AI Online Studio
Pay with Built-in tokens or your own fal.ai / WaveSpeed key, and build AI Online templates that turn guest photos into new scenes and short videos.
AI Online Studio sends the guest's photo and your prompt to an online AI model, and the finished picture (or a short video) comes back to the booth in seconds. You get the latest AI models without a powerful graphics card; in return you need a stable internet connection, and every generation is charged by the AI service.
Chapters (13)
- 0:00Intro · how AI Online works
- 0:29Built-in: no API key · tokens and prices
- 0:55Buy tokens on lumaqube.com
- 1:15Tokens added · refresh the balance
- 1:27Your own fal.ai or WaveSpeed key
- 1:45Build a template: output and model
- 1:59Test photo and a universal prompt
- 2:29Process and Compare
- 2:42From image to video
- 3:02Cover and Save to File
- 3:12Add it to an event
- 3:26Easy Mode · at the event
- 3:54Three things to remember
Chapters (12)
How AI Online works
- The guest takes a photo, Q-Booth sends it with the template's prompt to the AI model you chose, and the result is shown and shared like any other result.
- The AI runs online, so the booth computer does not need an RTX card. For AI on the booth computer itself, see AI Offline Studio.
- You choose how to pay under PROVIDER in the studio:
- Built-in (since 5.7.0.0): the QBooth Cloud AI service. No API key and no extra account. It uses tokens from your Q-Booth account; the studio shows your balance and what one generation costs before you press the button (see Built-in tokens).
- 3rd Party: your own account with fal.ai or wavespeed.ai, connected with an API key (steps below). After picking 3rd Party, a second row (Service) lets you choose the service. The service charges your account directly for each generation.
- Since 5.9.2.0, a new studio starts on the provider and model you picked last time, even after a restart. The very first time it starts on Built-in where it is offered. Opening a template doesn't change what is remembered.
- A saved template reopens on the provider it was saved with. The exception (since 5.9.2.0) is a Built-in template opened on a booth without Built-in (a dongle licence on 5.9.6.0 or older, or a rebranded edition without Built-in): it opens on fal.ai, on the same model when fal.ai has it. If fal.ai doesn't have it, no model is picked and Process says the model isn't available: pick one and save. Templates downloaded from the Template Gallery switch the same way.
Costs differ per model, quality, video length and options. Q-Booth shows the cost of one generation under the model, so you always know before you process. Video costs much more than a picture.
Built-in tokens
Built-in works on a cloud licence and, since 5.9.6.1, on a dongle licence too. A dongle booth uses the tokens of the lumaqube.com account the dongle is registered to; there is nothing to set up. If the dongle isn't registered to an account, Built-in says "This dongle isn't registered to a lumaqube.com account, so it can't use Built-in tokens. Contact support with your Dongle ID." Up to 5.9.6.0, a dongle licence doesn't show the Built-in | 3rd Party row at all, only fal.ai / wavespeed.ai. Some rebranded editions of the app don't offer Built-in. No Built-in? Update the app: 5.7.0.0 or newer on a cloud licence, 5.9.6.1 or newer on a dongle.
The balance card
When Built-in is selected, a balance card appears under PROVIDER:
- The big number is your token balance. Next to it you see tokens, or free trial tokens on a trial licence, or tokens (frozen) if the balance has been frozen (contact support).
- The round-arrow button refreshes the balance and the recent list. The balance also refreshes by itself after every Process.
- Buy tokens opens your account page on lumaqube.com, at the token page.
- Recent Generations lists the last 20 images this booth made with tokens. Each row shows its tokens, or not charged, or a status such as failed, running or expired. Open generations page goes to the full history on the web.
Under the model you see what one generation costs, for example "N tokens per image", or for a video "N tokens · 5 s video". The number follows your settings: a longer video or a higher quality costs more. Prices change from time to time, so always read the number on screen: the token price of each model shows under the model and in the Select Model list. Token packs are on the Products page.
How tokens work
- One balance is shared by all the licences on your account, dongles included. Top up once and every booth of yours can use it.
- Tokens are only taken when a result is delivered. A failed generation says "The generation failed (…). No tokens were charged."
