Tsumugi's🔨WishMallet (Web) Help

1What this app can do

• With one photo and a one-line description, you can create a short video with a voice and background music (BGM). Without a photo, you can also create a video from the description alone.
• The face, hairstyle, age, gender and build are carried over from the photo. The outfit is chosen automatically to suit the scene.
• You can also “transform” the person in one photo. Just enter an age or a gender in “Guided writing”, or tick one checkbox, and you get a video of a younger self, a different gender, a baby, a 100-year-old, a cat, an anime character and more (→ Transform the person in one photo).
• No photo at hand? No problem. Sample photos and voices are included from the start. There is also a photo of a 50-year-old man (JP_Mail_50.jpg) that makes it easy to try transforming.
• Videos are made on your GPU Server (Runpod, or this GPU Server if the Windows app is running). With the built-in LLM (the default), descriptions are not sent to OpenRouter. Only when you switch formatting to OpenRouter in Settings is that text sent over the internet (photos, voices and videos themselves are never sent). We do not store your Runpod API key or your videos.
• Videos are saved on the GPU Server and backed up to this browser automatically. After you stop the server, you can still review them in History (→ Backups).
• You can also finish a video before you publish it: hide faces and license plates, add mosaic for stores, delete frames you do not need, and lay mask images over it so only part of it moves. You do all of this on the “Edit video” screen, and it creates one more video for sharing while the original is kept.
• You can also make a “motion comic” where comic panels move and talk one at a time, from a single comic page. It is perfect for vertical shorts on TikTok and similar apps (→ 🎞 Make a motion comic).

You need a browser with internet (PC, Mac, iPhone or Android), a WishMallet sign-in, and a GPU Server to create videos (your own Runpod account is recommended; a Windows PC with the paid Windows app can also share its GPU Server). You do not need a powerful graphics card in the device you browse on.

G⚡ Start the GPU Server (the key step)

In the web version, videos are made on a GPU Server (a powerful computer). Your browser gives instructions and backs up finished works into this browser.

Recommended: start a Runpod GPU Server “from this screen”

1. Open the ⚡ GPU Server tab (it is the first tab when you open the app).
2. Use top-right 🔑 Sign in with your email and the 6-digit code (even without a paid license, the first sign-in registers you on the free plan: videos have a watermark, and you can start the GPU Server once per day).
3. Follow step ① to create a Runpod account and add credit (Open Runpod).
4. In step ②, create an API key in Runpod Credentials → API Keys (permission All) and paste it here.
5. In step ③, choose how you want to use it (the default “take my time” is fine). Or pick a GPU yourself and press OK.
6. Press Start GPU Server. The first start can take several to about fifteen minutes while software and models are pulled. Progress and an ETA are shown.
7. When the top bar says the GPU Server is running and the badge shows “● GPU Server connected”, you can make videos.

If this PC has the Windows app: “Use the GPU Server on this PC”

On a Windows PC with Tsumugi's🔨WishMallet (Windows, paid) running, you can connect the browser to that PC's GPU Server. There is no Runpod charge. The Windows app copies works to the Windows save folder.

1. Start the Windows app first and wait until it shows ready.
2. In the web app: ⚡ GPU Server → “Use the GPU Server on this PC” → Connect to the Windows app.
3. If the browser asks to allow access to devices on the local network, allow it.

While starting or running

• The rented GPU name and approximate hourly cost appear.
• ⏸ Stop engine: stops only the video engine to free GPU memory (the server keeps running; the next video starts with engine prep).
• On the right you always see 🧹 Clear inputs, 💡 Tips, and ❓ Help (also usable on phones without horizontal scrolling).

B🗂 Backups in this browser

A Runpod GPU Server is temporary. Stopping or recreating it deletes works and history on the server. The web version therefore backs up finished works to this browser automatically, and the history and album show them from here.

• Press 🗂 Backup in the top bar for counts, size, back up now / ZIP / load / delete.
• Backups exist only in this browser on this device. Other devices and browsers are separate. Private mode clears them when closed. Clearing browsing data also deletes them.
• To keep them for good or move PCs, use ⬇ Save as ZIP (the Windows app's data sharing also works).
• While a video is being made, Backup is disabled (it returns when generation finishes).
• Works on this PC's GPU Server (connected to the Windows app) are copied by the Windows app, so no browser backup is taken.

2The screens

The diagram below is a simplified view of the screen when the “Create” tab is open. The red circles match the explanations under the diagram.

• ACurrent status: top-right “● GPU Server connected / not connected” and “🔑 Sign in / Signed in / Free plan”. If not connected, open the GPU Server tab and start it. While creating: “● Creating”; while formatting: “● Formatting”.
• BTabs: switch screens (table below). The web version starts with ⚡ GPU Server.
• CRight end of the top bar: always shown on every tab (table below). While running, the bar also shows cost and stop controls.
• 1① Reference pictures: add a photo of the person you want in the video, and that person's voice (→ ① in 3.).
• 2② Description: write the video you want. If you cannot, use 🧭 Guided writing or ✨ Auto prompt (→ ② in 3.).
• 3③ Reference videos: videos used as a reference for motion. You can create a video without them (→ ③ in 3.).
• 4Settings: choose the detail, length, shape, quality and so on. How long it will take is shown at the bottom (→ 4.).
• 5Create the video: press it to start. 💾 Save to history next to it keeps what you entered in the history without creating a video (bring it back later with 🔁 Load these settings in “History”). 📤 Export is for advanced users (→ 7.).

Use the tabs along the top to switch screens.

While the “Create” tab is open, the following buttons appear at the right end of the tabs (💾 Save to history is at the bottom of the “Create” screen, next to Create the video).

Hover the mouse over a button or an input box to see a short explanation. When in doubt, hover first.

3How to create a video

Open the “Create” tab and work from top to bottom.

① Add a photo (reference picture)

• Press + Add. At first, the folder with the sample photos and voices opens, so you can try the app just by choosing one photo you like.
• To use your own photo, go to its folder and choose it. From then on, the folder you used last time opens (after 🧹 Clear all inputs, it starts from the sample folder again).
• Besides photos, you can also choose a voice or sound (MP3 and so on) or a video (MP4 and so on). A sound alone becomes a “Sound only (no picture)” slot, and you can add a photo later with + Add a picture (just as you add a voice with + Add a sound after adding a photo first). A video opens the “Choose a picture and a sound” screen, where you pick one moment of the video as the photo and a part as the voice (up to 5 seconds); both go into one slot together. All of these work the same as drag and drop in the table below.
• Use + Add a sound in the photo's slot to attach that person's voice. The samples include voices whose names match the photos (for example 01_tsumugi.jpg and 01_tsumugi.mp3, or 05_aoi.jpg and aoi_jikoshokai.mp3); choose them as a pair and the character speaks in that person's voice (the sample voices speak Japanese).
• We recommend starting with one photo. The more photos you add, the longer the video takes to finish. You can also create a video from the description alone, without any photo.

