Write the launch brief once, attach your reference images, and let Supercomputer plan the kit in a chat: a hero image, ad variants, a short clip, a voiceover and captions. Each piece is rendered with models from the Picasso IA catalog and checked by you before anything goes live.
A launch rarely needs one asset. It needs a hero image, a few ad variants in different shapes, a short clip, a voiceover and captions, and all of it has to read as the same campaign. Supercomputer, the agent in Picasso IA, takes the brief you would hand a designer and works through that list in one chat, keeping every piece tied to your reference images.
The slow part of a campaign is rarely any single image. It is the handoffs. The hero image is approved, then someone has to remember its palette for the square ad, the vertical ad, the clip, the voiceover script and the captions, and each handoff is a chance for the look to drift. A kit built from one brief removes most of those handoffs, because the brief and the reference images sit in the same conversation as every asset that comes out of them.
Supercomputer is an agent in a chat. You describe the result in plain words, it plans the job, picks models from the catalog, renders the steps and shows you what came back. Then you steer in the same conversation: make the clip calmer, try the ad on a darker background, change the voice. It is a separate product from the toolkit, where you pick the model yourself, but it runs on the same catalog of 502 models and the same credit system, so a generation costs credits that depend on the model and the settings.
A kit is five kinds of asset, and each one needs something different from the brief. The table lists what each piece asks of you and which catalog models fit it. The models named are examples from the catalog; the agent chooses for the job, and you can ask for a different one in the chat.
| Asset | What the brief has to say | Models that fit |
|---|---|---|
| Hero image | Subject, mood, palette, where it will be seen | Nano Banana Pro, Seedream 5 Pro |
| Ad variants | The same subject in new shapes: square, vertical, wide | Nano Banana Pro with your reference images |
| Short clip | One action, one camera move, a start image | Veo 3.1 starting from the hero image |
| Voiceover | The script, the tone, the language | Speech 2.8 HD |
| Captions | Font, position, language | Autocaption on the finished clip |
Read the table as a dependency chain. The hero image sets the look. The variants and the clip inherit it. The voiceover is written for the clip, and the captions are generated from the clip's audio. Change the hero image late and everything downstream has to be redone, which is why the first step is a brief that is settled before any rendering starts.
This is the journey from a document to a folder of files. It works for a product launch, an event, a seasonal sale or a new feature, and it assumes you have a brief and at least one reference image.
The toolkit runs the same catalog, so you can test every step above by hand there today: generate the hero image with Nano Banana Pro or Seedream 5 Pro, animate it, record the voiceover and add captions, one model at a time. It is slower, and it shows you which prompts and models suit your brand.
Consistency is the whole point of a kit, and it is the part most likely to go wrong. Three habits carry most of the weight.
First, use reference images instead of adjectives. A phrase like warm, playful, premium means something different to every model. A photo of your product and a mood image mean the same thing every time. Nano Banana Pro accepts several reference images alongside the prompt, and Veo 3.1 can start a clip from an image or hold a subject steady from reference images, so the look can travel from still to motion.
Second, make the hero image the source of truth. Approve it before the rest of the kit is built, and tell the agent to treat it as the reference for the variants and the clip. If a character recurs across assets, the same discipline applies as in consistent AI characters: lock the reference, vary the pose and the scene.
Settle the hero image before anything else is rendered. Every variant, the clip and the captions inherit its palette and subject, so a change made late costs you the whole kit, while a change made at the plan stage costs you a sentence.
Third, keep the words consistent too. The line that appears in the voiceover should match the line in the captions and the line in the ad. Put that line in the brief once, in quotation marks, and say it must be used exactly. Models paraphrase unless told not to.
Faces, voices and likenesses need a rule of their own. Use only people you have the right to use: yourself, a team member who agreed, or a fictional character. Never build a campaign around a real, identifiable person's face or voice without their permission, and do not imply an endorsement that never happened. For AI voice work, pick a stock voice or one you are entitled to.
Most disappointing kits trace back to the brief, not the models. The common mistakes are predictable.
A brief that describes a feeling and not a thing produces assets with different subjects. Name the subject, name the setting, and name what must not appear. A brief that skips the format gets assets that fit nowhere: say which asset is square, which is vertical and which is wide, because the same composition rarely survives a change of shape. A brief with no tone for the voice gets a default delivery that may not match the picture; say calm, brisk or warm, and say the language.
Another frequent mistake is asking for too much in one pass. A kit of five asset types with three variants each is a lot of credits and a lot to review. Start with one hero image and one variant, confirm the look holds, and then ask for the rest. The same goes for the clip: one action and one camera move is the safest request.
The last mistake is skipping the text check. Models render short words well and long lines badly, and a headline baked into an image can come out with a wrong letter. Proofread any text inside a generated image. The static ad creatives page goes deeper on layouts, and AI video ads for social covers the clip.
A kit from one brief is a strong first draft of a campaign, not a finished one. These are the places it stops helping.
Models also differ, and no single one is best at all five asset types. That is why the agent picks per asset, and why you review each result.
Supercomputer is an agent in a chat. You describe the result you want in plain words, it plans the job, picks models from the catalog, renders the steps and shows you the result, and you steer it in the same conversation. The toolkit is the manual workspace where you choose the model yourself, set the options and generate one piece at a time. They share the same catalog and the same credits.
A hero image, ad variants in the shapes you ask for, a short clip, a voiceover and captions. The images come from text-to-image models, the clip from a video model that can start from your hero image, the voice from a text-to-speech model, and the captions from a captioning model that reads the clip's audio. You can ask for fewer pieces.
No. Supercomputer is a chat in the web app, so there is no key to create or protect. Keys belong to the API, the MCP server and the connectors, where access works with a key you create in your account and should treat like a password. You write the brief and talk to the agent.
Through the brief and the references. Your reference images travel with the conversation, the hero image acts as the source of truth for the variants and the clip, and the exact line of copy is written once and reused. Models interpret images loosely, so review every asset against the hero image and ask for a redo of anything that drifted.
It draws on the same catalog as the toolkit, which holds 502 models. Fitting examples are Nano Banana Pro or Seedream 5 Pro for images, Veo 3.1 for a clip, Speech 2.8 HD for the voice and Autocaption for burned-in captions. The agent chooses for the job. If you prefer a particular model for a particular asset, say so in the chat and see whether the result suits you.
It depends on the models, the settings and how many revisions you ask for, so there is no single number to quote here. Video usually costs more than images, and every redo spends credits again. New accounts get free credits, and plan details live on the pricing page. The plan step helps, because you see what is about to be generated before it is rendered.
Only your own, or someone who has clearly agreed, or a fictional character. A campaign built on a real, identifiable person without their permission is a legal and ethical problem. If you need a spokesperson, a fictional character kept consistent with a reference sheet is the safer route.
You can try each step today in the toolkit, which runs the same catalog. Generate the hero image, make the variants from it with reference images, animate the best one, write the voiceover and add captions to the finished clip. Supercomputer is built to run those steps for you from one brief, and the toolkit shows you which models and prompts work for your brand beforehand.
Not automatically. Sizes, lengths, file types, text limits and content rules differ by platform and change over time, and nothing on this page quotes them. Tell the agent the shapes you need, then check each platform's current requirements before you export and upload.
Yes, and that is the point of working in a chat. Tell the agent which asset to change and what to change, and the other pieces stay as they are. The exception is the hero image: since the variants and the clip inherit its look, changing the hero image means regenerating what depends on it, which is why it is worth approving first.
Your next campaign does not need six separate tools and six handoffs. Open the toolkit to test the hero image and the clip on your own brief today, and ask support if you want a hand choosing models.
Choose your language
