Guides

    A prompt-built slideshow is planned slides, not your photo dump

    create_slideshow plans and fills slides from a prompt. Uploading your own photos is a different job, and mixing them wastes credits.

    Versely Team4 min read

    One prompt. A complete, curated slideshow. That is the job. Generate a slideshow from a prompt is one named agent capability. create_slideshow (and its baked-in-text twin create_slideshow_caption_in_image) plans per-slide queries, fills each slide, and applies overlay text if you asked for a style. It does not want your camera roll. Turn photos into a slideshow is that job. Feeding a prompt-slideshow your JPEGs — or generating a slideshow when you already had the photos — wastes credits on images you were going to throw away.

    The AI slideshow maker is the family door. This capability is only the prompt-planned row.

    What the prompt job actually builds

    Describe the theme, the format (reel, story, post, carousel), and how many slides. The agent plans a per-slide brief, curates or generates a matching image for each, and applies overlay text if you gave it a style. Required field on create_slideshow is prompt. Count, content type, and text style are optional.

    Two rendering styles, still one job:

    • Overlay (default): text stamped on top of the slide.
    • Caption in image: the image model renders the type into the picture. This path is async — status comes back as slideshow_generating, slides land over the next 30–90 seconds. Ask to see it on the next turn. Do not chain search_pinterest or extra image-gen tools on top; the tool's own instruction is end-to-end, and chaining double-renders.

    Cost is priced per image generated across the slideshow, shown before you confirm. That is why slide count is a width decision, not a vibe.

    What this job is not

    Your photos, arranged. create_slideshow_from_uploaded_images takes image_urls you already have. Order in the array is slide order. If you walk in with six product shots, use turn photos into a slideshow. Running the prompt job "inspired by" those shots still generates new stills. You paid for six cousins of files you already owned.

    A finished video. Turn a slideshow into a video is a later named job. This capability returns slides. Asking it to "just make the reel" in one breath is how people skip the review of slide three, which is the one that is wrong.

    Pinterest as the generator. Search Pinterest for inspiration returns URLs. create_slideshow is instructed not to be chained with that search. If you need a look, search first, then decide whether the prompt job or the upload job is the builder. Do not ask Pinterest to be a slideshow.

    A single hero generate. One image is generate an image from text. A slideshow is a planned set. Prompting six separate image jobs "for a carousel" is six uncoordinated stills and six charges without a planner.

    Let the planner pick the stills — or do not use this job

    The waste is half-using the planner. You specify the six captions you already wrote and ask it to generate cinematic stills and also drop in two of your own photos. The planner cannot see your photos the way the upload tool can. The generate path cannot protect your product label the way an attached JPEG can. Pick a contract.

    If the story is a theme you do not have pictures for — "five morning habits," "cozy fall aesthetic, eight slides" — this is the job. Brief the theme. Set a count you will actually review. Do not ask for twenty slides to "see options." That is a generate sampler billed as a slideshow.

    If the story is these pictures, stop. Upload job. Overlay captions there if you need type. Save the prompt-slideshow credits for a week you do not have the photos.

    FAQ

    Can I mix my own photos into a prompt-built slideshow?

    Not as this capability's contract. The prompt job plans and fills slides. The upload job takes your URLs. Mixing them in one ask is how you generate images you will replace, which is a wasted generate per replaced slide.

    Overlay text or caption baked into the image?

    Overlay if you may still change the words. Baked-in if the type has to live inside the scene. Baked-in is async; do not chain a second text-overlay tool on it or you double-render the line.

    Why not generate each slide as a separate image job?

    Because those jobs do not see each other. The slideshow planner is instructed to make a set. Six generate_images calls are six local optima. You pay six times for a carousel that does not read as one.

    Does this post the carousel for me?

    No. This job returns a slideshow. Posting is post or schedule to social media. A full campaign brief is set up a full content pipeline. Do not treat slide generation as publishing.