Using an image or video prompt from the catalog

An image or video prompt is text you paste into a generator. It is not installed in a coding agent: you copy it from its page, adapt a few passages, then start the generation.

Where to paste the prompt

Any generator that accepts a text description works: OpenAI's GPT Image, Google's Nano Banana (the name for Gemini's image capabilities), Sora, Veo, Kling, Seedance, and others. The prompt's page says which output it was written for (still image, clip, image to video). Models change often: the tool's official page lists the settings currently available.

Two starting points, depending on the output:

• Image: paste the prompt into the description field. To edit an existing image, most generators also accept a reference image alongside the text, as OpenAI's and Google's guides describe.
• Video: paste the prompt into the text-to-video mode, or into the image-to-video mode if the page says so. In that second case the text mostly describes motion: Kling recommends naming the subject, then its movement.

Replace the variables

Catalog prompts mark the passages to personalize: the subject, the brand, the colors, the text to display, the format. Replace them one by one before you run, then reread the whole prompt: a leftover instruction ends up in the image.

A few pointers from the official guides:

• Describe the scene in sentences rather than a list of keywords. Google suggests a structure of subject, action, setting, composition, style.
• For text inside the image, write it exactly, in quotation marks, and describe the font style.
• With a reference image, say what the generator must keep and what it may change. OpenAI's guide recommends this for edits.
• With Gemini, phrase it as a request for an image ("create an image of..."), otherwise the model may answer with text.

Set what is not in the prompt

Some parameters are set outside the text. For Sora through the API, resolution, duration and reference characters are set in the request parameters: writing "make it longer" in the prompt changes nothing, according to OpenAI's guide. In a web interface these settings are menus or buttons. Check them before you run, especially the aspect ratio (portrait for social networks, landscape for a standard video).

Iterate

A first generation is rarely the right one. Work in small corrections:

1. Change one thing at a time (lighting, framing, color).
2. In a chat interface, ask for the change as a follow-up: Google recommends short follow-ups such as "make the lighting warmer".
3. For video, add what the clip should show over time: action, camera movement, and sound if the model handles it. The Veo guides detail these elements.
4. Keep the version that works. A prompt you have tuned becomes yours: save it in your notes.

The cost of an iteration depends on your plan with the generator. Start with the lowest quality setting when one exists.

Text prompts for assistants

The catalog also contains text prompts meant for an assistant (Claude, ChatGPT, Gemini...). Paste them at the start of a conversation, or into a project's instructions if the tool allows it. Replace the variables, run, then fix the result with follow-ups. If you paste the same text every day, a skill will be more practical: see resource or plain prompt and the prompts catalog.

Rights on the images produced

PalateKit provides the prompt, not the images your generator makes from it. Your rights therefore depend on two things.

• The generator's terms. Each service sets what you may do with outputs. For example, OpenAI's Terms of Use state that, as between you and OpenAI and to the extent permitted by law, you own the output. For other tools, read their terms and your plan before commercial use.
• Copyright law in your country. In the United States, the Copyright Office considers that prompts alone are not enough to protect an output; creative modifications or arrangements by you can count. The rule differs from country to country.

Also worth knowing: Google states that videos generated with Veo carry a SynthID digital watermark that marks them as AI-generated. Do not ask for the imitation of an artist, a brand or a real person without the right to do so. For a client project, keep the generator's terms with the deliverable.

Going further

To see what a prompt offers compared with a skill, read resource or plain prompt. To install a first resource in your coding agent, see install your first resource.