You can also drag and drop files onto the ① area. What happens depends on what you add:

• You can use up to 3 voices and sounds in total, each 2–15 seconds and 15 seconds in total (a reference video with sound counts as one). Anything longer than the video length is not used.
• Drop another photo or sound onto a slot to replace it.
• Files you add are copied into the input folder inside your save folder and used from there, so it is fine to delete or move the originals (📁 Input folder opens it).

② Describe the video you want

Write the video you want to see in the description box. A single line is enough (for example A fantasy story). There are three ways to write it.

🧭 How to use Guided writing

Press 🧭 Guided writing at the top right of the box and a screen with the following three columns opens. Answering only what you know is fine (anything can stay empty).

• When you are done, press ✨ Format prompt. The description is written for you and goes into the ② box (with the built-in LLM, about 30 seconds to 2 minutes; the first time only, it takes a few minutes longer because it downloads the parts it needs, about 5 GB).
• When you add background music, set the video length to 5 seconds or more.
• What you typed stays even after you close the guide. The next time you open it, you can continue from where you left off. To start from empty, press 🧹 Clear all inputs inside the guide.
• To close it, press ✕ Close at the top right (clicking outside does not close it).
• Japanese quotation marks (「 」) work too when you write yourself, and words inside them are always kept character for character.

✨ How to use Auto prompt

• ✨ Auto prompt rewrites the text in the box into English wording that suits video generation. Even a few words are enough: it fills them out into a description for that video (→ 9. Handy tips). The spoken lines in quotes are kept in their original language.
• If you do not like the result, press Undo to go back (Redo cancels the undo).
• To stop partway, press Stop to the right of the waiting message. You can press the button again once the app has confirmed that everything has stopped (your description stays as it was before formatting).
• When the description is very long (for example when the same long explanation is repeated in every scene), it is translated into English scene by scene before it is formatted, so it takes a few minutes longer. If it is still too long, an error appears, and from 📋 How to convert it with ChatGPT or Gemini you can copy text to paste into ChatGPT or Gemini, which converts it into the official format.

Transform the person in one photo (Guided writing)

You can make the person in the photo younger, change their gender, or turn them into a cat or an anime character. No difficult description is needed. Just fill in “① Characters” in 🧭 Guided writing as shown in the table below and press ✨ Format prompt.

③ Add a video as a reference for motion (only when you need it)

• Just add a video file (MP4 / MOV / MKV / WebM) with + Add under “③ Add reference videos if you want”. You can also drag and drop it there. The video serves as a reference for motion and scenery.
• You can add up to 3 reference videos, each 2–15 seconds and 15 seconds in total. Anything longer than the video length is not used.
• If the video has sound, that sound is used as a reference too. It counts toward the limit of 3 together with the voices in ①.

To use a video as the source of a “face and voice” rather than as a motion reference, add it to ① instead of ③ (→ the “What you add to ①” table above).

④ Choose the detail, length and quality

At first they can stay as they are (the starting values are detail 0.9, length 1 second, shape Landscape 16:9 and quality 20, which is “Speed”). If you are not sure, just press one of the purpose buttons under “Quality” (Fast preview to Best). For details, see 4. Detail, length and quality, and how long it takes.

⑤ Create the video

• Press Create the video at the very bottom. The screen shows the current stage (checking your files → starting the video engine → creating the video → saving the finished video), the progress (%), the time left and the expected finish. The graphics card usage is shown on the right.
• The first time, or after you have not used the app for a while, starting the engine takes a few minutes.
• While you wait, you can look at the other tabs (History, Settings). The video keeps being made.
• To stop partway, press Stop, then Stop it on the confirmation screen. It stops within a few seconds (on the free plan, a stopped video does not count toward your daily limit).
• If you try to close the app while a video is being made, a confirmation appears. Press Stop and quit to discard the video being made and quit (press ✕ Close to stay).
• When the video is ready, you can use ▶ Play, ⬇ Save video and 🔁 Make another like this right there. The video is kept on the GPU Server and backed up to this browser.

4Detail, length and quality, and how long it takes

This table shows at a glance which items affect the “time (⏱)” and the “video memory (🧠)”. If a video fails for lack of memory, lower a 🧠 item; to wait less, lower a ⏱ item.

On a smartphone, scroll the table above sideways to see the ⏱ and 🧠 columns.

The fast model for Turbo is installed together with the rest during setup. On a PC that still has an older setup, turning Turbo on downloads about 19.5 GB (the progress is shown next to “Turbo”).

When you add background music, set the length to 5 seconds or more

Video length (seconds)

When you use spoken lines, ambient sound and background music all together, 5–6 seconds works reliably. This caution also appears when you hover over “What about background music?” in Guided writing (from app 1.12.37).

Seed (Random / Same as last time / Custom value)

Even with the same description, the framing and motion change every time you create a video. The number that decides this is called the “seed”.

• Random (the initial setting): you tend to get a different result every time. Good when you want to see various ideas.
• Same as last time: the result tends to stay close to the previous framing and motion. Use it when you want to change only the description or the quality a little and re-create the video with the same feel. When you choose it, the previous number appears in the box on the right.
• Custom value: creates with the whole number you enter in the box on the right (0–4294967295). Decimals and letters cannot be entered. Good when you want to note down the seed of a video you liked and use it again any time.

Even with the same seed, the result changes if you change the description, detail, length, quality, Turbo or photos. It is not for creating exactly the same video again; think of it as a way to get a similar feel.

Purpose buttons (under Quality)

How long it takes

At the very bottom of the settings (under the Seed box) you see how long it will take. There are two kinds of estimate.