- If the template makes a picture and then a video, Q-Booth checks that you have enough tokens for all the steps together before it starts, and charges nothing if you don't.
- A trial (demo) licence gets free trial tokens to try Built-in. There is nothing to buy on a trial, so Buy tokens and Recent Generations are hidden. When the trial tokens run out, Q-Booth says so: get a full licence to keep using Built-in, or use your own fal.ai / WaveSpeed key.
Buying tokens
- In the studio press Buy tokens, or log in to your account on lumaqube.com and open Cloud AI › Tokens in the member menu. The member Shop has the same packs under AI Tokens, with the same Confirm your order step.
- The Cloud AI Tokens page shows Your balance and how many licences share it.
- Under Buy tokens, pick a pack and press Pay.
- Check the order (Confirm your order) and press Continue to payment. Payment goes through Midtrans and is billed in Rupiah.
- As soon as the payment is confirmed, the tokens are added to your account. Every top-up and every AI generation is listed under Token history.
- Back in Q-Booth, press the refresh button on the balance card.
If you closed the payment window, the unpaid order shows with Pay now / Cancel order. Pay now reopens that order's payment window (with the way to pay already chosen, if you chose one). An unpaid order disappears by itself when its payment link expires (24 hours). There is no Check button: the page re-checks unpaid orders with the payment provider each time it opens, so a payment that hasn't shown up yet appears on your next visit. If you close the payment window before choosing a way to pay and press Pay on the same pack again, the same order reopens instead of a second one. Token purchases need a licence or a dongle on your account. Dongle-only accounts can buy tokens too; the dongle shows under Shared by N licences, tagged (dongle).
The member menu also has Cloud AI › Generations: every AI result your booths made with tokens, newest first, with filters by licence, status and date. Old results are removed after the storage period; the row stays but shows "Result removed after the storage period".
Get a fal.ai API key
Only needed for 3rd Party. You do this once per computer.
- Open fal.ai in your browser and click Login. The easiest way is to sign in with your Google account.
- Add credit to your fal.ai account (in fal.ai's billing page). Every image or video you generate uses credit from that account.
- Click API Keys in the menu, then Add key.
- Type a description if you like, for example Q-Booth Online Studio, and click Create Key.
- Copy the key immediately and keep it somewhere safe.

fal.ai shows the key only once. After you close the window you cannot see it again; if you lose it, create a new key. Treat the key like a password: anyone who has it can spend your credit.
Enter the API key in Q-Booth
- Open Settings › General and scroll to AI Online Service.
- Under Your own API keys, paste the key into the fal.ai row (or the WaveSpeed row for a WaveSpeed key). Both rows are always shown (since 5.8.0.0). Get a key opens that provider's key page. Tick Show key if you want to check what you pasted.
- Click Save Settings.
- In AI Online Studio, choose PROVIDER › 3rd Party and the matching service.
Older versions have one API Key box: pick the provider in the AI-Online API Provider dropdown first, then paste its key.

Keys are stored separately per provider. If a template uses a model from the other service in the list, enter that service's own key as well; otherwise processing stops with "API key for … is not set". When you open an event whose templates need a key that is missing, Q-Booth also asks for it before the event starts. See also General settings.
The studio screen
The screen has three parts: the left panel with template-wide settings (see Left panel settings), the middle panel with the AI itself, and the canvas on the right with the tabs Source Image, Result and Compare. At the bottom are Reset, Process (F5) and Share. The left panel folds away to a thin strip with the arrow at its top right, and you can resize the panels by dragging the lines between them.
Since 5.9.6.0 the middle panel is one page with no tabs. From top to bottom: OUTPUT, MODEL TYPE, then the picture settings (provider, model, prompt, and for the models that use them References, Motion Reference Video and Background). Below them is the optional video step, added with Turn into video. A section that is missing simply isn't supported by the chosen model.
On 5.9.5.0 and older these are tabs: Model (always), plus References, Motion, Background and Video Chain depending on the model.
