> ## Documentation Index
> Fetch the complete documentation index at: https://docs.moxus.cloud/llms.txt
> Use this file to discover all available pages before exploring further.
# Conversation
> Test models, prompts, image generation, images, PDFs, Markdown files, parameters, and API request snippets from the web UI.
Conversation is the web testing surface for Moxus AI. After you sign in, it automatically uses an internal virtual key for test requests, so you do not need to paste your own API key into Conversation. Calls are billed like API calls and appear in Usage.
## What to test
* Model availability and response quality.
* Prompt wording and system instructions.
* Image, PDF, and Markdown inputs.
* Image generation from a prompt when you choose an image model.
* Video generation from a prompt or reference image when you choose a video model.
* Parallel answers from several models for the same question.
* Parameters that match the selected model type.
* Current model price information.
* API request snippets that show how parameters are structured.
## Attachments
Supported attachment types include:
| Type | Formats | Use cases |
| -------- | -------------------- | ---------------------------------------------------------------------------------- |
| Images | JPEG, PNG, GIF, WebP | Vision QA, screenshots, receipts, charts |
| PDF | `.pdf` | Summarization, QA, field extraction |
| Markdown | `.md`, `.markdown` | Appended as text in the conversation for README files, notes, changelogs, and docs |
The attachment size limit is about 20MB. Markdown files are read as text and sent with the message. Image and PDF understanding still depend on the selected model's input support.
## Image and video generation
When you choose an image generation model in the model selector, Conversation sends the request through the image generation endpoint. The text you enter becomes the generation prompt, and returned images appear directly in the conversation.
Image models only show **Image Count**. Enable it to set the number of images for the request, from `1-4`; when it is disabled, the request does not include an image count. Size, quality, and response format depend on the selected model. Conversation does not send text-chat parameters such as Temperature or Top P to the image generation endpoint.
With a video model, the text input becomes the video prompt. Any uploaded image is also sent as a reference image and can be used for image-to-video generation. Video generation is asynchronous: Conversation shows the task status first, then displays the video after the task completes.
| Video parameter | Values | Meaning |
| --------------- | ------------------------------------------------- | -------------------------------------------- |
| Resolution | `480p`, `720p`, `1080p`, `4k` | Output video resolution |
| Aspect ratio | Auto, `16:9`, `4:3`, `1:1`, `3:4`, `9:16`, `21:9` | Output frame ratio |
| Video duration | Auto, `4s-15s` | Requested video duration |
| Generate audio | On or off | Whether to generate audio in the video |
| Watermark | On or off | Whether to include a watermark in the output |
| Priority | `0-9` | Priority value sent with the video task |
Video models do not all support the same resolution, ratio, duration, or extra features. Start with the defaults or Auto. If a model rejects a combination, retry with values that model supports. Video models do not use text-chat parameters such as Temperature, Top P, or Max Tokens.
Use Conversation to test the prompt and model output first. When you are ready to integrate from code, use the `/v1/images/generations` examples in [Vision and image generation](/en/guide/vision-and-image).
## Workflow
1. Sign in and open Conversation.
2. Search for and select a model in the model list. For images or videos, choose the matching image or video generation model. To compare results, use **Add Model** to add more models.
3. Open the parameter panel beside each model and configure that model independently.
4. Enter text or upload an image, PDF, or Markdown file. A video model can use both a text prompt and reference images.
5. Send the request and review the response. With multiple models selected, the same question is sent to each model and Conversation shows each answer, image, or video-task status separately.
6. Open the API snippet panel when you want to understand the request shape. External code should still use an `sk-` key created in API Keys.
## Multi-model comparison
Use **Add Model** to put several models in the left-hand list. When you send one message, Conversation makes a separate request to every selected model, so you can compare their output in the same turn.
* Each response identifies the model that generated it.
* System prompts, parameter toggles, and parameter values are stored per model. Changing one model does not change the others.
* Remove is available only when two or more models are selected; at least one model must remain.
* Each selected model is an independent billed request, so adding models increases total usage and cost.
* Regenerating an answer requests only the model that produced that answer.
## Parameters
The parameter panel changes with the selected model type. The checkbox beside a parameter decides whether it is included in the request. A cleared checkbox means the parameter is not sent, which avoids errors from models that do not support it.