• “On this GPU Server” looks for past videos with exactly the same detail, length, shape, quality and Turbo, and exactly the same number of materials (reference pictures, sounds and reference videos). When no run with the same settings is found, it estimates from the closest settings and shows “(estimated from 2 past runs with similar settings)”.
• The number of materials also changes the time. Adding reference pictures, sounds or reference videos makes it take that much longer, and the estimate changes when you add or remove materials.
• The higher the detail (the higher the picture quality), the longer it takes (it also uses more video memory).
• While a video is being created, you see “Predicted time”. Once creating the picture and sound (denoising) starts and the time per step is known, it is updated to match the actual speed at that moment. “Expected finish” shows the time of day when it is likely to finish.
• The actual time varies with the performance of your graphics card and with how other apps are being used. The first time, preparation time is added, so it takes longer than the estimate.

5Using History

The “History” tab lists the videos you have created, newest first. Type a photo's file name, part of a description or a creation date (for example 2026/9/9) in the search box at the top to narrow the list.

A failed entry keeps its “Failure details”. They are a useful clue when you contact support.

Edit video (hide faces, lay mask images, delete frames you do not need)

This screen hides “parts you do not want to show” in a finished video and removes scenes you do not need, to create one more video for sharing. The original video is not changed. All editing happens inside that PC only, and no video or image is sent anywhere.

How to open it: in “History”, press 🖼 Thumbnails for the video, then press 🎬 Edit video at the bottom left.

The first time you use “Sensitive body parts” or 🔞 Review mosaic, a screen asks to download detection data (about 12 MB). Press I have read the license, fetch it and it is ready from then on (the detection data for faces, people and license plates is included from the start). Your videos and images are never sent anywhere.

Step 1: Watch the video and pick a moment

Step 2: Hide what you do not want to show

There are three ways to hide things. You can combine all of them (adding one later does not remove the earlier edits).

You can review the hidden places with the following two drop-down lists.

Step 3: Delete frames you do not need

Things that got into the shot by accident and moments that went wrong can be removed frame by frame.

• Frames are actually removed when you save. On the screen you can bring them back at any time.
• The sound is cut at the same places, so voices and background music stay in sync with the picture (the joins are made very short and smooth).
• You cannot delete every frame (keep at least one).

Step 4: Lay mask images over the video (cinemagraph / motion comic)

What motion comics are good for and tips for making them are in 🎞 Make a motion comic.

When you lay a partly transparent image (a mask image) in front of the video, the video moves only in the transparent parts. You can make a comic page where “only the panel being read moves” (a motion comic), or a photo where only one part moves (a cinemagraph).

• Use transparent PNG / WebP images. If an image has no transparent parts, you get a warning (as it is, the video is hidden where it is laid over).
• The image is stretched to the size of the video. If you make it from a frame of the video (such as a photo saved with ⭐ Save to album), it lines up exactly.
• 🛡 Privacy blur and 🔞 Review mosaic apply only to the video, not to mask images (check what is in your mask images yourself).
• When you save, you get one video that looks just like the preview (the sound stays as in the original).
• “Make a mask image from this frame” makes an image the same size as the video, so it lines up exactly. Hidden places stay hidden in the image (mask images are not blurred or mosaicked). The PNG you make also goes into the input folder.
• Images you pick are copied into the input folder inside the save folder, just like reference images. The last mask images you set (images and positions) are remembered for each video, and the next time you open “Edit video” for that video, it starts in the same state (closing the screen does not lose them). If input already has an image with the same name, it is replaced by the one you picked.

Step 5: Save it as a video for sharing

• Press Save video for sharing at the bottom right to save it as a separate file. The original video stays as it is.
• A mark is added to the end of the file name: _privacy for Privacy blur, a name for each store for Review mosaic, and _edit when there are no hidden places (only deleted frames or mask images).
• You can save with frame deletions or mask images alone, even if there is not a single place to hide.
• While the video is being created, you can stop with Cancel. The original video is unchanged and the unfinished file is deleted.
• The finished video appears in “History” under the original video as a 🛡 For sharing row. You can Delete them one at a time.
• To close the screen, press ✕ Close at the top right (you cannot close it while a video is being created; cancel first).
• 🎁 On the free plan you can mark places to hide, delete frames, lay mask images and check the result, but Save video for sharing is a full-version feature (pressing it shows how to buy).

AI can miss things. Before you publish a video, always watch it yourself from start to finish and check it.

🎞 Make a motion comic (bring comic panels to life one at a time)

A “motion comic” is a video of a comic page where only the character in the panel being read moves and talks. With this app, you can turn a single comic page into a vertical short video where the panels come to life one after another.

Why use mask images?

Even if you ask the video AI (MiniMax H3) to “move only this panel”, when the same character's face appears several times on the page, the faces in other panels may move or lip-sync too. To the AI, every face looks like the same person.

That is where mask images help. When you lay a “still picture” with only the panel you want to move made transparent over the video, the video shows only inside that transparent panel and the other panels stay still. Prepare one mask image per panel, line them up in order, and you have a motion comic whose panels move one at a time.

How to make one (4 steps)

For step-by-step operations, see “Step 4: Lay mask images over the video” in 5. Using History.

Use only comics and illustrations that you own the rights to or are allowed to use (→ 11. Using the app safely).

6Best shots

A place to collect “favorite single frames” from your videos as photos. Photos are added with 🖼 Thumbnails → ⭐ Save to album in “History”.

• Click a photo to show it large. Turn to the previous or next photo with the ‹ › buttons on the left and right, the ← → keys or a mouse drag. Your position is shown below, like “3 / 10”.
• In the large view you can use ▶ Play this scene (plays the original video from that moment), 💾 Save full quality (enlarges it with AI for your purpose and saves a PNG; details below) and 🗑 Remove from album.
• Press ☑ Select to pick several photos, then 💾 Save all as ZIP or 🗑 Delete selected.
• Use “Sort” (Newest first / Oldest first) and “Video” (which video the photos came from) at the top to narrow down the photos shown.
• 📁 Album folder opens the folder that holds the photos (bestshots in your save folder).

💾 Save full quality (enlarge with AI for your purpose)

Press 💾 Save full quality on a photo in the large view, and a screen asks “What will you use it for?” so you can pick a purpose. Just pick one: the AI enlarges that moment of the original video to just the right size and saves it as a lossless PNG.

The sizes are for portrait. For the master image and the posters and art, you can also choose landscape (for example 10800 × 7200) under “Orientation” (for a landscape video, it starts as landscape).

