are placed to match the hands and waist. That makes scenes like typing on a phone with both hands, chugging from a plastic bottle or writing at a desk easy to set up without hunting for separate references. In practice, this is the kind of detail many pose libraries struggle to offer in one place.
The preset poses are the 42 poses from Name Gacha turned into actions, including crossing arms, scratching the head, waving, bowing, raising both hands and hugging the knees. After choosing a pose, drag the “motion” value (“動き”) and small movements appear, such as scratching, waving, nodding bows or breathing, and a value of 0 gives the original pose. A “left/right” switch swaps which hand is used, which is great for subtle variations.
What is the difference between the three gacha modes?
Pose Gacha has three ways to pull, a cross-action gacha, a single-action gacha and a pull near your current values, and they differ in how wide the draw is. All three show 12 options in the large preview area for you to pick from.
| Mode | Official button name | What it pulls from | Best for |
|---|---|---|---|
| Cross-action gacha | “ポーズ横断ガチャ” | Every action, with each option using a different action | You have no idea yet and want to see what actions exist |
| Single-action gacha | “ポーズ指定ガチャ” | The value ranges of the current action; it builds 4 times 12 candidates, drops unnatural ones, then picks 12 whose limbs point in different directions | You have chosen the action and want different variations |
| Nearby gacha | “今の数値の近くで” | Only small changes around your current values | You are almost happy and want to fine-tune |
Take the single-action gacha as an example. When you pull “sitting,” kneeling, cross-legged and hugging-the-knees poses can all show up together. That is fun, but if you only want kneeling, use the checkbox on the left to lock the sitting style, then keep pulling the other values. In practice, this lock is the most important switch for controlling what the gacha gives you.
My take is that the three modes map to three stages. Use the cross-action gacha when you have no idea, the single-action gacha once you know the action, and the nearby gacha when you are close to done. Following that order is more efficient than re-pulling blindly, and it is also what makes the tool more useful than a typical random-pose site.
How do you choose characters, expressions and camera angles?
Pose Gacha can apply a pose to a bundled VRoid character, your own VRM file or a 3D mannequin, and expressions and the camera can be adjusted separately. The table below shows how the three character options differ.
| Character option | Details | Expressions | VRMA export |
|---|---|---|---|
| Bundled characters (Female 01, Male 01) | Made by the author in VRoid Studio; usable only inside the tool, and the models cannot be extracted or redistributed | Yes | Yes |
| Your own VRM file | Load a .vrm file with “自分の VRM を読み込む…” (load my own VRM); it is processed only on your device and is not sent anywhere | Yes, if the model has expression data | Yes |
| 3D mannequin | Built from joint angles; you can show the skeleton and the “center of gravity and support” views to check balance | No | No (a VRoid character must be selected) |
VRoid characters can be drawn in four looks: VRoid colors, name lines (color with outlines), grayscale plus lines, and screentone plus lines. The last two match how Name Gacha and Comicalizer draw characters, and screentone plus lines turns shadows and color depth into halftone dots for a strongly manga feel. If your goal is a rough underlay, I would try grayscale plus lines first because it competes the least with your own linework.
Expressions come straight from the ones built into a VRoid character, including joy, anger, sadness, fun, surprise, blinking and the mouth shapes for the Japanese vowels, with adjustable intensity, eye closing or winking, and mouth opening. There is also an expression gacha at the bottom of the page that makes 8 small variations of your chosen expression in a row. A mannequin has no face, so expressions are not available for it.
The camera supports six control schemes, including the standard drag-to-rotate, schemes close to 3D software or drawing apps, and a mode that lets you stand on the floor and look around. Direction buttons give you front, three-quarter, side, rear three-quarter and back views, and you can change the focal length from wide angle to telephoto. My advice is to use the direction buttons for a rough angle first and then fine-tune, which is faster than only rotating with the mouse.
Where can you use the exported results?
