Generate images, video and audio
3 minute read · Chats & Local Agents
Connect OpenRouter, select a model for the kind of result you want and describe that result in the composer. Specialized models use Versions, keeping each generation with its prompt and inputs.
- Make your first generation
- Match the model to the job
- Add frames and references
- Inspect and download a result
- Continue without losing earlier results
Make your first generation
- Select OpenRouter and the intended account in the conversation.
- Open the model picker. Use a category and search to find a compatible model.
- Write a complete description of the result. Choose the settings offered for that model, such as aspect ratio, duration, resolution or voice.
- Add supported reference files if needed. Check their roles and remove files the selected model cannot use.
- Choose Generate version. Wait for the result or read the reported error before sending another request.
Switching models can change the available controls and accepted file types. Unsupported leftover files must be removed or replaced before sending. Provider charges and limits apply to each submitted request.
Match the model to the job
| Kind of model | Typical workflow |
|---|---|
| Images | Describe an image and add references when supported; inspect the result in the image gallery. |
| Video | Describe a clip, select available duration and size, and optionally add frames or references. Some models edit or upscale a source video. |
| Speech | Enter the words to speak and choose an available voice or supported reference audio. A voice reference can replace the voice setting. |
| Audio | Describe or supply the supported inputs for that model; listen to the returned compatible audio. |
| Transcription | Attach supported audio and read the returned transcription. This is separate from local microphone dictation. |
| Embeddings, rerank or decisions | Use the task-specific inputs shown by the model and inspect the structured result or downloadable file. |
The catalog describes available capabilities; it is not a guarantee that every provider accepts every request. Image, video, audio or PDF input is shown only where supported by the chosen model.
Add frames and references
A compatible video model can show independent First frame, Last frame, Images, Video and Audio boxes. Add, preview or remove a file in its own box. Replacing the first frame leaves the last frame and reference files in place. Adding video or audio does not silently move images to another role.
The general input limits are up to 10 files, 8 MiB per file and 24 MiB per message; lower model-specific limits also apply. Video editing and upscaling can require a single source clip. Some upscaling models do not require a prompt. Follow the actual controls and validation for the selected model.
When a provider requires different image dimensions or proportions, Vortex can prepare a resized or padded copy for the request. The original attachment and its frame/reference role remain intact. This preparation does not guarantee provider acceptance.
Inspect and download a result
Compatible video and audio results have playback controls. The file row shows its name, type, size and download action. Images open in the gallery. Other results, including structured files and unsupported media formats, remain available as downloads.
If a preview does not load, use its reload/retry action. Reloading an existing file is different from generating again. Results are stored as encrypted conversation attachments; supported cloud synchronization follows the attachment lifecycle.
Continue without losing earlier results
Select an earlier version to inspect it. Use inputs restores its prompt, original references, account, model and generation settings to the composer. Edit the draft and submit deliberately to create the next version. See versions, references and video extensions.
For microphone input rather than generated speech, see dictation and voice conversations.