Model
The top of the middle panel (the Model tab on 5.9.5.0 and older):
| Control | What it does |
|---|---|
| OUTPUT | Image or Video. |
| MODEL TYPE | For images: Image → Image (starts from the guest's photo), Text → Image, Others. For videos: Image → Video, Text → Video. |
| PROVIDER | Built-in (tokens) or 3rd Party (your own API key; choose the Service). |
| MODEL | Opens Select Model (see below). Each model has its own look, speed and cost; the cost of one generation is shown under the model. Next to the model button, Info opens the model's web page (not shown for Built-in models). |
| ASPECT RATIO | The shape of the result. Auto leaves it to the model and the photo. When you take the source photo with the camera, the camera also crops it to this shape. Hidden for models that don't accept a shape (many video models). |
| Video Duration | Video models only (since 5.6.2.0): how long the video is. A slider for models that accept any length in a range, a dropdown for models with a few fixed lengths. Longer costs more. |
| Generate Audio | For video models that can add sound effects, ambience and speech. Its tooltip says it costs the same with or without audio. |
| PROMPT | Basic (one fixed prompt) or Dynamic (guests fill in blanks, see Dynamic Prompt). Prompt Library stores prompts you want to reuse. Prompt Builder (Basic only, since 5.8.0.0) builds the prompt by picking from lists, see Prompt Builder. Hidden for models that take no prompt. |
| Model settings | Extra settings for some models, for example GPT SETTINGS with Quality (Low/Medium/High), which directly changes the price, and on some GPT image-edit models Fidelity (Low/High: how closely the result keeps the original photo's details, such as faces; High costs much more). On Built-in, Quality lists only the qualities the service sells, and the token price follows your pick. |
Built-in models can show locked settings (with a lock icon): settings the service fixes for that model. You can't change them.

The Select Model window
- It lists only the models that match OUTPUT, MODEL TYPE and PROVIDER, with a count ("N models") and a Search models... box.
- Each row shows what differs between models: the price (tokens for Built-in, or what your own fal.ai / WaveSpeed account pays), multi-image (it accepts several pictures), no prompt (it takes no text), and for videos the lengths it can make (for example "5-10s").
- Info on a row opens that model's page on the provider's website.
- Pick a row and press Select, or double-click it.
- The model list updates itself (since 5.7.0.0): new models can appear, and old ones disappear, without an app update. It is checked about once a day; press Refresh models to get the newest list right away. The text next to it says when the list was last updated.
Under the chosen model you see its price. For Built-in that is the token price ("N tokens per image"). For 3rd Party it is what one generation costs on your own account (since 5.6.2.0); ≈ means an estimate and N/A means the cost isn't known yet. Video models add a second line with the lengths they can make, for example "Duration: 5 / 10 s".
If the provider removes a template's model, the template opens with no model selected and Process says "This template's AI model is no longer available. Open the template and choose another model." A model that is only switched off for a while shows "This template's AI model is temporarily unavailable…"; the template still saves and runs again once the model is back.
Prompt Builder
Since 5.8.0.0, the Prompt Builder button next to Prompt Library (Basic prompts only) builds the prompt by picking instead of typing. You can pick how much to change, keep the face, outfit, expression, pose, background, event theme (wedding, birthday, Lebaran, Christmas…), lighting, time, weather, mood, camera framing and angle, background blur, art style, colours, and text to write into the picture. Every choice has Custom... for your own words. The prompt appears on the right as you pick; press Use This Prompt to put it in the prompt box.
- Prompt Language: the prompt is written in the app's language by default (English, Bahasa Indonesia, Filipino, German, Chinese or Spanish), and you can switch it. Words you type yourself stay as typed. If a model follows one language badly, try another.
- Paragraph / List: the same prompt in two layouts. Both give the same result.
- Lighting is three rows: one light source, one brightness, and any number of effects. So it can't ask for "bright" and "dark" at once.
- When the model takes reference images, each one gets a "use it for" choice: its outfit, as the background, its art style, its pose, add this person, add this object, or Custom. The prompt then points at it as @Image2, @Image3… (see @ references).
- Copy (under the prompt on the right) copies the prompt, to paste it somewhere else.
