From the API key to the finished video in your shop: set up providers, use templates and storyboards, keep costs under control and know what to do when a job gets stuck.
Shopware 6.6 or 6.7 with Storefront, PHP 8.2 or newer. For the videos you need your own API key with at least one video provider: Google (Veo via the Gemini API), Runway, fal.ai or Higgsfield. Billing runs directly with the provider.
Recommended but not required: ffmpeg on the server. It enables poster images from the video, smaller video files, the hover video for the listing, background music and the social version. Storyboards can be stitched via fal.ai without ffmpeg.
scheduled-task:run and messenger:consume should run as cron jobs or services. For long videos and storyboards, 512M memory and 300 seconds runtime are sensible.Upload the ZIP in the admin under Extensions, install, activate. Then enter the licence key in the module. You reach the AI Product Videos module through the Plugin Suite and the main menu. It has three tabs: Overview, Videos and Settings. On the product there is also an AI Video tab.
Under Settings, section AI video provider, choose the provider, enter the API key and pick the model. Each provider has direct links to the account, the key and the price list. Test connection checks the key without causing costs.
You can store keys for several providers. The generator then offers every model with a key, each with its price for the chosen length and resolution.
The video is created in the background, usually within one to five minutes. You can leave the page. Finished videos are stored in the media library in the folder KI Videos and wait for you under Awaiting approval. The generator saves unfinished settings automatically, they appear in the overview under In progress.
Six fixed templates define the motion:
Under Settings, Own templates, you save named prompts, for example for a fixed studio look. They appear in the generator and in bulk generation next to the fixed templates. The prompt of an own template fully replaces the motion of the fixed templates.
With an optional end image the video runs from the start image to the end image, ideal for a change from front to back. With Veo the length is set to 8 seconds. Create for all variants gives every variant its own video from its main image.
If a text AI is set up under Settings (OpenAI, Anthropic Claude, Google Gemini or Mistral), a click on Write prompt from product data creates a fitting instruction from name, description, manufacturer and properties. With Send product images it sees small previews and matches the prompt to the start image.
German texts are translated into English automatically before sending, which the video models understand best. With Google Gemini the text AI key field can stay empty, the Veo key is used then.
Cut-out product images often look sterile as a video. Under Scene image by image AI an image AI first places the product in a real setting, for example on a terrace. You describe the setting yourself or let the text AI suggest one.
If you like the result, take it over with Use as start image. With the setting Save as product image it is also added to the product. The image AI uses the Google Gemini, fal.ai or Runway key stored under AI video provider.
In Storyboard mode one video is made of several scenes. Optionally give an idea and let the AI write the storyboard, or create the scenes by hand. Each scene is generated as its own clip and then stitched into one video.
Create a video with an inexpensive model first. If you like it, choose Create final version on the finished video: same job, same images, same format and same music, just with a higher-quality model. That way you only pay for the expensive version when the video really works.
Background music comes from your media library. Upload royalty-free tracks as MP3, for example from Pixabay Music or the YouTube Audio Library, and check the licence. The music is laid under the video with a short fade in and out and replaces the original sound, the volume is adjustable.
The social version turns a finished video into a portrait version in 1080 x 1920 for Reels, TikTok and Shorts. The video sits in the middle on a blurred background made from itself, optionally with the product name on top. It is created with ffmpeg on your server, without AI costs.
Under Videos, Bulk generation, select products by category (including subcategories), a dynamic product group or your own selection. Decide whether main products, variants or both are included and whether only products without a video are considered.
Calculate preview shows how many videos will be created, what they will likely cost and which products are skipped, for example without an image. Monthly limit and budget cap the selection. With Write prompt per product with text AI every video gets its own prompt. The jobs run one after another in the background, even if you close the page.
With automation switched on, every newly created main product with an image automatically gets a video. Each product is handled once, for the existing catalogue you use bulk generation. Model, template and a daily cap are set separately. Ideally choose an inexpensive model and create good videos as a final version later. Monthly limit, budget and approval apply as configured.
With Approval before publishing switched on, every finished video waits for you. You can approve, reject, regenerate, download or delete it, optionally together with the file in the media library. With Show automatically after approval an approved video immediately replaces the product's previous video.
The list under Videos can be filtered by status and exported as CSV. Possible states: Queued, Generating, Awaiting approval, Approved, On the product page, Rejected and Failed. With Several videos per product more than one video can be active, and you change the order with Move forward and Move back.
.product-detail-buy, at the start or end inside the element, or as a replacementAutoplay, loop, controls and a maximum width are set per sales channel. Videos start muted in the shop. Variants without their own video can show the main product's video.
With Structured data the video appears as a VideoObject in the JSON-LD of the product page. The plugin also provides a video sitemap with all active videos. The addresses are ready to copy in the settings, submit them in Google Search Console under Sitemaps.
Poster images are created from the first video frame and delivered as WebP. Videos load lazily with a fixed size, so there are no layout shifts.
A subtle note in the top left of the video marks it as AI-generated. That creates transparency and meets labelling duties for AI content. You change the text in the settings or per language via the snippet stawKiVideo.aiNotice.
Costs depend on model, length and resolution and are charged per second. An example: according to its price list Veo 3.1 Lite costs 0.05 US dollars per second in 720p, so a 6 second video costs about 0.30 US dollars. Before every job the generator shows the estimate, the overview shows this month's and total costs.
