How an Extra becomes a video: from the owner's content to the guest showcase

A short video can make an Extra more visible, more concrete and easier to imagine.

But between the initial photograph and the clip the guest actually sees, there isn't just an automatic generation. There's an offer to present correctly, a visual source, a model to choose, a duration, a movement suggestion, an external processing step, a human review and a final publication.

It's this sequence that makes the difference.

StayFast doesn't just add a video generator to the Extras editor. It ties content, technology and distribution into a single flow: from the owner's idea to the showcase and catalogue the guest uses.

It all starts with the Extra, not the AI

Before the video comes the offer. The owner needs to be clear about:

• what they're offering;
• what's included;
• how much it costs;
• when it's available;
• how it can be requested or purchased;
• which image properly represents the experience.

A video can't fix an unclear Extra. If the title, description, conditions or price are confused, movement only draws attention to an offer the guest can't evaluate.

The owner's first role remains editorial and commercial: define a real service and present it transparently. Only then does visual content come into play.

Two paths: upload a video or start from the photo

Not every property owner works the same way. Some properties already have a good vertical clip — recorded while preparing breakfast, during a treatment, at a tasting or on a local experience. In that case the owner can upload their own video.

That's often the best choice when you need to show:

• real people during the service;
• a preparation;
• a place in motion;
• a specific gesture;
• an experience actually delivered at the property.

Other owners only have a good photograph. StayFast then lets you use that image as the starting point for a short video generated via fal.ai.

A real video when it exists; a video generated from the photo when it doesn't.

AI doesn't necessarily replace original production. It shortens the distance between owning a photograph and being able to present the Extra with moving content.

The owner picks the model that fits the photo

Not every photo needs the same kind of processing. An empty room, a set table and a person performing a massage are different problems. That's why StayFast distinguishes two models.

Pika 2.2 for people, gestures and experiences

Pika is recommended when the photo contains people, a treatment, a gesture, an ongoing activity or a scene where human movement matters: a massage, a served breakfast, a toast, a class, a tasting, someone preparing or presenting something.

The owner can pick 5 seconds for a short gesture or 10 seconds for a more complete action. Ten seconds is the default, because it more easily lets a gesture start, develop and end.

Seedance 2.0 Fast for rooms, objects and products

Seedance is recommended for rooms, spaces, terraces, products, set tables, empty spas, details and atmospheres. Here movement is mostly about the camera, depth and how the scene is presented.

Available durations are 5 seconds for a quick movement, 7 seconds as the recommended choice, 10 seconds for a more developed scene and 15 seconds for a slow, continuous movement.

The choice stays explicit. StayFast doesn't automatically switch from one model to the other after an error, because every new generation can result in a new charge on the tenant's fal.ai account.

The initial suggestion doesn't consume AI credits

Writing an effective prompt can be hard for anyone who doesn't work with generative models daily. That's why StayFast prepares an initial suggestion based on the Extra's title, description, selected model, duration and the owner's language.

The suggestion is generated locally through StayFast templates and rules. No other AI model is called, no fal.ai credit is consumed and no content is sent outside just to prepare the text.

For Pika, the suggestion focuses on the gesture and on the people already in the photo. For Seedance, it focuses on camera movement, depth and preserving the environment.

The owner can use the suggestion as is, edit it, pick a preset, restore the initial suggestion or update it after changing the model or duration. Manually written text is never silently replaced: if the owner has edited it, StayFast asks for confirmation before inserting a new suggestion.

StayFast keeps a second layer of protection

The text visible in the editor isn't the only instruction sent to the model. Before generation, StayFast adds a conservative set of server-side directions. They ask the provider to:

• preserve the photo's visual identity;
• keep rooms, furniture, colours and objects;
• not add new people;
• not introduce text;
• not invent rooms or architectural elements;
• avoid unnatural transformations and movements;
• keep a continuous scene.

This second layer doesn't replace the owner's review, but it reduces the chance that a short or ambiguous prompt produces a result far from the original image.

The owner's prompt can go up to 600 characters. The initial suggestion stays shorter, leaving room to personalise it.

The tenant uses their own fal.ai account

StayFast doesn't sell internal packages for these generations. The property connects its own fal.ai account and uses its own credit. The economic relationship stays direct between the tenant and the provider.

StayFast doesn't front the cost, doesn't top up the credit, doesn't trigger new generations automatically, doesn't hide the chosen model or duration, and shows a notice before submission.

The API key is stored securely and isn't returned to the browser after connection. Today the credential is linked per property: the same fal.ai account can be used across multiple properties, but the key must be associated explicitly to each tenant.