A prompt made with the builder is locked in the prompt box ("Made with Prompt Builder"). To change it, press Prompt Builder again: your picks are saved with the template. Unlock turns it into normal text you can edit by hand, but the picks are gone for good. Picking a prompt from the Prompt Library also unlocks it (the app asks first).
On a template whose prompt was typed by hand, Prompt Builder first asks whether to continue: the builder can't edit typed prompts, so it starts from scratch and its prompt replaces the typed one.
Prompt Builder is for image models only. The button is hidden on a video model and for the video step's Motion Prompt: video prompts describe motion (what the camera does, what moves, what stays frozen), which the builder's picks can't say. For guests answering questions, use a Dynamic prompt and its Dynamic Prompt Editor instead.
References
A fold-out section under the prompt (since 5.9.6.0); it shows how many files it holds even when folded. On 5.9.5.0 and older it is the References tab (since 5.6.2.1; before that the tab was called Image References and held images only). Shown only for models that accept several pictures. These files go to the picture model only; the video step has its own.
- Image References: extra pictures sent along with the guest's photo, so the model follows a style, costume, product or composition. Press Select Image (Camera, Clipboard, Gallery, File or a recent file); every press adds another image. Remove one with its X.
- Video References and Audio References: only for models that accept them. Add Video (mp4 / mov) and Add Audio (mp3 / wav). Each row shows the file, its length and a small preview player.
Limits for reference videos and audio: each clip about 1.8–30 seconds; all videos together up to 30 s, and all audio together up to 30 s; up to 10 of each. Reference video time is billed too, and the cost line includes it.
Reference files are saved inside the template, so the template works on another PC.
Since 5.9.4.0, big reference pictures are shrunk automatically to 2048 pixels on the long side, when they are added and when an older template is opened. The AI works at about that size anyway, so nothing visible is lost. On an older version, if generations fail with a connection error after you add a reference picture, use a smaller copy of that picture (2048 px is plenty).
@ references: pointing at a specific picture
Since 5.8.0.0, for models that take several pictures (marked multi-image), you can tell the model which picture you mean, by number:
@Image1is always the guest's photo (in the video step: the AI picture from the step before).- Reference images under References are
@Image2,@Image3and so on. Reference videos are@Video1,@Video2…, reference audio@Audio1,@Audio2….
Example: Put the person from @Image1 in the outfit from @Image2, in the setting of @Image3.
- Type @ in the prompt and a list pops up with the photo and each reference file. Pick one to insert its name. The hint under the box says "Type @ to point at the photo or a reference file."
- Each reference card shows its number as a small badge (for example @Image2). Click the badge to copy it.
- Removing a reference renumbers the others in your prompt automatically, so each name keeps pointing at the same file. If the prompt still uses the file you're removing, the app refuses: "@Image2 is used in the prompt. Take it out of the prompt first, then remove the file."
Dynamic prompts are not renumbered automatically.
Motion Reference Video
Only for "Motion Control" video models (since 5.5.3.0). They take a video of someone moving and make the person in the guest's photo perform the same movement. Since 5.9.6.0 it is a section right under the model; on 5.9.5.0 and older it is the Motion tab.
- Motion Reference Video: pick the clip (mp4, mov, webm, m4v or gif). It is saved inside the template. Use a clip with one person, whole body visible.
- The clip decides the result's length and shape. The line under the model shows the allowed clip length; a clip under 3 s is billed as 3 s, and a clip that is too long is refused.
- You pay per second of the result, so a longer clip costs more.
A Motion Control template can't be saved or processed without a clip: "Please select a motion reference video."
Camera Motion
Since 5.9.2.0. Only the H3 Max Camera Controls video model lets you edit the camera path, on Built-in and on fal.ai, under OUTPUT Video › Image → Video. The guest's photo becomes a video where only the camera moves: people and objects stay frozen, like a "bullet time" shot. It also works as the video step (the camera flies around the AI picture). Other video models have no camera editor; describe the camera move in their prompt instead.
Under the model a CAMERA MOTION card shows the current move, for example "Orbit 360° right" or "Custom path · 5 keyframes". Click Edit Motion to open the editor.


