Design tall pin images with a headline built into the picture, then turn the best ones into short vertical video pins. Test the same idea across several image models, keep the version that reads well at thumbnail size, and export it ready to upload to your boards.
Pinterest is a search engine where every result is a picture, and the pictures that win are tall, clear at thumbnail size and honest about what is behind the click. Making a few good ones used to mean a designer, a template and an afternoon. With AI you can describe the pin, get several versions in a minute, and animate the best one into a short video pin, then decide which one deserves a place on your board.
A pin is a small poster. It has one image, usually a line of text, and about a second to explain itself to someone scrolling fast. That is a very narrow brief, and narrow briefs are where image models do their best work: one subject, one mood, one headline, one tall frame. If you run a blog, a shop, a recipe site or a design portfolio, you probably need many pins for each page, each with a different angle, and that volume is the part that wears people out.
Picasso IA puts 212 text-to-image models and 121 text-to-video models behind one workspace, 502 models in total across the catalog, so the same pin idea can be tried on several engines instead of fighting the quirks of one. New accounts get free credits to test a direction before spending anything, and the full model catalog is open if you want to see what else is there. The point is not to make more pins than anyone can manage. It is to make a handful of good variations of each idea and see which ones people actually save.
Most of the work in a pin is composition. The image has to be tall, the subject has to survive being shrunk to a small card, and any words have to be big enough to read on a phone. Models handle this well when you ask for it directly: say that the picture is a vertical pin, name the subject, leave room for the headline, and describe the text exactly as you want it to appear, inside quotation marks.
For the text-heavy ones, Ideogram v4 Quality is a good first stop. Its listed output sizes include tall portrait options such as 1664x2496 and 1440x2560, so you can generate in a vertical frame from the start instead of cropping a square later. The gpt-image family from OpenAI, including gpt-image-2.0 and gpt-image-1.5, is the other obvious pair of candidates when words are part of the picture. Run the same prompt through both and keep whichever spells the headline correctly and leaves the cleanest layout.
Check the platform side separately. Pinterest changes its recommended pin sizes, ratios and video lengths from time to time, so read its current guidelines before you settle on a format, then pick the closest resolution the model offers.
Not every pin has the same job. A recipe pin, a quote pin and a product pin ask for different pictures, and a different model often fits each. This table is a starting point for what to try first, not a ranking.
| Pin type | What the picture needs | Models to try first |
|---|---|---|
| Headline pin | Big readable title, simple background, clear hierarchy | ideogram-v4-quality, gpt-image-2.0 |
| Photo-led pin | Strong single subject, natural light, room above for text | flux-2-pro, seedream-5-pro |
| Product pin | The real item on a clean or styled surface | nano-banana-pro (edits your own product photo) |
| Step or list pin | Several small scenes or numbered panels in one tall frame | gpt-image-2.0, ideogram-v4-balanced |
| Video pin | A still that has one thing worth moving | wan-2.6-i2v, kling-v2.6 |
This is the shortest route from an idea to a pin you can upload. It assumes you already know what the page behind the pin is about.
A video pin works best when the still already looks like a pin and exactly one thing in it should move: steam rising from a bowl, a curtain in a breeze, a hand pouring something, a slow push toward the subject. Wan 2.6 I2V takes your image plus a text prompt and produces a clip of 5, 10 or 15 seconds at 720p or 1080p, so you can describe the motion in a single plain sentence and let the model handle the rest. The wider workflow is covered in the page on turning photos into videos.
Keep the motion small. Big camera moves and crowds of moving objects are where image-to-video models drift, and a drifting clip looks worse than a still. Generate two or three takes of the same prompt and keep the calm one.
Tip: design the still as a pin first and as a video second. If the headline and subject are not readable on the first frame, the video will not rescue them, because many people see only that frame before deciding to scroll on.
Check the video result for shape before you upload it. Confirm that the clip kept the tall frame of your source image, and trim or re-export it if it did not, since a pin that arrives in the wrong proportions is cropped or letterboxed. If your pin needs sound, the page on background music for videos covers that side.
Pins are a distribution game as much as a design game, and the picture is only half of it. Here is where this workflow stops helping:
If you need an exact brand font or a locked template, finish the pin in your design tool and use the generated image as the background.
It can make pins that are clean, tall and legible, which covers most of what a pin needs. What it will not do on its own is match a strict brand system, so exact fonts, logo placement and fixed templates are better finished in a design tool. A sensible split is to let a model produce the picture and headline layout, then compare a few versions and keep the one that reads best on a small screen.
There is no single winner, which is why running the same prompt twice is worth the credits. Ideogram v4 Quality and the gpt-image models are the usual first choices when words are part of the picture, because they are the ones people reach for when the headline has to read correctly. Write the exact text in quotation marks, keep it short, and check each letter. Photo-led pins without much text are fine on FLUX 2 or Seedream.
Tall, and Pinterest decides the exact recommendation, which it updates from time to time, so check its current guidelines before you export. On the generation side, models such as Ideogram v4 Quality list tall portrait resolutions like 1664x2496 and 1440x2560, so you can pick a vertical size at the start rather than cropping a square afterwards. Cropping after the fact usually cuts into the headline.
Yes. An image-to-video model such as Wan 2.6 I2V takes a still and a short motion prompt and returns a clip of 5, 10 or 15 seconds. Pick a still where only one thing should move, write the motion in one sentence, and generate a couple of takes. Check that the output keeps a vertical frame, and check Pinterest for its current video length and file requirements.
Nobody can promise that, and this page will not. Saves depend on the topic, the title, the board, the audience and plenty of things outside the picture. What AI changes is how many versions you can test and how quickly you can discard weak ones. Treat each pin as a small experiment, keep the ones that people respond to, and stop making the ones that nobody opens.
Yes, and an editing model is the right tool for it. Upload your own product photo to a model like nano banana pro and ask for a cleaner background, a styled surface or a vertical layout with room for a headline. Keep the product itself unchanged. If the pin sells something, shoppers should get what the picture shows, so do not let the model change the item's colour, shape or features.
Use one reference image and reuse the same wording for palette, lighting and layout in every prompt, changing only the subject and the headline. Edit models that accept an input image help here, because the reference carries the look rather than the text alone. Expect to try several times, and review the set side by side so a stray colour or crowded layout is obvious before anything goes live.
Use faces that are yours or fictional characters, never a real identifiable person without their permission, and do not generate other brands' logos or trademarks. If a pin uses your own logo, add it afterward in a design tool so it stays exact, because models redraw logos with small errors. When in doubt, a pin that shows the product, the food or the place is safer than one built on a person.
Generation on Picasso IA gives you files you can use in your own content, but rules about commercial use, disclosure of AI imagery and what Pinterest accepts are set by the platform and by your country, and they change. Read Pinterest's current policies before you publish, and the terms linked from the model page you used. If the pin promotes a real product, make sure every claim in the picture is true.
Generation costs credits, and the amount depends on the model and the settings you pick, so no single number would be reliable. Images sit at the cheaper end of the catalog and video costs more. New accounts start with free credits, which is enough to test a pin or two, and the pricing page is the place to check current plans.
Pick one page you want people to find, write its pin in one sentence, and run it through the Picasso IA toolkit to see the first set of tall options.
Choose your language