Pose Gacha exports three kinds of results, PNG images, VRMA pose files and share links, which suit rough underlays, other VRM tools and saving or handing off a pose. Check the table first, then I will go through each one.
| Result | Good for | Things to watch |
|---|---|---|
| PNG image | Rough underlays for manga and illustration, pose inspiration, body-movement practice, reference material for games or videos | Choose 1200px, 1600px or 2400px; follow the model terms when you use a VRM character |
| VRMA pose file | Carrying the same pose into other tools that support VRM; a VRoid character's expression is saved too | A VRoid character must be selected; support for VRMA varies by tool, so test on a small scale first |
| Share link | Passing the current action, values, camera and expression to someone else, or saving a pose you like | Does not include a VRM you loaded yourself; poses may shift slightly after updates, so save PNG or VRMA for the important ones |
The most direct use is the PNG. According to the official site, you can put the PNG under your storyboard rough and trace it, so even hard poses stop being scary. If you open the tool from the sister tool Name Gacha, the image is also exported at the aspect ratio of that panel. For anyone working to a deadline, that saves a lot of time spent looking things up.
VRMA is a pose and animation file for VRM, and the official site says other tools that support VRM can read the pose. For people making games or 3D projects, that means carrying the same pose into another tool and continuing from there. I have not tested each tool's support, so I suggest trying one small file before relying on it.
The link with the sister tools is also worth a mention. In Name Gacha's panel editor (“コマの編集”), pressing “ポーズガチャで編集” (edit in Pose Gacha) opens the pose and expression with that panel's camera, and after you adjust it, you send it back, even to other panels where the same character appears. A “Pose” button at the top of Comicalizer works in a similar way.
What about licensing? The official site says exported PNG and VRMA files can be used freely, commercially or not, and credit is up to you, so manga and illustrations do not need it. For games and videos, the author would appreciate a credit in the end roll or description, for example “Tool used: ポーズガチャ (MIYADAIKU).” If you use a VRM character, though, that model's own terms of use come first.
Can you use Pose Gacha images as pose references in ComfyUI?
A PNG exported from Pose Gacha can technically be used as a reference image in ComfyUI, but the official site does not mention this use, so it is an advanced pairing you need to test yourself. What follows is my own inference from common ComfyUI workflows, not official guidance, and I have not verified each step.
The principle is not complicated. ControlNet in ComfyUI cannot take a reference image directly, so the image first goes through a “preprocessor” that turns it into a control image, for example OpenPose or DWPose to extract a skeleton, Lineart to extract line art, or a depth preprocessor to extract front-to-back relationships. ComfyUI does not bundle every preprocessor, and a common approach is to install the comfyui_controlnet_aux extension through ComfyUI Manager.
| Approach | Pose Gacha output to use | Notes |
|---|---|---|
| Pose skeleton | A regular PNG with an OpenPose or DWPose preprocessor | The most intuitive way to make a generated character take the same pose, but it depends on how well the pose is detected |
| Line art | A PNG exported with the “name lines” or “grayscale plus lines” look, with Lineart control | These looks are already line drawings, so they should be easier to use for non-realistic renders |
| Depth | A PNG with a depth preprocessor | Keeps the front-to-back order of limbs and is often stacked with pose control |
There are a few cautions. First, OpenPose-type preprocessors were built for photos and ordinary illustrations, while Pose Gacha outputs renders of 3D characters or mannequins, so back views, rear three-quarter angles or exaggerated manga-style runs may fail to be detected; that is my guess, so please test it. Second, the official “skeleton” view (“骨組み”) is a display mode for checking balance, and I saw no description of it outputting an OpenPose-format skeleton, so do not treat it as an OpenPose image. Third, VRMA is a pose file, not an image, so it cannot go straight into ControlNet and would need to be applied in another 3D tool and captured first.
Last is model matching. Different base models, such as SD1.5, SDXL and FLUX, need matching ControlNet weights, and the preprocessor output has to fit what the weights expect, otherwise the effect may be weak or even absent. Also, the official site only says exported images can be used freely and does not say whether using them as AI image generation references counts. The page points to a terms page, and I suggest reading it before this kind of use: https://oyasumi-gamedev.com/tools/terms/. I was not able to confirm what that page says, so treat the terms as the final word.
How is Pose Gacha different from other pose reference sites?