| In the editor | What it does |
|---|---|
| Presets | Along the top: Dolly in, Dolly out, Spiral in, Crane up, Orbit 360° right / left, Half orbit right / left, High orbit right, Low orbit right, Orbit 360° spiral in, Swing. Click one to start from it. |
| 3D view | Drag the purple camera to place it, scroll on it to move it closer or further, drag the background to look around. The Camera view box in the corner shows what the camera sees. |
| Timeline | Each diamond is a keyframe: where the camera is at that moment. Click a diamond, then move the camera, the dial or the sliders to change it. Moving the camera between two diamonds adds a keyframe there; drag a middle diamond to change its timing. Keyframes sets how many (2–12), Remove keyframe deletes the selected one. |
| Azimuth / Elevation / Distance | Around the subject (full turns count), up or down (−90° to 90°), and how far (0.1 close to 2 far; 1 = the photo's own framing). |
| Play | Previews the move. The striped end of the timeline is the video's last second, where the camera holds its final position. |
| Reset / OK / Cancel | Reset starts a blank path (camera standing still). OK keeps the changes, Cancel throws them away. |
A new template on this model starts on Orbit 360° right. Length: 3–10 s on Built-in, 1–15 s on fal.ai. The prompt is optional: empty means "keep everything frozen, only the camera moves".
Big moves (a full orbit, looking from behind) make the AI invent the parts the photo never showed. Test them before the event. On an app older than 5.9.2.0 the Built-in model shows without this card and the AI picks the move itself: update the app to choose it.
Background
Only for a background-removal model (under Others). A section under the model since 5.9.6.0; on 5.9.5.0 and older, the Background tab. Original keeps the cut-out with a transparent background; Change Background puts it on a Background Image you pick. Without that image Process refuses to run: "Please select a background image."
Turn into video: make the AI picture move
A second AI step that runs after the first: the image model makes the picture, then a video model makes it move, in the same run. Shown for image models that support it.
Where it is (since 5.9.6.0): press Turn into video, the dashed button under the picture settings. A purple 2 Video card appears below, under an arrow that says "AI picture goes into step 2", and the picture settings become card 1 Picture. Remove on card 2 takes the step away again.
- There is a shortcut under OUTPUT › Video too: a hint asks "Want the guest's AI picture to move, not the raw photo?", and its AI picture → video button switches to a picture model and adds the video step.
- The video step starts on the same provider as the picture (Built-in with Built-in).
- Under the card, a Per guest line shows what one guest costs: the tokens of both steps and their total when both are on Built-in, tokens plus the fal.ai amount when they are mixed, or the dollars added up when both use your own keys.
On 5.9.5.0 and older it is the Video Chain tab (the last tab): turn on Generate video from the image result. The settings below are the same.
| Control | What it does |
|---|---|
| Turn into video / Remove | Adds the video step after the image model, or takes it away (5.9.5.0 and older: Generate video from the image result). |
| PROVIDER / MODEL | The video model to use: Built-in (where it is offered), fal.ai or wavespeed.ai. Any image step can chain into any of these. Its price and lengths show under it. |
| Aspect Ratio / Video Duration / Generate Audio | When the video model supports them. You pay per second. |
| Motion Prompt | A short motion prompt: what moves and how the camera moves. It has its own Prompt Library. Many video models refuse an empty prompt, so Q-Booth stops you first: "Video prompt is empty. Describe the motion in step 2 (Video) on the Model tab…" (5.9.5.0 and older: "…Open the Video Chain tab and describe the motion…"). |
| IMAGE RESULT IS THE: First Frame / Last Frame | Since 5.9.1.0, only for video models that can end on a given picture: on Built-in Kling 2.5 Turbo Pro and Gemini Omni Flash (same token price as usual); on fal.ai / wavespeed.ai also Seedance 2.5, Seedance V1 Pro and Minimax Hailuo-02. First Frame (the default, and every older template) is the usual chain: the video starts on the AI picture. Last Frame: pick a Start Frame picture; the video starts on it and ends on the AI picture. Without a start picture the template can't be saved: "Please select a start frame picture in step 2 (Video)…" (5.9.5.0 and older: "…on the Video Chain tab…"). |
If the video model takes several pictures, an Image References section appears here too (plus video and audio references for models that take them). If it is a Motion Control model, a Motion Reference Video section appears.
