Image, video, and audio generation

Image generation, video generation, and audio each have one global selection, found under Settings → Models → Default models as "Image model", "Video model", and "Audio model". A model source's login status and interface configuration decide which options are available; when a request actually runs, only the source selected here is used. All chats share these three selections — a chat using Claude, Codex, or another language model does not change the image, video, or audio source.

Custom interfaces can keep a Base URL, API Key, and model configuration, and a custom audio interface additionally has a voice. Its endpoint and key are used only when "Custom model" is selected; saving or editing a custom interface does not override an already selected subscription source, nor does it automatically change the default selection.

The image, video, and audio tools are always callable, and every request reads the current default selection. When the selected source is not logged in, its configuration is missing, or the request fails, the tool returns an explicit error; it never quietly switches to another configured source. After you change the default selection, subsequent requests use the new selection — there is no need to switch the chat's language model.

Generation has a dedicated display in chat: image, video, audio and music each carry their own icon, and the status row follows the progress — from awaiting approval to generating, then generated, failed or stopped. Several generations in one message are grouped together, and any failure within the group is shown as it is.

• Image generation can use the subscription interfaces of Codex, Qoder, QwenWork, or QoderWork, or a configured custom interface. Subscription interfaces use the corresponding source's login credentials, with no separate custom backend to fill in.
• The Qoder, QwenWork, and QoderWork subscription interfaces support text-to-image only and do not accept reference images. Codex and custom interfaces that support image editing can take reference images; the exact support depends on the interface.
• Video generation can use MiniMax M Plan or a custom interface. MiniMax uses subscription credentials, and video requires the Explore or Build effort level; it keeps the video model name you set, but the endpoint and credentials always come from the MiniMax subscription. A custom interface uses its own saved configuration.
• Audio can use MiniMax M Plan, the Alibaba Cloud Bailian Token Plan, or a custom interface. MiniMax's audio models spend from the same subscription allowance as text turns, need no separate interface, and likewise only keep the model name and voice. Bailian audio goes through the plan gateway's own speech synthesis route and also spends plan allowance; it accepts only qwen-audio-3.0 series models and returns a download link valid for 24 hours. For a custom interface you must first pick a provider, since the four protocols' request shapes are not compatible: "OpenAI compatible" /audio/speech returns the audio file directly; "Zhipu BigModel" GLM-TTS has the same shape but returns wav only, with a limit of 1024 characters per request; the "Alibaba Cloud Bailian" pay-as-you-go interface uses the platform's own protocol, returns a download link valid for 24 hours, and has a limit of 600 characters per request; "MiniMax open platform" uses a pay-as-you-go open-platform API key and its own protocol, with no endpoint to fill in (the M Plan subscription is a separate source and is not configured here).
• Zhipu and Volcano Ark are not offered as audio subscription sources: a Zhipu coding-plan key can pass audio interface authentication, but the plan itself includes no audio allowance (in practice it returns an insufficient-balance error); the Volcano Ark domain has no speech synthesis interface. A Zhipu open-platform key can still be used in a custom interface by selecting "Zhipu BigModel".
• Each speech request can also set a voice, emotion, and speed for that clip without changing the global settings: voice applies to every source; emotion (calm, happy, and so on) is honored only by MiniMax; speed (0.5 to 2 times) is honored by MiniMax and OpenAI-compatible interfaces. Unsupported parameters are ignored.
• Generated audio, like video, is returned as a saved path inside the session directory; generated audio and video play inline in the conversation and can also be opened from the Files panel.
• Local image, video, and audio generation is not available in the current version. A retained legacy local configuration does not block another explicitly selected subscription source; selecting that legacy configuration shows an unavailability notice.
• Generated files count as artifacts of the turn and appear in session artifacts — see Session artifacts.