• Picture type: choose 📷 Photo / live action for real footage and 🎨 Anime / illustration for anime or illustrations for a cleaner result.
• Finish: not sure? Choose 🌿 Natural (recommended). Eyes and mouths stay natural and close to the original video. ✨ Crisp sharpens fine details, but small faces may look slightly distorted around the eyes and mouth.
• It takes a few seconds to about a minute (about 10 seconds to 2 minutes with “Natural”). Press Stop to stop partway (no half-made file is left behind).
• Images for books and posters are saved as PNG files with the print setting (300 PPI). A master image or poster is tens of MB per image, so keep an eye on the free space where you save. The first folder offered is the exports folder in your save folder. ⬇ Save folder / Save, to the left of “💾 Save”, opens the folder you last saved to.
• When an image is enlarged a lot, as for a poster, fine details may look slightly soft up close (the screen tells you when this applies). The more detail you create the original video with, the cleaner the result (the difference is biggest for small faces).
• You cannot press 💾 Save while a video is being created or the description is being formatted (they all use the graphics card). Save after it finishes. Likewise, while it is saving, you cannot start creating a video, Auto prompt or auto formatting.
• The image is automatically split into smaller pieces to fit the free graphics memory, so it can be saved even on PCs with little free memory (it may take a little longer).
• The parts for AI enlargement come with the app. Nothing extra is downloaded, and your images are not sent over the internet.
• If the original video has been deleted or moved, the album photo is used instead (the quality may be a little lower).

🎁 On the free plan you can still open 💾 Save full quality and see the list of purposes and the preview. When you press the final 💾 Save, it shows how to get the full version. 💾 Save all as ZIP is also a full-version feature. Saving to the album and viewing it work.

7Export (for advanced users)

The instructions and messages in exported files follow the Japanese or English language selected in Settings. Your prompts, material names and file names are kept as entered.

This chapter is for people who want to re-create a video on another PC or on a cloud service. You do not need it for everyday use.

📤 Export is a button that writes the settings you are using now (description, which photo is which, detail, length, quality and so on) into one file. Open the file in another environment and you can re-create the video with the same settings. It is not a button that exports the video itself (the video is in your save folder).

• “Create” screen: at the right end of the row of buttons at the very bottom (to the right of Create the video and 💾 Save to history). Exports the settings that are on the screen now.
• “History” screen: in each entry's row of buttons. Exports the settings used when that video was made.

🎁 On the free plan you can open the export screen and check its contents. Export (writing the file) is a full-version feature.

Three kinds of files

When you open the Export window, “③ For Google Colab” is selected at first (to export for ComfyUI, choose another one under “Format”).

How to use them

1. ① ComfyUI (standard): copy the exported .json to the PC with ComfyUI, and put the photos and voices into ComfyUI's input folder with the same names. Open ComfyUI in a browser, drag and drop the .json onto the screen and run it.
2. ② ComfyUI (high quality): open the ComfyUI screen on Floyo, drag and drop the .json, upload the photos and voices, and run it. If “Missing Models” appears, follow Floyo's guidance to get the parts.
3. ③ Google Colab: in Google Colab, load the .ipynb from “File → Upload notebook”, choose a GPU (G4 is the best choice; if you cannot select it, A100) under “Runtime → Change runtime type”, turn on “High-RAM” if the option is there, and choose “Run all”. When it pauses partway, upload the photos and voices with the names shown. When it finishes, the video is downloaded to your PC automatically.

Choosing a GPU in Google Colab

A guide to which of the GPUs offered under “Change runtime type” suit this app. For video creation, the amount of video memory matters most (→ How do they compare with home cards?).

How do cloud graphics cards compare with home ones?

This helps you decide when “my own PC is too slow” or “I am not sure whether to buy a new graphics card”. The first thing to know is that what matters in video creation is “the amount of video memory” rather than speed.

The full set of AI parts this app uses is about 42.5 GB even in its smallest configuration. Whatever does not fit in video memory is calculated while being swapped to and from the PC's main memory each time, and that waiting makes it slower. Even the top home card has 32 GB of video memory, so a home graphics card cannot hold the full set of parts. This is the biggest difference from the cloud.

On a smartphone, you can scroll the table above sideways.

The larger the detail, the longer the video and the more photos you use, the more this difference shows. In order, they rank as follows.

RTX 4090 < RTX 5090 < Colab (A100 40 GB) < Colab (A100 80 GB) < Colab (G4) / Floyo (H100 NVL)

Which one should you choose?

• Your PC is not powerful enough, or videos fail for lack of memory → try ③ Google Colab first (G4 is the recommended GPU). Even for long videos or large detail, the full set of parts fits as it is on Colab's G4 or on ② Floyo (high quality).
• You are thinking of buying a new graphics card → we recommend trying the cloud before you buy. Even the top home card (RTX 5090, 32 GB) cannot hold 42.5 GB, so the “swapping” does not go away completely even after you upgrade. You can experience how comfortable it gets first, and then decide. For the cost side, also see “💡 Could a GPU actually cost you less?” below.
• You already have ComfyUI on another PC → ① ComfyUI (standard). Where that PC's video memory falls in the table above tells you directly what to expect.
• Your PC does not meet the requirements (for example, it has no NVIDIA GPU) → during installation the app tells you what is missing. If you choose to go on (“Yes”), the app is installed anyway, and with Export (a full-version feature) you can make files for ③ Google Colab or ② Floyo and create videos on an outside GPU (the video-creation setup on that PC is skipped).

Sources of the figures: the 42.5 GB full set of parts is from the official ComfyUI blog; Floyo's H100 NVL (94 GB, 3.9 TB/s) is from Floyo's official information; the RTX 5090 (32 GB, 1.792 TB/s), A100 and L4 (24 GB, 300 GB/s) are from NVIDIA's official specifications, and Colab's G4 (NVIDIA RTX PRO 6000 Blackwell, 96 GB) is from Google Cloud's official specifications (as of October 2026). Besides video memory, there are also differences in how AI calculations are done and in hardware generations, so please treat the comparison above as a guide.

When Google Colab stops or shows an error

Even if an error appears in red, the work done up to that point is kept. You do not need to start over. Press the ▶ to the left of where it stopped and run just that block (cell) again.

At the very bottom of the notebook there is also 10_Troubleshooting. Please take a look there first too.

If it still does not work (contacting support)

Prepare the following three things and send them to the support contact written in your purchase email. The cause can then be found quickly.