The same choice is under the model when you pick one of these video models directly (OUTPUT Video › Image → Video), as INPUT IMAGE IS THE: First Frame / Last Frame. There, Last Frame makes the video end on the guest's photo; without the start picture you see "Please select a start frame picture under the model…"
You get both the AI picture and the video. The video is the main result and is shown first (since 5.5.3.0). Video takes longer and costs more than a picture, so test before the event.
Want the guest's AI picture to move? Use Turn into video. Picking OUTPUT Video › Image → Video is different: it moves the guest's raw photo, with no AI picture first. That mix-up is the usual reason the AI photo doesn't move.

Example: a "being drawn" video
Since 5.9.1.0, Last Frame can make a video of a hand sketching the guest's portrait:
- Take a photo of a blank sheet of paper on a table, from straight above. This is the Start Frame.
- Image model: a multi-image model (for example Nano Banana). Add the blank-sheet photo as an Image Reference, and prompt it to draw the guest as a marker or pencil portrait on that paper, keeping the paper, table and framing exactly the same.
- Video step (Turn into video): Kling 2.5 Turbo Pro (10 s) or Seedance 2.5, Last Frame, the same blank-sheet photo as Start Frame, and a motion prompt such as
overhead static camera, time-lapse of a hand sketching the portrait in pencil, then inking, then colouring with markers; the paper never moves.
If the paper in the AI picture moves or changes size, the video morphs instead of drawing. Fix the image prompt first.
Build a template step by step
- Open Studio › AI Online.
- In Model choose OUTPUT › Image and MODEL TYPE › Image → Image, so the guest's photo is the starting point.
- Choose the PROVIDER and click MODEL to pick a model. Note the cost shown under it.
- On the right, load a test photo in Source Image with Select Image (file, camera, clipboard or gallery).
- Write the prompt (example below) and set ASPECT RATIO.
- Press Process (F5). A few seconds later the result is in; check it with Compare.
- Optional: press Turn into video under the picture settings (5.9.5.0 and older: Video Chain tab › Generate video from the image result), choose a video model, a duration and a short motion prompt, and process again.
- Optional: under Options turn on Before → After Video (photo hold, result hold and transition times can be set).
- Set a frame in Overlay, fill in Caption and Cover image in Event, and click Save. AI Online templates are saved as
.qsofiles.
Example prompt that works for one guest or a whole group:
Transform this photo into a royal winter fairytale. Dress every person as royalty in elegant regal outfits that suit each of them (ball gowns, royal uniforms or noble attire) with sparkling details, and tiaras or crowns where fitting. Place them in front of a snowy palace at dusk, with soft falling snow and warm golden lights. Keep every person's face, expression, pose and the number of people exactly the same. Photorealistic, high detail, cinematic.
Describe the scene, the clothes and the light, not the people themselves, and ask the model to keep every face the same. More in Writing good AI prompts.
Process and results
The Source Image tab is hidden for Text → Image and Text → Video models, which use no photo. You can also drag an image onto the canvas. At the event the guest's photo takes its place; the test photo is not part of the template.
Process (F5) checks the template first and stops with a Cannot Process message if something is missing, for example "No model selected.", "Prompt is empty.", "No dynamic prompt to process." or "Source image is empty.". With a dynamic prompt, the questions window opens first.
On success the canvas switches to Result:
- Generated in shows how long it took.
- With several results (for example picture + video, the original photo, a before → after video), use the arrows to step through them ("1 / 3").
- The Compare tab is a before / after slider: the source photo on one side, the AI picture on the other. It appears when the model uses a photo and there is an image result. With several image results, a Result dropdown picks which one.
- Share sends, prints or exports the result. Reset clears the whole studio to a new template.
