> ## Documentation Index > Fetch the complete documentation index at: https://magic-post.mintlify.site/llms.txt > Use this file to discover all available pages before exploring further. # AI models > Store provider API keys, switch writing and media models on or off, test them, set credit prices and add fal.ai endpoints. MagicPost's AI caption writer and its Image and Video Studio run on models from outside providers. You connect the providers and decide which models customers can use on two admin pages under **Publishing**: * **Writing Models** for captions and text. * **Media Models** for pictures and video. > [!NOTE] > > Both pages share one set of provider keys. A key you save on **Media Models** for OpenAI, Google or xAI is the same key **Writing Models** uses. How customers use the models is covered in [AI writing and media](https://magic-post.mintlify.site/user-guide/ai-tools.md). ## When a model is usable A model appears to customers only when both of these are true: 1. The provider's API key is stored. 2. The model's own switch is on. Each page opens with a line that counts how many models are ready, or tells you that none is usable yet. ## Writing models **1. Open the page** Go to **Publishing**, **Writing Models**. Under **Provider keys** there is one field for each provider on the list: **OpenAI GPT**, **Google Gemini**, **Anthropic Claude** and **xAI Grok**. **2. Paste the keys** Paste each key you have into its field. The eye icon shows the value. Click **Save keys**. Leave a field empty for a provider you do not use. Clearing a field removes that key. **3. Check the provider card** Each provider has a card with a **Key stored** or **No key** badge, and a row for each of its models. **4. Test a model** Click **Test** on a model row. MagicPost sends the smallest possible real request. A success message names the model and says it answered. A failure shows the provider's reason. The request is tiny, but it is made with your provider account and counts as usage there. **5. Switch it on** Turn on the switch at the end of the row. Models from the starting list come switched on, so a missing key is usually the only thing keeping them unusable. Getting keys: [OpenAI](https://magic-post.mintlify.site/integrations/ai/openai.md), [Anthropic](https://magic-post.mintlify.site/integrations/ai/anthropic.md), [Google Gemini](https://magic-post.mintlify.site/integrations/ai/gemini.md) and [xAI Grok](https://magic-post.mintlify.site/integrations/ai/xai.md). ### Edit a model Click **Edit** on a row to change it. | Field | Notes | | - | - | | **Model identifier** | The model name exactly as the provider names it in their API. Up to 191 characters. | | **Display name** | What customers see. Up to 191 characters. | | **Description** | Optional, up to 500 characters. | | **Tier** | **Premium**, **Standard** or **Fast**. Shown as a badge. | | **Credits per run** | The price of one generation, a whole number from 0 to 1000. | Provider model names change over time. If a test reports an unknown model, correct the identifier here. When a model turns down one of the settings MagicPost sends, the test says which ones, and MagicPost stops sending them to that model. ## Media models **1. Open the page** Go to **Publishing**, **Media Models**. If the list is empty, click **Load the starting list**. A fresh install already has it. **2. Paste the keys** The **Provider keys** form lists **OpenAI**, **Google**, **xAI** and **fal.ai**. Paste the keys you have and click **Save keys**. **3. Test a picture model** Click **Test** on a picture model. MagicPost makes one small picture to prove the key and the model identifier work together. Video models and models that need a source picture have no Test button. **4. Switch models on** Use the switch on each row. Pictures are priced per picture and video clips per second. Credits are held while the work runs and charged only for what arrives. See [Credits and allowances](https://magic-post.mintlify.site/saas/credits.md). Each row shows the model name, a tag with its price (for example "credits a picture" or "credits a second"), the shapes it makes and, for video, the clip lengths. Further tags say **Edits**, **From a picture** or **Needs a picture**. **Refresh the starting list** reloads the built-in models. It adds models that are missing and resets display names, descriptions and capabilities to the starting values. It keeps the credit cost and the on/off switch of models already on the list. ### Edit a media model | Field | Notes | | - | - | | **Model identifier** | Exactly as the provider names it. Unique within the provider. | | **Display name**, **Description** | As for writing models. | | **Credits per picture** or **Credits per second** | Whole number from 0 to 10000. | | **Pictures per request** | Picture models only. **Up to** 1, 2, 3 or 4. | | **Can edit a picture** | Picture models only. | | **Clip lengths, in seconds** | Video models only. Comma-separated whole numbers of up to two digits, for example `4, 6, 8`. | | **Can start from a picture** | Video models only. | | **Shapes it makes** | Tick at least one of 1:1, 4:5, 3:4, 4:3, 3:2, 2:3, 16:9, 9:16. | ### Add a fal.ai endpoint fal.ai serves many picture and video models behind one key, so you add the ones you want by endpoint id. **1. Fill in the form** At the bottom of **Media Models**, under **Add a fal.ai endpoint**, enter the **Endpoint id** exactly as fal.ai shows it, for example `fal-ai/flux-2` (at least two parts separated by a slash, made of letters, digits, dots, hyphens and underscores), a **Display name**, what it **Makes** (**Pictures** or **Video**), and the credit price. **2. Add it** Click **Add endpoint**. The model is added switched off and its editor opens. **3. Check the request mapping** The **Request mapping** box holds JSON that tells MagicPost which fields the endpoint expects. It must be a JSON object that contains `inputs.prompt`, the field that takes the description. Compare the field names with the endpoint's API page on fal.ai. **4. Save and switch on** Tick **Only works from a picture** if the endpoint cannot run without a source image. Click **Save**, then turn the model's switch on. The mapping uses these keys: `inputs` names the endpoint's fields for prompt, shape, count, seconds and image (end a field name with `[]` to send a list); `shape_values` renames shapes; `seconds_as` sends the length as a `string` or a `number`; `preset` goes with every request; and `output` says where the files are: `images`, `image` or `video`. Use the trash icon on a fal.ai row to remove it. You cannot remove a model while something is still being made with it. Pictures and clips already made stay in the libraries. ## Provider keys are secrets Provider keys are stored encrypted in the database. Never paste them into support tickets or screenshots. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.