1. The error text: copy the last few lines of the red text (the line that starts with “❌” and the lines below it are enough). A photo of the screen is fine too.
2. The run log: press the folder icon at the left edge of Google Colab, right-click wishmallet_studio_export_report.txt and download it. The log up to the point where it stopped is saved automatically.
3. The exported file: the .ipynb you used, itself. It records when and with which version it was made.

If the app itself also seems to be misbehaving, please also send the ZIP made with Save support information under “Settings” → “Support” (→ 10. Troubleshooting). It does not include photos, voices, videos, descriptions or your activation code.

Exported files go into the exports folder inside your save folder. In another environment the graphics card is different, so the picture will not be exactly the same. Lowering the length or detail makes failures less likely.

8The Settings screen

The “Settings” tab lists the following, from top to bottom.

🔄 Updates

When “Check for updates at startup” is ticked, the app checks for a new version every time it starts (it is ticked at first). Check now checks by hand.

🌐 Language / 表示言語

Switches between Japanese and English. It changes the moment you choose. This page, which opens from ❓ Help and 💡 Handy tips, and the check tool that opens from Check PC performance are also shown in the language you chose.

🔔 Sound effects

When “Play sound effects” is ticked, a short sound plays once when a video is finished or “Format prompt” ends (on success and on failure), and when something needs your check. It is ticked at first. Untick it and no sound plays.

🔑 License

Shows your sign-in status and email. Sign in from the top-right “🔑” button. To register with a purchase code (WMS-…), use Enter a code under Settings → License. A Windows license works on the web via the purchase email (→ Free plan & license).

💻 Your PC

Shows the connected GPU Server. Choose the GPU type in the ⚡ GPU Server tab.

💽 Storage

Free space, and the space used by videos and temporary files. Clean up temporary files deletes files that are no longer needed (finished videos and the history are kept).

📁 Save location

The folder where finished videos go (at first Videos\MiniMaxH3). Change picks another folder and Open save folder opens it. Inside it, samples holds the sample photos and voices, docs the guide, input copies of the files you added, exports exported files and bestshots the album photos (samples and docs come back when you restart the app if you delete them).

⏸ Auto shutdown of the video engine

If a certain time passes without creating a video, the mechanism running in the background is stopped automatically to free the graphics card's memory. When you want to use the graphics card for another app, press Stop now.

🩺 Diagnostics

Check the video engine checks whether the connected GPU Server is ready to create videos. If something is wrong, stop and start the GPU Server again. Use ⚡ Open the GPU Server tab to jump there.

🗂 History

Restore history from the save folder restores videos that disappeared from the list, using the save folder.

📮 Support

When you are in trouble, press Save support information to make a file (ZIP) for investigation. It does not include photos, voices, videos, descriptions or your activation code.

ℹ️ About

The current version, Open the guide (opens Read me first.html, the getting-started guide in your browser) and Take the tour (shows the short first-start walkthrough again).

🤝 Affiliate program

Apply for the program that pays you a commission for introducing this app, and see your current status (not applied / pending / registered rate). See “Affiliate program” below for details.

For the “LLM used for formatting” setting (choosing the AI that Auto prompt and Format prompt use), see Format with OpenRouter (sign-up and creating an API key).

🎁 Free plan & license

Sign in with the top-right 🔑 Sign in using your email and a 6-digit code. Even without a paid license, the first sign-in registers you on the free plan.

• You can still open the 🔒 features and see what they can do (for example, the list of purposes and the preview in 💾 Save full quality). When you press the final button, it shows what the feature does and how to buy.
• The number of videos is counted on our server, so you need to be connected to the internet during the trial. A video you stop partway is not counted.
• For support, during the trial the app sends your PC's configuration (CPU, GPU, memory and Windows version) to our server (model names and sizes only; no PC name or other personal details). Your photos, voices, videos and descriptions are never sent.
• After you enter a code (“Settings” → “License” → Enter a code), there is no limit on days or videos.

Affiliate program (introduce this app and earn a commission)

Introduce this app on social media, a blog or in videos, and earn a commission on every sale made through your own link. Sales and payouts are handled by Gumroad (the store). You apply from “Affiliate program” at the bottom of Settings.

Amounts are “price × rate” as a guide. What you actually receive changes with Gumroad's fees, refunds and so on. At the 10% rank, 10 sales add up to the price of this app (9,600 yen).

How to apply

1. Create a Gumroad account and finish its payout settings.
2. In the app, open Settings → “Affiliate program” and press Apply (available once you have registered a full-version license; on the free plan you can only read about the program).
3. Enter your Gumroad account email, a contact email and your name or handle, then press Submit. Your status becomes “Application pending”.
4. When we have registered you on Gumroad, Gumroad sends you your affiliate link and the app shows “Registered (10% commission)”.
5. When you reach 10 or 100 sales, press Request a rank-up in the same place. We check the Gumroad sales records and change your rate to 30% or 50%.

Your current status (not applied / pending / registered rate) is checked automatically each time the app starts. Press Refresh status to check right away. If you typed the wrong Gumroad email, you can correct it with Edit application while the status is “Application pending”.

Questions about the affiliate program: affiliate@aicouturelab.com

🌐 Format with OpenRouter (sign-up and creating an API key)

By default, ✨ Auto prompt and “✨ Format prompt” in 🧭 Guided writing use an AI that runs inside your PC (the built-in LLM) to polish your description. If you switch to OpenRouter in Settings, formatting becomes much faster. This chapter walks you through everything, from signing up to setting up the app, so that even a first-time user can follow it.

① What is OpenRouter? What you need, and how it differs from the built-in LLM

OpenRouter is an internet service that lets you use many different AIs through a single door (it is run by a company other than the maker of this app). This app asks one of them, deepseek/deepseek-v4.1-flash (DeepSeek V4.1 Flash), to polish your description.

What you need

• An OpenRouter account: signing up is free. You can sign up with Google, GitHub or an email address (→ ②).
• Credits: a prepaid balance from which only what you use is deducted. One formatting run costs about $0.0007, and with a card, buying $15 or more at a time is the better deal (→ ③).
• An API key: a long string, like a password, that lets the app use OpenRouter (→ ④).
• An internet connection: needed every time you format.

How it differs from the built-in LLM (the default)

② Sign up for an OpenRouter account

1. In your browser, open https://openrouter.ai/.
2. Press Sign in at the top right of the page.
3. Choose whichever is easiest for you: Google, GitHub or an email address (if you have a Google account, that is the easiest).
4. Follow the on-screen instructions. If you sign up with an email address, also confirm the email that is sent to you.
5. When the OpenRouter pages open while you are logged in, you are done.