Left panel settings
| Section | What it does |
|---|---|
| Template | Open (From Gallery, Browse File or a recent file), Save / Save As, and Notes. The top right of the studio shows whether the template is saved (Not saved yet, Modified, Saved). The template can't be saved while required parts are missing, for example a model that needs a prompt but has none. |
| Overlay | Disabled / Default (the global overlay from Settings › Overlay & Filters, or the event's own overlay inside an event) / Custom (pick a PNG for this template with Select Overlay). Since 5.9.6.1, with Custom, the Put the photo inside the frame's window switch puts the AI picture (or its video) inside the frame's window instead of under the whole overlay; see Put the photo inside a frame. The Include original photo and Before → After Video overlays have the same switch. |
| QR Overlay | Prints a QR code with the photo's share link onto the result. Default (follows the global setting, or the event's) / Disabled / Custom (drag it into place on a preview with Position & Size...). Needs Web QR to be set up, or no QR is added. Called "QR Code" before 5.5.5.4. |
| Camera Filter | Use Global / Off / Custom: the colour filter for photos taken with the camera for this template. Filters apply to camera shots only, not to files picked from disk. |
| Camera Filter › Beauty Filter | Since 5.9.0.0: Use Global (the app's beauty look) / Off / Custom (this template's own look; Configure... opens the Beauty Filter window). Saved inside the template. Guests can still turn it off. See Overlay & Filters settings. |
| Event | Caption and Thumbnail (Select cover image): what guests see for this template in the event's template picker. |
| Options › Include original photo in result | Also delivers the guest's untouched photo. When it is on, an Overlay choice appears for that photo: Disabled (the default, a clean photo), Default (same overlay as the result) or Custom. |
| Options › Before → After Video | Since 5.1.1.0: makes an extra video that starts on the original photo, transitions, then shows the AI result: Photo hold and Result hold (0.5–6 s each; when the template makes a video, the AI video plays for its own length instead of the result hold), Transition (Fade, Dissolve, Wipe Left, Slide Left, Circle Open), and its own Overlay (Disabled / Default / Custom, since 5.5.0.0, so the video can have a different-shaped frame). |
When AI Online Studio is opened from another studio or the event editor (for example from a Collage layer or Create in Studio), the left panel starts with a big Save & Return button and Cancel, and Save becomes Save to File (save a copy without closing).
Since 5.9.6.0, when the studio that opened it needs a certain kind of result, a note at the top of the Output section says so:
- Needs a picture (Collage layer, Lenticular layer, Art Studio source): Video output is greyed out and Turn into video is hidden. A template that still makes a video (one opened from a file, or a video step left on) turns the note red, and Save & Return refuses it until Output is Image and the video step is removed.
- Needs a video (Flipbook, Motion's Glam / Boomerang source): nothing is greyed out, but Save & Return refuses a template that only makes a picture. Pick a Video model, or press Turn into video.
- Motion's ordinary Studio Template clip and the event's template list take either, so there is no note.
Save & Return always hands the template back to the studio that opened it. A template you open inside this window (Open / Gallery) is used as a starting point; your file itself is not changed (since 5.9.6.0; before, Save & Return wrote the changes back into that file).
Use it in an event
- Expert mode: in the event open Templates and click Open File, or Create in Studio › AI Online and finish with Save & Return.
- Easy Mode: the AI Photo online card: choose a model, write the prompt and use Try it on a photo before you save. New Easy AI templates use Built-in tokens where available. Since 5.8.0.0, Build my prompt above the prompt box opens the Prompt Builder, so no typing is needed.
- Inside a Collage: add a layer of type Studio Template › AI Online. The template must make a still picture; one set to a video model (or with a video step) is refused with "…Collage layers must produce a still image — pick an image-output model for this layer." Its original photo and before → after video come out as extra results next to the collage. Motion, Lenticular and Flipbook can also run an AI Online template inside them.
At the event the guest picks the template and takes the photo; the AI works for a few seconds and the result is shared as usual. Dynamic prompt questions are asked on the event's Input Page, before the camera. A template's Overlay and QR Overlay on Default follow the event's own settings. Q-Booth checks templates before an event starts, so a template whose model is gone is reported before guests see it.
Running out of tokens
Since 5.8.0.0:
- When you start an event that uses Built-in and fewer than 100 tokens are left, a warning appears with Buy tokens. You can still start the event.
- If a generation is refused because the tokens ran out, the error has a Buy tokens button. In a studio it opens the token page on lumaqube.com.