The system check in the settings shows at a glance whether everything is ready: ffmpeg and its features (H.264, text overlay, WebP, font), the scheduled task, the background queue, memory, runtime, temporary folder, PHP extensions and all API keys. Test all keys checks every access at once.
If ffmpeg is not found, for example without root access, upload a ready-made ffmpeg binary to your web space, make it executable and enter the full path under Path to ffmpeg.
scheduled-task:run and messenger:consume.Questions and requests straight to us - support comes from the developer, not from a queue. You can also request a feature through the Plugin Suite in the admin.
Every plugin setting with its default and an explanation, in the same order as in the administration. Based on version 1.0.41.
Defines which sales channel the settings apply to. Without a selection they are the default for all, single fields can be overridden per channel.
Fields with the chain symbol inherit from the default. A click sets an own value for the channel.
Switches the display of videos in the shop on or off. Videos already created are kept.
Shows approved videos on the product page. When off, the generator in the admin stays usable.
Provider, API key and model for video generation.
Google Veo via the Gemini API, Runway, fal.ai or Higgsfield.
The key from your provider account. It is only used on the server.
The recommended model offers the best balance of quality and price. A custom model ID is possible.
Checks the key without causing costs and shows the credit with some providers.
Defaults for new videos, approval and limits for costs and parallel jobs.
This template is preselected in the generator.
Rounded to the lengths the model supports.
Landscape fits most tabs, portrait suits mobile and social networks.
Higher resolution costs more with some models.
Some models create matching sounds. Videos start muted in the shop, sound raises the price with some providers.
Appended to every instruction, such as light, background or mood. Most reliable in English.
What should not appear in the video. Not every model evaluates this.
Finished videos wait for your approval. Strongly recommended, as the AI can alter details.
An approved video immediately replaces the product's previous video.
Several videos can be active. All appear in gallery and tab, the listing shows the first.
Cap on new videos per calendar month. 0 means no limit.
New videos no longer start if the estimated cost would exceed the budget.
Failed jobs that had already started at the provider count with their estimate.
This many videos may be in progress at the provider at once.
If the provider delivers no video by then, the job counts as failed.
On overload or server errors at the provider the job restarts after three minutes, up to twice.
Converts images to JPEG, enlarges images that are too small and shrinks those that are too large.
Places the image centred on an area in the chosen format so nothing is cut off. The padding colour comes from the image corners.
Creates the poster from the first video frame. Needs ffmpeg, otherwise the product image is used.
Re-encodes the video with ffmpeg (H.264, fast start). Often saves more than half of the file size.
Only needed if ffmpeg is not found automatically.
Without ffmpeg the scenes are stitched via fal.ai. A fal.ai key is required for that.
Where and how the video appears on the product page and in the listing.
As its own tab, via CSS selector, both or neither (then gallery and listing only).
Right after the description or as the last tab.
Leave empty for the translatable default title.
Element on the product page where the video is inserted, for example .product-detail-buy.
Before, after, at the start or end inside the element, or as a replacement.
Plays muted as soon as the video is visible and pauses outside the visible area.
Restarts the video after it ends.
Shows play, sound and fullscreen.
Limits the width of the video.
Variants without their own video show the main product's video.
Also shows the video as a slide in the product image gallery.
Before or after the product images.
Plays the video in the product box while the mouse rests on it.
Counts plays and hovers without cookies and without personal data, just one counter per video and day.
Marks videos in the shop as AI-generated.
Shows a subtle note in the top left.
Leave empty for the translatable default text.
Makes videos discoverable for search engines.
Outputs the video as a VideoObject in the JSON-LD of the product page.
Provides a video sitemap with all active videos. The addresses to submit are shown below.
Sends a short email as soon as videos are finished or fail. Several videos are combined.
Recipient of the notification. Leave empty to send no emails.
Includes failed videos with their reason in the email.
Optional: a text AI writes prompts and storyboards from the product data.
OpenAI, Anthropic Claude, Google Gemini or Mistral.
With Google Gemini the field can stay empty, the Google Veo key is used then.
A fast, inexpensive model is enough for short prompts.
The text AI sees small previews. Prompts then match the start image, and in the storyboard it picks the image per scene.
Places a product image in a real setting, the video then starts from that photo.
Google Gemini, fal.ai or Runway. The key under AI video provider is used.
The recommended models keep the product most faithful.
Accepted scene images are also stored as an image on the product.
Named prompts that appear in the generator and in bulk generation.
Name and prompt per template. The prompt fully replaces the motion of the fixed templates.
Automatically creates a video for newly created main products with an image.
Applies to products created after switching it on, not to the existing catalogue.
Ideally an inexpensive model, then create good videos as a final version.
Motion type of the automatically created videos.
Limits the number of automatically created videos per day.
Checks ffmpeg, background processing, memory, runtime, temporary folder and API keys.
Runs the check again or tests all stored accesses at once.
All prices net, plus VAT.
You buy the licence directly from us and receive an invoice from us. AI Product Videos is in beta: fully usable, features and handling may still change. Installation by upload in the admin without SSH, updates through our licence platform, support straight from the developer.
The plugin will be listed in the official Shopware Community Store. Installation and automatic updates then run through the plugin manager in the backend, billing through your Shopware account. As soon as it is available, we will link it here.