③ Add credits

With OpenRouter you buy credits (a balance) first, and only what you use is deducted. Without credits, formatting stops with “not enough credits”.

1. Open https://openrouter.ai/settings/credits (if you are not logged in, the login page appears first).
2. If you pay by credit card, we recommend adding $15 or more in one purchase. A fee is charged on every credit purchase, so one purchase of $15 or more costs less than several small ones (see the note below). How many runs $15 covers (a guide): formatting costs about $0.0007 per run, so $15 covers about 20,000 runs on average. With short descriptions only, that is about 28,000 runs; even with long descriptions of about 3,000 characters every time, it is about 9,000 runs. Long descriptions or many scenes mean fewer runs, and the number also changes with the provider and with the time. When it runs low, just add more.
3. Enter a payment method (such as a credit card; choose from the ones OpenRouter shows) and complete the payment.
4. When the balance has gone up, you are done.

About the fee: a fee is charged on every credit purchase. For credit cards it is 5.5% of the amount (with a $0.80 minimum), according to OpenRouter's FAQ (as of October 2026). Because of the minimum, a purchase under about $14.55 pays a higher percentage than 5.5%. From $15 up the rate stays at 5.5%, so with a card, buying $15 or more at a time is the better deal. Fees and payment methods can change, so what OpenRouter's own pages show is the correct information.

④ Create an API key

1. Open https://openrouter.ai/settings/keys.
2. Press Create API Key.
3. In Name, enter a name you will recognize later (for example: WishMallet).
4. We recommend setting a limit (Credit limit). With a limit, even if someone else used your key by mistake, no more than that amount could be spent.
5. Press the button that creates the key.
6. A long string that starts with sk-or-v1- (the API key) appears. Copy it, for example with the copy button.

⑤ Set it up in the app

1. Open the Settings tab in the app and find “✨ LLM used for formatting”.
2. In “Type”, choose OpenRouter (deepseek/deepseek-v4.1-flash).
3. Paste the key you copied in ④ into the API key box (Ctrl + V).
4. Press Save. The saved key is shown after “Saved:”, and the box is emptied.
5. Press Test connection to check that it connects. When it says “Connected”, you are ready (it does not format anything, so it costs nothing).
6. From now on, press ✨ Auto prompt or ✨ Format prompt as usual.

• To choose the provider yourself (optional): the Provider box to the right of “Type” normally stays empty (it shows “Cheapest”), and the app automatically picks the cheapest of the providers that do not store data (ZDR). If you enter a provider name such as DeepInfra, only that provider is used, with no switching to others even when it is unavailable (formatting then fails with an error). Providers that do not support ZDR cannot be used. Use the provider name shown under Providers on the model page on OpenRouter (https://openrouter.ai/deepseek/deepseek-v4.1-flash) (ASCII letters, digits, spaces and . _ / + - only, up to 64 characters). It is saved when you press Enter or leave the box. Empty it to go back to the cheapest.
• To go back to the built-in LLM, just choose it again in the same place.
• To remove the saved key, press Delete key. This removes only the key saved in this browser. The key itself stays on OpenRouter, so if you no longer need it, delete it there too (on the keys page).
• The key is stored encrypted in this browser. It is not included in the settings file or in the support information.
• You cannot change the choice while a video is being created or a description is being formatted (please try again after it finishes).

The names of buttons and boxes may differ slightly depending on the version of the app.

⑥ If it does not work

When an error appears, please also read the “How to fix it” steps shown on the screen.

⑦ Privacy: what is sent and what is protected

When you format with OpenRouter, unlike the built-in LLM, your description and similar text are sent outside your PC. Please use it only after you understand the following.

• What is sent: the description being formatted (the few words you wrote for “Auto prompt”, or the description Guided writing built), the content of the scenes, and the names (file names) of the reference materials. Your photos, voices and videos themselves are never sent (only the file names are used). As a precaution, also check that your file names do not contain real names.
• Where it goes: OpenRouter, and the provider that runs the model. The app asks to use only the cheapest of the providers that do not store data (ZDR: Zero Data Retention) (if you enter a name in “Provider” in Settings, only that provider; the no-data-storage condition stays the same). A provider's policy is based on information from OpenRouter and can change.
• OpenRouter itself does not record prompts (your descriptions) or replies by default. Whether to allow recording is something you choose in your account's privacy settings. It is reassuring to open https://openrouter.ai/settings/privacy once and check that you have not allowed it.
• Before you use it, please read OpenRouter's terms (https://openrouter.ai/terms) and privacy policy (https://openrouter.ai/privacy) and make sure you agree.
• Depending on the content, it may be refused (403 in ⑥). In that case, switch to the built-in LLM.
• Do not write personal information that you do not want others to know, or business secrets, in the description. For that kind of content, it is safer to format with the built-in LLM (it never leaves your PC).

The wording and layout of OpenRouter's pages may change without notice. If a page looks different, look for a button or item with a similar name. For fees, charges and conditions, OpenRouter's pages and terms are the correct information.

9Handy tips

⭐ Get a prompt for the video you want from just a few words (✨ Auto prompt)

Write the video you want to see in a few words and press ✨ Auto prompt. The AI rewrites it into a prompt (description) in the officially recommended format for making that video. No photos or voices are needed.

The AI plans scenes that fit the length and shape you set and fills out the description. Reading the result shows you how a description should be written to get the video you have in mind. Once you are used to it, add places, people, movements and lines, or add photos and voices, to get closer to exactly what you imagine.

Recommended steps

1. Set the conditions first. To start, a very short video is best: Detail 0.5 / Length 1 sec / Shape Landscape 16:9 / Quality 20. It finishes quickly, so you can soon see whether your words become the video you imagined.
2. In “② Describe the video you want”, write the video you want in a few words and press ✨ Auto prompt (with the built-in LLM, 30 seconds to 2 minutes; the first time takes a few minutes longer because the needed parts are downloaded).
3. When the prompt is ready, press Create the video.

If formatting takes a long time, you can switch “LLM used for formatting” to OpenRouter in Settings to finish in much less time (it needs a sign-up and a small fee; → Format with OpenRouter).

For example, you can start from words like these.