Generation doesn't block the editor

Producing a video takes longer than editing a normal card. That's why StayFast doesn't leave the form waiting for the full response. When the owner presses Generate AI video, StayFast checks tenant, plan and permissions, validates the source image, the model, the duration and the prompt, creates a job, sends a single request to the fal.ai queue, saves the request id and returns control immediately.

The owner can keep working, navigate away or close the editor. The job stays persistent. When the editor is reopened, StayFast retrieves the state from the database and shows whether the video is submitting, queued, in preparation, generated, saving, ready or failed. It doesn't depend only on the browser's temporary state.

Webhook and reconciler work together

Normally fal.ai signals completion through a signed webhook. StayFast verifies the signature before processing the result, records the produced video and copies it into its own Storage. This step matters: the clip must not depend permanently on a temporary provider URL.

But things happen: the webhook may arrive late, the browser may be closed, a download may be interrupted, the Storage upload may not finish, or the database update may fail after the file was already saved. For these cases there's a periodic reconciler.

The reconciler doesn't create a second generation. It can only check a request already sent, retrieve a result already produced, retry download and upload, or complete an interrupted update. This distinction protects the tenant from the risk of a second unintended charge.

Errors aren't all the same

A network error is different from an image the provider didn't accept. StayFast distinguishes at least three families of problems.

Temporary error

It may come from network, timeout, temporary limit, provider unavailability or an upload issue. In these cases StayFast can recover the same job without creating a second generation.

Unknown submission outcome

The connection may drop right as the request is being sent. StayFast doesn't know for sure whether fal.ai received the job. To avoid a possible double cost, it doesn't automatically re-submit.

Moderation block

The provider can reject an image or its content. In that case the job is closed as failed and doesn't keep cycling through recovery. The owner sees a clear message and can use a different model, edit the prompt, pick another image just for the video or upload their own clip. The Extra's original image isn't automatically removed or changed.

The result stays a preview

When the video is ready, it isn't published directly. The owner must first watch it and check that people look natural, gestures are credible, objects aren't distorted, the environment is still recognisable, no elements were added that don't exist, and the content really represents the Extra.

If the result isn't good enough, it can be discarded. The previous video, if any, stays active until the new one is confirmed. Only the Use this video button updates the Extra.

From preview to showcase

After approval, StayFast checks that the file really exists in Storage, updates the Extra's video fields, prepares the URLs needed for guest rendering, marks the job as accepted and keeps the relationship with the source image and the model used.

From that point the clip can appear on the guest showcase, in the Extras menu and in the offer's detailed card. The guest doesn't see the job, the prompt, the model or the technical steps. They just see the final result. Technology disappears behind the experience.

Movement has to stay selective

Not every Extra needs to become a video. And not every video should play at the same time. In a showcase made mostly of photos, a few animated items can catch the eye. If everything moves, movement loses value and the interface becomes noisy.

Video is particularly well suited for massages and treatments, breakfasts, romantic experiences, tastings, products, spa, local activities and services communicated through atmosphere or gesture. A photo can still work best for purely operational services, where movement doesn't add understanding or desire.

Who does what

The flow works because each actor keeps a clear role.

The owner

Creates the offer, picks the image, decides the model, chooses the duration, edits the suggestion, reviews the preview, approves or discards, and is responsible for the commercial correctness of the content.

StayFast

Organises the flow, protects the credential, validates image and prompt, manages job, webhook and recovery, saves the video, prevents automatic publication and links the result to the guest showcase and the Extras catalogue.

fal.ai

Receives the image and the prompt, runs the generation with the selected model, applies its own moderation rules, charges the consumption directly to the tenant and returns the technical result.

The guest

Sees the Extra, perceives the movement, opens the card, reads price and conditions and decides whether to request or buy. They don't need to know the architecture that produced the clip.

The value isn't only in the model

Pika and Seedance can generate a video. But StayFast's value is what happens before and after: before, the Extra already exists, the image is connected, the model is chosen in the right context, the prompt is prepared, duration and cost are visible. After, the job is recovered, the file is saved, the owner reviews it, the video is accepted and the showcase displays it.

A standalone generator produces a file. StayFast turns that file into part of the Extra's commercial path.

A simpler flow, without removing responsibility

The goal isn't to make every decision invisible. It's to make the right decisions simple. The owner doesn't need to know webhooks, queues, signed URLs or Storage. But they need to know which model they're using, which duration they've chosen, that consumption is external, that a new generation can cost again, that the result must be reviewed and that publishing requires a confirmation.

The owner knows the experience, StayFast organises the path, fal.ai generates the movement and the guest sees only the Extra ready to be discovered.

To understand why video can make Extras more visible, read also <a href="/blog/photo-to-video-property-extras">From photo to video: making your property's Extras feel more alive</a>.