Take the cover you already have and the blurb you already wrote, and turn them into a short video teaser: an image-to-video clip for the opening, a voiceover read from your own copy, and a music bed to hold it together. Everything runs in one workspace, so you can test a few moods before you pick one.
A book trailer used to mean a budget, a camera crew and a month of back and forth with an editor, so most authors skipped it and posted the cover again. You can now build a decent teaser from two things you already own: the cover and the blurb. The cover becomes an animated opening shot, the blurb becomes a voiceover, and a short music track holds the mood together.
A cover is a still image and a blurb is a block of text. Neither moves in a feed that mostly shows moving things. A fifteen to thirty second teaser gives a reader a reason to stop scrolling and a feel for the book before they read a single page, which is the job a trailer has always had.
The honest advantage of AI is not polish, it is iteration. Picasso IA runs 502 models behind one workspace, so you can animate the same cover with two different video models, narrate the same blurb in three voices, and compare the results the same afternoon. New accounts get free credits to try this, and a draft you dislike costs you a few minutes rather than an editor's invoice. If you still need the cover itself, the book cover design page covers that part.
Not every book wants the same trailer. A thriller needs restraint and a ticking tempo, a cozy mystery wants warmth, and a children's picture book wants bright motion. Choosing the style before you generate anything saves most of the wasted attempts, because the style decides what you feed the models.
| Trailer style | Built from | Voice and music | Watch out for |
|---|---|---|---|
| Animated cover | Your cover as the first frame of an image-to-video clip | Short tagline, one music bed | Title lettering can warp as it moves |
| Mood montage | Four or five generated scenes about setting and atmosphere | Low narration, rising music | Scenes drifting into a different book |
| Blurb read aloud | Slow visuals under a narrated blurb | Calm voice, sparse music | Giving away the twist in the last line |
| Character teaser | A fictional character image animated in a few short shots | Voice of the narrator, tense music | Faces changing between clips |
| Quote card video | A pull quote over gentle motion | No narration, music only | Text rendered inside video, so add it in an editor |
If you are unsure, start with the animated cover. It needs one image, one clip and one voice line, and it already tells you whether the idea works. The short films and trailers page goes deeper on longer cuts, and photos into videos covers the animation step on its own.
This is the order that avoids rework. It assumes a finished cover file, a blurb and a free afternoon.
The prompt is where a trailer goes wrong quietly. If you paste the blurb straight into a video model, it will try to illustrate every event, and the result is a muddle of half scenes that spoils things you wanted to hold back. Write a separate prompt for each shot instead, and build each one from a place, a light and a single movement: a lighthouse at dusk with the beam turning, a kitchen table with a letter sliding into frame.
Characters need more care. Describe a fictional figure by clothing, silhouette and mood rather than by a detailed face, and show them from behind or in shadow when the shot allows. This keeps the character consistent across clips more reliably than a close-up, and it keeps your trailer from resembling any real person by accident. The consistent AI characters page explains how to hold one look across several images.
Pro tip: read your finished script and ask whether someone who has never opened the book could repeat the ending after watching. If yes, cut the last line. A teaser should leave the question open, not answer it.
The voice does more of the selling than the visuals do. gemini-3.1-flash-tts has 30 voices and a style prompt, so you can ask for a slower, quieter delivery for a literary novel or a brisker one for a thriller, and markup tags like whispering let you shape a single phrase. It works across many languages, which helps if your book is also published in a translation. The voiceover page covers how to direct a synthetic voice in more detail.
For music, lyria-3 turns a text description into a short instrumental clip, and you can attach reference images, for example the cover, to steer the mood. Ask for instrumental only, name the instruments and the tempo, and avoid naming a real artist or a famous film score. Because a clip is short, you may need to generate two and let a video editor loop or crossfade them. The background music page has more on that.
A teaser built this way has real limits, and knowing them early saves a lot of retries:
None of this stops a good teaser from working. It means the human decisions, what to reveal and where to cut, still decide whether it is any good. Check each platform's current requirements for video length and format before you upload.
Yes, that is the simplest version. The cover becomes the first frame of an image-to-video clip, the blurb, trimmed to a few lines, becomes a script for a text-to-speech voice, and a generated music track sits underneath. You will probably want two or three extra scenes to avoid showing one moving cover for the entire length, but the cover and the blurb carry the core of the idea.
Image-to-video models are the right family. Wan 2.6 I2V takes a still photo and a prompt and produces a clip of 5, 10 or 15 seconds, and Seedance 2.0 can use an image as its first frame and generate audio in the same pass. Try the same cover in two models, because they interpret motion prompts differently, and keep whichever one leaves the artwork looking most like your cover.
Write the script and the shot prompts yourself instead of pasting the whole blurb into a model. Stick to the premise, the setting and the central question, and stop before the turn. After you have a draft, ask a friend who has not read the book what they think happens next. If they guess the ending, the teaser says too much.
Only your own face, or a person who has given clear permission. Never a real identifiable person without it, and never a celebrity. For most fiction the safer path is a fictional character shown in silhouette or from behind, which also avoids the continuity problem of faces changing between clips. If you do use your own likeness, treat it as your own choice and check the platform rules where you post.
Yes. Text-to-speech models turn your script into a voice file. Gemini 3.1 Flash TTS offers 30 voices and lets you describe the delivery in plain language, such as slow and quiet. Pick a voice that suits the genre, listen to the whole thing once, and rewrite any sentence that sounds awkward, since spoken text needs shorter sentences than the page does.
It can make an original instrumental from a description. Lyria 3 produces a short clip from a prompt that names genre, instruments, tempo and mood, and it accepts reference images, so the cover can steer the tone. Ask for instrumental only. Avoid naming real artists or famous scores, and expect to generate a couple of options before one fits the pace of your cut.
There is no rule worth quoting, and platforms change their limits and formats, so check the current requirements of wherever you plan to post. As a practical guide, shorter is easier to get right: a hook, a few images of the world, one line of tension and a title card. Individual clips from these models run a few seconds each, which pushes you toward a tight edit anyway.
For a polished result, yes, though any basic editor will do. The models generate the clips, the voice and the music, but joining them, trimming the timing and adding the real title text happen outside the generation step. That last part matters, because lettering added in an editor stays sharp, while lettering generated inside a video can warp.
Generally that is what these videos are made for, but the rules depend on the model and on where you publish. Read the terms that apply to the models you used, and keep to the platform's current advertising and content rules. If your book features a real brand, a real person or someone else's artwork, remove them from the trailer before you post it.
Generations are paid in credits, and the cost depends on the model and the settings. Video usually costs more than images or audio, so test ideas at the shortest length and lowest resolution first, then re-run the winner. New accounts start with free credits, and current plans are on the pricing page, which is the only place to trust for numbers.
Your book already has a cover and a blurb. Open the Picasso IA toolkit, animate the cover and hear how the first line sounds out loud.
Choose your language