• A beautiful female samurai quietly raises her sword in a storm of cherry blossoms
• A beautiful angel girl stands by the shore of a lake
• A private beach with white sand. Girls in bikinis are enjoying a swim in the sea
• A cat in a spacesuit happily runs across the surface of the moon
• A giant dragon slowly flies above a sea of clouds
• A giant whale swims over an ancient underwater city
• A supercar racing across a post-apocalyptic wasteland makes a big drift
• A cute young witch draws a giant magic circle in the night sky
• A giant robot looks up at a rainbow in a city after the rain
• A giant castle floats above the clouds. A flock of white birds flies over a beautiful girl standing at its entrance

⭐ With your own photo: “I want to look younger!”, “I want to be an anime character!”, “I want to change my gender!”

All you need is one photo and 🧭 Guided writing. You do not need to write a description yourself. Start by setting the video length to 1–2 seconds so that you see the result quickly.

Common steps

1. On the “Create” tab, press + Add and choose one photo in which the face is clearly visible. A photo of just one person, facing the front, from the chest up is recommended. You can also try the sample JP_Mail_50.jpg.
2. Press 🧭 Guided writing at the top right of the description.
3. Fill in “① Characters” as described for each goal below.
4. In “Where are they and what is happening?” under “② Scenes”, write the place and the camera direction (for example: “They are sitting on a park bench. The camera shows the face from the front.”). Type a line in “What are they saying?” to make a talking video (it can be left empty).
5. Press ✨ Format prompt, and when the description is ready, press Create the video.

“I want to look younger!”

1. In “Age, gender and occupation?”, enter the age you want to be as a number only (for example 16), the gender (for example male) and the occupation (for example Japanese).
2. In “Features”, a short note of the look you want gets you closer (for example high school student).
3. Do not tick “Non-human” or “Anime”.

“I want to be an anime character!”

1. Enter the gender and occupation as they are for the person in the photo (for example male and Japanese). Age can stay empty (if you enter it, the character is of that age).
2. Tick “Anime”.
3. With several people in the video, tick only the person you want as anime. The other people and the background stay as in the photos, and only that person becomes anime. Tick everyone and the whole video becomes anime.

“I want to change my gender!”

1. In Gender, enter the gender you want (for example female).
2. Enter a number in Age too (your current age is fine). If Age is empty, the video is made with the look in the photo.
3. Write the hairstyle and outfit in “Features” and “What are they wearing?” to get closer to that look (for example long straight hair and high school uniform).

Other things you can do

The same list is also in 3. How to create a video, “Transform the person in one photo”.

When you use a photo of someone other than yourself, always get that person's permission (→ 11. Using the app safely).

⭐ Make a “motion comic” from a comic page

From a single comic page, you can make a vertical short video where only the panel being read moves and talks (a motion comic). It is a popular format that can go viral on TikTok, YouTube Shorts and Instagram Reels.

Even if you ask the AI (MiniMax H3) to “move only this panel”, other panels may move too when the same character's face appears several times on the page. Lay mask images, with only the panel to move made transparent, over the video in 🎬 Edit video, and the panels move one at a time.

⭐ I cannot write a good description

Press 🧭 Guided writing at the top right of the description on the “Create” screen. Just answer the questions and the description is written for you.

• In “① Characters” you answer “Age, gender and occupation?”, “Features” and “What are they wearing?” for each photo; in “② Scenes (scenario)” you fill in “Where are they and what is happening?”, “What are they saying?” and the subtitles for each scene; and above “③ The description” you fill in “What about background music?” for the whole video. When you add background music, set the video length to 5 seconds or more (under 5 seconds the video may be silent). Each character is shown as the face icon of its photo; hover over it to see its name (for example <Subject 1> = the person in <Picture 1>). Press a face icon at “Where are they and what is happening?” to insert that name into the text.
• Use “+ Add a scene” to add scenes in the order things happen, and “+ Add a line” to line up spoken lines. If you leave “Where are they and what is happening?” empty from scene 2 onward, the description is built with the same text as the previous scene (it continues the previous scene). If you type even one character, only what you typed is used. For a spoken line, “What are they saying?” appears to the right of the face icon, and you type the words in the box below it (several lines are fine). Press the face icon to choose the speaker. For someone who is not in a photo, choose “Other” and type a name. The resulting description appears right away in “③ The description”.
• Anything you do not know can stay empty. Only what you answered becomes text.
• Enter a number in “Age” and the person in the photo is created with the look of that age (the facial features still resemble them; if you also enter the gender and occupation, the result follows them). If it is empty, the look in the photo is kept.
• To make something other than a human, such as a dog, a cat or a robot, tick “Non-human” under the occupation and write something like “Shiba Inu puppy” in “Features”. The age is not used (write puppy, adult dog and so on in “Features”). Only the expression and the mood are taken from the photo, and the whole body from head to toe becomes that animal. Tick “Anime” too and it becomes an animal drawn in anime style.
• Tick “Anime” and only that person is drawn as a Japanese 2D anime character (line art and cel shading) whose face resembles the photo. The other people and the background stay as in the photos. Tick everyone and the whole video becomes anime. If you enter an age, the character is of that age.
• The words you type in “What are they saying?” are spoken exactly as written. Tick “Show subtitles” and those words also become the subtitle (to show a different subtitle, type it in the subtitle box to the right of “Show subtitles”).
• You are not asked about voices or reference videos. They are used automatically.
• Press ✨ Format prompt and the description is written. If you do not like it, press Undo to go back. You can redo it as many times as you like.
• What you typed stays even after you close the guide or quit the app. The next time you open it, you can continue from where you left off, so you can add to it little by little. To clear everything, press 🧹 Clear all inputs inside the guide (you will be asked to confirm).

⭐ I want to keep “one high-quality photo” without spending much time

Create a short 1-second video at good quality and keep one frame of it as a photo. It takes only a few minutes.

1. Create a video with these settings: just one photo (no voices or reference videos) / Detail 2 (1888x1088px: Full HD) / length 1 second / any shape you like / quality 20 or more.
2. Open 🖼 Thumbnails in “History”, drag the strip at the bottom to find a good moment, and press ⭐ Save to album.
3. Open the photo in “Best shots”, press 💾 Save full quality and pick a purpose (not sure? “⭐ Master image”). The AI enlarges it to just the right size and saves it as a PNG (details in 6. Best shots). 🎁 On the free plan you can see the list of purposes and the preview.

