Create one key in your account, name a model in the request and get the generated file back. Image, video, voice, music, 3D and chat models share the same door, so trying a different model is a change of name, not a new integration with a new vendor.
Most projects that touch generative AI end up with a pile of accounts: one for images, one for video, one for voice, one for a chat model, each with its own key, its own bill and its own habits. The Picasso IA developer API is built around the opposite idea. You create one key in your account, name a model in the request, and the generated file comes back, whether that model draws, animates, speaks, sings or answers.
Every extra vendor costs more than the signup. It is another key to store and rotate, another dashboard to check, another invoice to reconcile, another set of terms to read before you ship. For a small team or a solo builder, that overhead is usually bigger than the generation work itself, and it grows every time someone says the new video model looks better.
With one key, the integration is written once. The request says which model you want and what you are giving it, a prompt, an image, a script or a message, and the result comes back as a file you can download. When a better model appears, or the one you picked turns out to be wrong for the job, you change a name instead of starting a new project.
The same thing is true for credits. The API works with the same credit system as the toolkit, so one balance pays for an image, a video clip or a voiceover, and a generation costs credits that depend on the model and the settings. New accounts get free credits to start, and the pricing page is where the plans live. You will not find a price quoted here, because the right place for a number is the page that owns it.
The catalog behind the API is the same one the toolkit runs: 502 models today, and it is worth knowing how that breaks down, because the mix decides what you can build. Text-to-image is the biggest group at 212 models, text-to-video has 121, and large language models have 75. The rest is smaller but real: 28 video-editing models, 24 text-to-speech, 12 lipsync, 10 music, 9 super-resolution, 5 for 3D, 3 for speech-to-text, 2 video enhancers and 1 for background removal.
| Kind of model | A few examples from the catalog | What you give it | What comes back |
|---|---|---|---|
| Image | flux-2-pro, seedream-4.5, nano-banana | A prompt, sometimes reference images | An image file |
| Video | veo-3.1, kling-v2.5-turbo-pro, hailuo-2.3 | A prompt, sometimes a starting image | A video file |
| Voice | speech-02-hd | A script and a voice choice | An audio file |
| Music | music-2.6 | A style description, optionally lyrics | An audio file |
| 3D | hunyuan-3d-3.1 | A photo or a text description | A 3D model file |
| Chat | gpt-5, claude-4.5-sonnet | A message | Text |
You can browse every one of these on the catalog page, or by category in the collection, and each model page lists what that model accepts. That list is the thing to read before you build, because the models are not identical, and the next section explains why that matters.
The promise of one key is that the plumbing stays the same: the same account, the same credits, the same way of asking and the same way of receiving. A swap from one image model to another is a change of model name, and you then check that the settings you sent still make sense for the new one.
Treat the model name as a setting, not as a decision baked into your code. Keep it in one place, test a new model on a few real prompts first, and only then change it, because one API makes switching easy but it does not make two models behave the same.
That caution is the honest tradeoff. A voice model has a voice to choose and a speaking speed. A music model can take lyrics. A 3D model has a polygon count to decide. A video model may want a starting image, a length or an aspect ratio, and not every video model offers the same options. Between two models of the same kind the overlap is large, and between two kinds of model it is small, so a swap inside a family is usually a light edit while a swap across families is a different job that happens to use the same key.
The safest habit is to try first in the toolkit. Open the toolkit, pick the model, run your real prompt and see the result before you spend a minute wiring it up. If you are weighing an all-in-one catalog against a single chat assistant, the Picasso IA vs ChatGPT comparison lays out how they differ.
This is the journey in plain words, with no spec sheet required. It assumes an account and an idea of what you want to generate.
An honest page says where the tool stops, and a unified API has a few edges worth knowing before you promise anything to a client or a boss.
Faces and voices deserve one more line. Use your own likeness and voice, or fictional characters, and never a real identifiable person without their permission.
The pattern repeats across very different projects. A newsletter tool that makes a header image for each issue only needs one image model. A small studio that turns scripts into short videos wants a chat model to draft, a voice model to narrate and a video model for the picture, and the key is the same for all three. A shop that wants a 3D preview from a product photo needs a single 3D model and little else.
What these have in common is a catalog problem more than a coding problem: which model is right for this step? That is why the toolkit matters even for builders. It is the cheapest place to run the experiment, and the same models answer there and behind the key. If you already know the shape of the job, these pages show how people use the individual pieces: AI voiceover for videos, turning photos into videos and the photo upscaler.
It means one key from your Picasso IA account reaches the whole catalog, 502 models across image, video, voice, music, 3D and chat. You name the model in the request, give it the inputs it needs and get the generated file back. You are not opening a separate account with each vendor, and you do not maintain a separate integration for each one.
The intended coverage is the image, video, voice, music, upscale, background removal, 3D, lipsync and chat models of the catalog. By count that is 212 text-to-image models, 121 text-to-video, 75 language models, 24 text-to-speech, 12 lipsync, 10 music and so on. Browse the catalog page to see each one, and check a model page for what it accepts.
Not the integration, which stays the same, but you should still check the inputs. Models of the same kind overlap a lot, so swapping one image model for another is usually a small edit. Across kinds, say from an image model to a video model, the inputs differ and the job is different, so plan on testing the new model with your real prompts before you switch.
Yes, and it is the recommended first step. The toolkit runs the same catalog today, so you can pick any model, write a prompt or attach a file and see the result. If the output is not good enough there, wiring it into a project will not fix it, so decide on the model first and the integration second.
You create a key in your account and use it to identify your requests. Treat it like a password: do not paste it into a public repository, a shared chat or a screenshot, and keep it out of any code that runs in a visitor browser. If you think it has leaked, replace it from your account and ask support if you are unsure what to do next.
With credits, the same ones the toolkit uses. The cost of a generation depends on the model and the settings you choose, so a short image, a long video and a chat reply do not cost the same. New accounts get free credits to start, and the pricing page lists the plans. This page quotes no figures, because they change and the pricing page is the one to trust.
No, and it would be a mistake to expect it. Each model has its own parameters, limits and strengths, so the same prompt can give a clean result on one model and a poor one on another. That is exactly why a catalog is useful: you run the input through two or three candidates and keep the model that understood the brief.
It depends on the model and on what you ask for, and this page will not put a number on it. As a rule, a short image or a chat reply comes back sooner than a video, a song or a 3D model. Build your project so that it can wait for the file instead of assuming an immediate answer, and test with realistic inputs.
Only with your own likeness and voice, or with fictional characters. Do not generate content that imitates a real identifiable person without their permission, and do not present generated media as if it were real footage or a real recording. Check the rules of any platform you publish on, since each has its own requirements for AI content.
Ask before you build. The page describes a key from your account, a model name with its inputs and a file that comes back, and it makes no promise beyond that. For a specific need, such as a workflow that chains several models, write to support and explain the job, and meanwhile test the idea with the same models in the toolkit.
Ready to find out which model fits your idea? Open the toolkit and run your first prompt, or ask support if you have questions about building on it.
Choose your language