### Chat and reasoning models
| Parameter | Applies to | Values | What it does | Recommendation |
| ----------------- | --------------- | ----------------------------------------- | --------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------- |
| System Prompt | Chat, reasoning | Text | Sets a fixed role, rule set, or output format before the user message | Use it for a stable tone, response scope, or a required format such as JSON. You can set it separately for each model. |
| Temperature | Chat | `[0, 2]`, step `0.1` | Controls response randomness. Lower values usually make results more stable and conservative. | Use lower values for classification, extraction, or code edits. Raise it slightly for creative writing. |
| Max Tokens | Chat, reasoning | Chat `[1, 32000]`; reasoning `[1, 64000]` | Limits the maximum number of tokens the model can generate in this response | Raise it when responses are cut off. Lower it for brief replies to control usage. |
| Top P | Chat, reasoning | `[0, 1]`, step `0.01` | Limits the probability range of candidate tokens used during sampling | Usually keep the default. Avoid making large changes to both Temperature and Top P at the same time. |
| Frequency Penalty | Chat | `[-2, 2]`, step `0.01` | Reduces the tendency to repeat the same words or phrases | Raise it slightly when responses repeat themselves. Values that are too high can make phrasing unnatural. |
| Presence Penalty | Chat | `[-2, 2]`, step `0.01` | Encourages the model to introduce words and topics not already present in the conversation | Raise it slightly when you want broader coverage. Keep it low when the response must stay close to the source. |
### Image generation models
| Parameter | Values | What it does | Recommendation |
| ----------- | ----------------- | ------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------- |
| Image Count | `[1, 4]`, integer | Controls how many images the request generates from the same prompt | Start with `1` image to validate the prompt and model. Increase it after you are satisfied. When disabled, this parameter is not sent. |
Image models do not show System Prompt, Temperature, Top P, Max Tokens, Frequency Penalty, or Presence Penalty.
### Video generation models
| Parameter | Values | What it does | Recommendation |
| -------------- | ------------------------------------------------- | ------------------------------------------ | -------------------------------------------------------------------------------------------------------------- |
| Resolution | `480p`, `720p`, `1080p`, `4k` | Sets the output video's clarity | Start with `720p`. Higher resolutions can take longer or be unsupported by some models. |
| Aspect Ratio | Auto, `16:9`, `4:3`, `1:1`, `3:4`, `9:16`, `21:9` | Sets the output frame ratio | Choose Auto when unsure. `16:9` is common for landscape and `9:16` for portrait. |
| Video Duration | Auto, `4s-15s` | Sets the requested video length | Start with Auto or a short duration. The model determines the actual supported range. |
| Generate Audio | On, off | Requests audio in the generated video | Turn it off when you only need visuals. Success depends on the video model's capabilities. |
| Watermark | On, off | Requests a watermark in the output | Choose it according to your delivery needs and model requirements; some models may enforce their own behavior. |
| Priority | `[0, 9]`, integer | Sends a priority value with the video task | Keep the default `0` unless you know the selected model or service requires another value. |
Models of the same type can still accept different values. If a parameter causes a failure, clear text-parameter checkboxes, or reset image and video settings to their defaults or Auto before trying again.
## Troubleshooting
* If Conversation works but your client fails, compare Base URL, model name, request body, and key limits.
* If Conversation fails too, open Usage and read the failure reason, then check balance, model availability, and parameters.
* If an attachment is ignored, confirm that the model supports image or PDF input. For Markdown files, check that the file is readable text and not too large.
* If no image is generated, confirm that the selected model supports image generation, the prompt is not empty, and the requested image count is supported.
* If a video task fails, confirm that the selected model is a video model and the prompt is not empty. Start with the default resolution, Auto ratio, and Auto duration, then adjust one setting at a time.
* If multi-model answers differ, that is expected because they are independent requests. Check that their parameter panels use the same system prompt and enabled parameters before comparing output.
## Next steps
* [Quickstart](/en/overview/quickstart)
* [Vision and image generation](/en/guide/vision-and-image)
* [PDF files](/en/overview/modules/pdf-files)
* [Markdown files](/en/overview/modules/markdown-files)