⭐ The video does not look good

1. First, set the quality to Best and create the video again.
2. If that is still not enough, raise the detail as far as you can. With a length of 1 second, even a low-performance PC can try up to Detail 2 (1888x1088px: Full HD).
3. If it stops for lack of memory, buying a new graphics card is one option. But it is an expensive purchase, so before that, please also see “⭐ Try the cloud without spending money” below and 7. Export, “How do cloud graphics cards compare with home ones?”. Even if you upgrade to the top home card, the full set of parts does not fit.
   💡 On the other hand, if you create a lot, buying a GPU can cost less than paying for the cloud again and again → Could an expensive GPU cost less?

⭐ I want to re-create a video I like with a small change

You like the framing and motion, but you want to fix only the outfit or the spoken line, or raise only the quality. In such cases, use the “seed”.

1. On the “History” tab, press 🔁 Make it again for that video.
2. The settings come back to the “Create” screen, and the seed is set to “Same as last time”.
3. Change only what you want to fix (the description, the quality and so on) and press Create the video.

You will not get exactly the same video. Think of it as getting close to a similar feel. Conversely, when you want to see various ideas from the same description, leave it on “Random”.

⭐ Try the cloud without spending money

Write out the same settings with 📤 Export and re-create the video on a cloud service (→ 7. Export). You can start both without spending money. See “How do cloud graphics cards compare with home ones?” for how powerful a graphics card you can rent.

The Floyo code is the one published on Floyo's official site as of September 2026. It may end at the distributor's discretion.

10Troubleshooting

Look for the heading that matches. On every screen, “How to fix it” with numbered steps appears under an error; start by following those.

🩺 Diagnostics fails or you cannot connect to the GPU Server

First check the ⚡ GPU Server tab for connection status and any start error. Stopping and starting the GPU Server again is the usual fix. Follow the numbered steps under Settings → Diagnostics when shown.

🚫 “Create the video” cannot be pressed

• Check that the description in ② is not empty (a video cannot be made without one).
• Look for a message in red (up to 3 voices and sounds, each 2–15 seconds and 15 seconds in total; up to 3 reference videos).
• You cannot press it while a video is being created or while the description is being formatted. Please wait until it finishes.
• If the free plan says “You have used today's free videos”, you can create 3 more when the date changes.

🧠 Videos fail for lack of memory

1. Close other apps that use the graphics card (browsers, games, video players and so on).
2. On the failure screen, press Try again with detail … (lowers the detail) or Try again at … seconds (shortens the video). The “…” is the value the app proposes.
3. Reduce the number of photos. With a reference video, use about half your usual detail or length (a reference video needs about twice the memory).

⚡ The GPU Server will not start or stops mid-way

The GPU Server container cannot see the NVIDIA GPU. Stop it and start again; on Runpod try a different GPU type. With a local Pod, restart this PC.

1. Restart your PC.
2. Reinstall the NVIDIA driver (580 or later) from NVIDIA's website, then restart your PC.
3. On Runpod, try a different GPU type. With a local Pod, restart this PC.

🧩 “Guided writing” or “Auto prompt” stops with an error about a missing component

1. Stop the GPU Server in the ⚡ GPU Server tab, start it again, then retry.
2. When it asks “Check and repair now?”, type y and press Enter (the AI models are not downloaded again; it takes about 5-15 minutes).
3. When it finishes, reopen the app and try again.

If the black window does not open when you press the button, you can do the same thing by right-clicking setup_sglang_env.ps1 in the setup folder inside the app's install folder and choosing “Run with PowerShell”. Setup cannot start while a video is being created, a description is being formatted or a picture is being saved in full quality; wait until that finishes.

Even if this feature is unavailable, it does not affect creating videos. If you are in a hurry, switch “LLM used for formatting” in Settings to OpenRouter and you can carry on (to set it up, see Format with OpenRouter).

🌐 An error appears when formatting with OpenRouter

What to do for each message (wrong key, not enough credits, content refused, too busy and so on) is summarized in Format with OpenRouter, “⑥ If it does not work”. If you are in a hurry, switch “LLM used for formatting” in Settings to the built-in LLM and you can carry on.

📝 “The description is too long to translate and convert into the MiniMax H3 official format here.” appears

It exceeds what the formatting AI in this app can handle at once. Repeating the same long description in every scene reaches the limit quickly.

1. Shorten the description, or write the parts repeated in every scene only once, and try again.
2. To use it as it is, press 📋 How to convert it with ChatGPT or Gemini under the error, copy the text that appears with 📋 Copy and paste it into ChatGPT or Gemini. Paste the resulting text into the description box, then press Create the video.

⚡ “Create the video” cannot be pressed even though Turbo is on

If the fast model for Turbo is not on this GPU Server yet (for example, a PC with an older setup), about 19.5 GB is still being downloaded. The progress is shown next to “Turbo”; please wait until it finishes.

🌐 “The remaining free videos could not be checked” appears

On the free plan, the number of videos is counted on our server. Check that you are connected to the internet and press Check again.

🔄 The update fails

Check that you are connected to the internet, wait a while, and then try “Settings” → “Updates” → Check now.

📤 An exported file gives an error in another environment

How to fix Google Colab when it stops, and what we would like you to send us, are summarized in 7. Export, “When Google Colab stops or shows an error”.

💾 “Save full quality” fails

1. Close other apps that use the graphics card, such as games or video playback.
2. Press Stop now under “Settings” → “⏸ Auto shutdown of the video engine”, then try again.
3. If it still fails, update the NVIDIA driver.

If it says “The files for AI enlargement are missing.”, reinstall the app (run the installer again). Until then, you can still save a PNG with the “📷 Original size” purpose.

📮 If it is still not solved

Under “Settings” → “Support”, write what the problem is and press Save support information. Attach the ZIP file it makes and contact the support address written in your purchase email.

11Using the app safely

When you prepare photos, voices and videos

• Use only your own, or ones that the owner has given you permission to use.
• When you use the face or voice of a real person, get that person's consent.
• Do not use material that shows minors.

License and disclaimer

• The video AI (MiniMax H3) is provided under the MiniMax H3 Community License. There are regions where it cannot be used and purposes it must not be used for. Please check the “Terms of use and AI conditions” screen shown at first start and MiniMax-H3-LICENSE.txt in the install folder.
• Because the AI creates videos automatically, the result may not be what you intended.
• You are responsible for the content of the finished videos and for how you use them.

12Glossary in plain words