- During an event it shows a QR code instead. Scan it with your phone, buy tokens, then press Retry. The guest's photo is kept, and you don't need to close the event.
Waiting times and timeouts
- Most pictures take a few seconds; videos take longer. While the AI works, the screen shows progress such as "In queue" and "Generation in progress…".
- Max Timeout (default 180 seconds) and Cancel Button Delay (default 60 seconds) are in Settings › Other (turn on Show all settings if you do not see it). See Other settings.
- If a generation takes longer than usual, Q-Booth shows Still Generating: nothing has failed yet, and Keep Waiting lets it finish.
Troubleshooting
"API key for … is not set."
In Settings › General › AI Online Service, paste the key into that provider's row (older versions: pick the provider first) and click Save Settings. Remember that each provider has its own key.
"Not enough tokens."
You are using Built-in and the balance is too low for this model (with a video step, both steps together; the Per guest line under the video step shows the total). Top up, or Remove the video step (older versions: turn off Video Chain). To top up, use Buy tokens on lumaqube.com, then refresh the balance in the studio. Nothing was charged for the failed attempt. During an event (since 5.8.0.0): press Buy tokens on the error, scan the QR code with your phone, buy, then press Retry.
"Prompt is empty." / "No dynamic prompt to process."
Write a prompt, or build one with Prompt Builder (Basic) or the Dynamic Prompt Editor (Dynamic).
"Generation failed" or the service rejects the request.
Check the credit on your provider account, the internet connection and the prompt (some models refuse certain content or an empty prompt). Try again or choose another model.
"The server took too long to respond."
The internet connection is slow or unstable. Check the venue connection, have a backup ready (a network cable or a mobile hotspot), and try again.
"This template's AI model needs QBooth X or newer."
Since 5.9.2.0: this Q-Booth is too old for that model. Update Q-Booth, or pick another model.
A template shows the token card, but this booth has no Built-in | 3rd Party row and can't switch to fal.ai.
The template was saved on Built-in, and this booth doesn't offer it (a dongle on an older version). Update the app: dongles get Built-in from 5.9.6.1. Versions 5.9.2.0 and newer at least open it on fal.ai. On older versions, make a new template and copy the prompt across.
"This dongle isn't registered to a lumaqube.com account…"
Since 5.9.6.1: contact support with your Dongle ID, so they can register the dongle to your account. Then press Retry; no restart is needed.
"Cloud AI needs a cloud licence. Dongle booths are not supported yet."
A dongle booth on 5.9.6.0 or older. Update the app (dongles get Built-in from 5.9.6.1), or use 3rd Party with your own key.
"Please select a start frame picture in step 2 (Video)…"
Since 5.9.1.0 (older versions say "…on the Video Chain tab…"): in the video step, pick the Start Frame picture, or set IMAGE RESULT IS THE back to First Frame.
Faces or the number of people change.
Add "Keep every person's face, expression, pose and the number of people exactly the same" to the prompt, and try a higher Fidelity if the model offers it.
I don't see Built-in.
Update the app: Built-in needs 5.7.0.0 or newer on a cloud licence, and 5.9.6.1 or newer on a dongle licence. Some rebranded editions don't offer Built-in; use 3rd Party with your own fal.ai or WaveSpeed key there.
"Your free trial tokens are used up."
You are on a trial licence. Get a full licence to keep using Built-in, or use your own fal.ai / WaveSpeed key.
"Video prompt is empty."
Write a short Motion Prompt in the video step, card 2 (older versions: on the Video Chain tab).
"Please select a motion reference video."
Pick a clip under Motion Reference Video below the model, or in the video step's card (older versions: the Motion tab, or the Video Chain tab's motion section), within the length shown under the model.
"Collage layers must produce a still image."
The AI Online template in the Collage layer is a video template. Use an image model with no video step (older versions: Video Chain off).
Three things to remember
- Check your balance (tokens or provider credit) before every event.
- Test each template on a real photo first, on the venue's internet.
- Using your own account? The API key goes in Settings › General.
Next steps
Still need help?
Our team answers on WhatsApp. Tell us your Q-Booth version and what you see on screen.