How to Use Multiple AI Services in One FMX Workflow

FMX for Windows can start several AI services from one captured photo and wait for all of them at a single AI Processed gate. This creates richer experiences without adding a separate loading screen after every AI request.

You can also place an AI Processed gate between services when the second service must process the first service’s result. The placement of the gate determines whether the services share the original source or form a sequential chain.

Terminology: The Workflow Builder does not use a separate visible screen called AI Processing. The wait is created with an AI Processed trigger, normally on an Animation step that acts as the guest’s loading screen.

Before You Start

Prepare:

• A workflow that creates a Raw Photo before any photo-based AI service
• The required AI service configurations and prompts
• Enough Cloud AI credits for every service that will run per guest
• A loading animation, transparent animation, or branded waiting screen
• A plan for which AI result the guest will preview, place in a layout, or share

Every AI service request is billed and processed separately. Running three services for one guest normally creates three AI jobs.

Choose the Correct Pattern

Goal	Workflow pattern	Source used by the AI services
Create several alternatives from one capture	Several AI services, then one AI Processed wait	The same frozen captured photo
Apply one AI transformation after another	AI service, AI Processed wait, next AI service, another wait	The later service can use the prior AI result
Do not keep the guest waiting for a long video	Start AI video, then show a QR handoff	The guest retrieves the result later

Pattern 1: Several AI Results from One Photo

Use this pattern when you want several independent interpretations of the same guest photo—for example, a cartoon, a professional headshot, and a themed AI portrait.

Build the main path like this:

Take Photo → AI Service 1 → AI Service 2 → AI Service 3 → Loading Animation → Preview or Selection

Configure the connection that leaves the Loading Animation to use the AI Processed trigger.

When FMX reaches the AI services:

1. The first service freezes the current source photo for this AI batch.
2. Each following AI service before the gate uses the same frozen source, even if an earlier result finishes first.
3. FMX registers every service as part of the active batch.
4. The Loading Animation remains active until all jobs in that batch settle.
5. The workflow continues only after the shared AI Processed wait is satisfied.

This prevents a fast service from advancing the workflow while another result is still processing.

Let the guest choose an AI result

If the workflow produces several AI Raw Photos, add a guest-choice step after the AI Processed gate. For example, Drag & Drop Layout can expose AI Raw Photos so the guest can place the preferred result into a layout.

See How to Use Drag & Drop Layout in FMX for Windows.

Pattern 2: Sequential AI Chaining

Use a sequential chain when AI Service 2 must transform the result produced by AI Service 1.

Build the path like this:

Take Photo → AI Service 1 → Loading Animation 1 → AI Service 2 → Loading Animation 2 → Preview

Set the outgoing connection from each Loading Animation to AI Processed.

The first gate closes the original source batch. The AI service after that gate begins a new batch and can consume the available AI result instead of returning to the original capture.

Do not remove the gate between sequential services. Without it, both services belong to the same batch and intentionally use the same original source photo.

Configure the Loading Animation

1. Add an Animation step after the AI services that belong to the batch.
2. Select a branded animation, or leave the media empty to use the built-in transparent animation.
3. Set the outgoing trigger to AI Processed.
4. Make sure the animation can remain on screen for the full processing time. FMX loops an AI Processed gate animation while jobs are still running.
5. Connect the successful path to Preview, Drag & Drop Layout, Save Final Photo, or another appropriate step.

For animation options, see Animations in the Workflow Builder.

Configure AI Error Routes

Configure an error route for each AI service when the guest must receive a clear recovery path.

A practical error route can:

• Show a friendly “processing unavailable” animation
• Offer a retry path
• Continue with the original Raw Photo
• Return to a Selection Screen so the guest can choose another experience

If a routed AI job fails, FMX stops the affected batch and follows that service’s error route. A service without an error route cannot send the guest to a dedicated recovery screen, so test failure behavior before the event.

Long AI Video Processing

Some video models can take several minutes. Seedance 2.0 may take more than three to four minutes, which is usually too long for an unattended guest to remain at a loading screen.

For these services, use a handoff workflow:

1. Capture the guest’s source photo.
2. Start the AI video service.
3. Present a QR Code or collection method that lets the guest leave the booth.
4. Deliver or display the result after processing completes.

Do not add an AI Processed wait merely to keep the guest at the booth for the entire render.

Test Before the Event

Run at least these tests:

1. Confirm that every AI service receives the intended source.
2. Confirm that the loading screen remains until the slowest service finishes.
3. Confirm that all desired AI outputs are available after the gate.
4. Test one failed service and verify its recovery route.
5. Review AI credit use for one complete guest run.
6. Repeat the workflow with a second session to confirm that results never carry over between guests.

Troubleshooting

Preview shows the Raw Photo instead of AI output

• Make sure the AI Processed gate is after the AI service.
• Do not connect Preview immediately after an AI service without a waiting animation or trigger.
• Verify that the Preview or layout is configured to use an AI result where required.

The second AI service ignores the first result

Add an AI Processed gate between the services. Services placed before the same gate share the frozen original source by design.

The loading animation ends too soon

Confirm that its outgoing connection uses AI Processed, not Playback Ended or Time.

The guest waits several minutes

Check the selected AI model. Use a QR handoff for long-running video services rather than keeping the guest at an AI Processed gate.

AI costs are higher than expected

Count every AI service that runs in one guest path. A multi-AI workflow spends credits for each submitted service, including alternatives the guest does not ultimately select.