The biggest difference is that Pose Gacha generates poses from numbers and mass-produces them with a gacha, instead of offering ready-made photos or asking you to pose a 3D model by hand. The table below compares it with two common pose tools.
| Tool | How it works | Good for | Notes |
|---|---|---|---|
| Pose Gacha | Adjust numbers and pull; unnatural poses are filtered out automatically | Finding pose options and making rough storyboard underlays | Free, no sign-up, in development |
| Posemaniacs 30-second drawing | Shows random 3D poses at set time intervals | Gesture drawing and anatomy practice | The site says it offered this on the web as early as 2005 |
| PoseMy.Art | You pose a 3D model yourself and view it from any angle | When you need a very specific pose and angle | Basic models are free and more options are paid (per a Wacom community article) |
My take is that these three are not substitutes for each other. For gesture practice, a timed tool like Posemaniacs fits better. If you need one very specific pose and angle, posing a 3D model yourself is the most direct. And if your thought is “I know I want to draw a run, but I am not sure what a good-looking run is,” Pose Gacha's pulling is a natural fit.
What limits should you know before using Pose Gacha?
Pose Gacha is still in development, so the main things to watch are version changes, character model terms and device limits. The official site says directly that after an update, the same numbers may produce slightly different poses, so save any pose you want to keep as a PNG or VRMA right away. That is the most practical warning for long-term use, and it also means a share link should not be your only backup.
On the character side, the two bundled characters can be used only inside the tool and cannot be extracted or redistributed. If you load a VRM made by someone else, the official site specifically asks you to follow that character's terms of use. Expressions work only on VRoid characters, and your own VRM will not offer them if the model has no expression data.
On privacy, the official site says pose calculation, VRM rendering and export all happen inside your browser, and a loaded VRM is not sent anywhere. The bundled characters are loaded from the site when you select them, and a share link carries the action, values and expression but not your own VRM. In my view, that design is a plus for creators who do not want to upload a character model to the cloud.
What are the most common questions about Pose Gacha?
This FAQ covers six points people ask about most: cost, character choices, gacha modes, commercial use, ComfyUI and interface language.
Is Pose Gacha free, and do I need an account?
According to the official site, Pose Gacha is free to use and needs no sign-up or login. Pose calculation, VRM rendering and export all run inside your browser. It is currently labeled as in development, so features are still being added.
Can I use it without a VRoid character?
Yes. The tool includes two bundled VRoid characters (Female 01 and Male 01) and a 3D mannequin, so you can start right away. The mannequin has no face, so it cannot use expressions or export VRMA files.
What is the difference between the three gacha modes?
The cross-action gacha pulls one option from each of many actions and suits the moment you have no idea. The single-action gacha pulls within the current action's value ranges and automatically drops unnatural poses. The nearby gacha only nudges your current values and suits fine-tuning.
Can I use exported images commercially?
The official site says exported PNG and VRMA files can be used freely, commercially or not, and credit is up to you. When you use a VRM character, that model's own terms of use apply. The bundled characters can be used only inside the tool and cannot be extracted or redistributed.
Can I use it with ComfyUI?
The official site does not describe this use. Technically you can use a PNG as a reference image and run it through an OpenPose, Lineart or depth preprocessor to make a control image, but how well it is detected needs your own testing. VRMA is a pose file, so it cannot go straight into ControlNet.
Is there an English interface?
The language menu on the tool page lists nine languages, and English is one of them. I did not check how complete or accurate each translation is, so go by what you see on screen. The usage guide on the author's site is written in Japanese.
Is Pose Gacha worth bookmarking?
Pose Gacha is worth bookmarking for creators who want to find natural character poses quickly, especially people who draw manga or work with VRoid characters. Looking at the whole picture, its point is not drawing something pretty but shortening the part where you work out the pose, which is where the time goes. Being free, requiring no sign-up and keeping your data in the browser also make it easy to slot into an existing workflow.
It is still in development, though, so save the poses that matter, and if you want to pair it with an AI tool like ComfyUI, read the terms and run your own tests first. It does not draw for you; it saves you the time of figuring out the pose. The value of the gacha is not luck, but that it throws out the unnatural poses before you ever see them.