Provider and model matrix
Which providers you connect depends on what you want to generate. Not every provider does everything, some are chat-only, some are image or video specialists. This page maps each one to the modes it powers, so you know what a given key unlocks before you connect it. See Connecting a provider for how to add a key.
By provider
| Provider | Chat | Image | Video |
|---|---|---|---|
| Qwen (DashScope) | Yes | Yes, native | Yes |
| OpenAI | Yes | Yes, native | No |
| Google Gemini | Yes | Yes, native | No |
| Anthropic Claude | Yes | No | No |
| Kimi (Moonshot) | Yes | No | No |
| DeepSeek | Yes | No | No |
| GLM (Zhipu) | Yes | No | No |
| Custom endpoint | Yes | No | No |
| BytePlus | No | Yes | Yes |
| fal.ai | No | No | Yes |
Qwen, OpenAI, and Google are the only providers that generate images natively. BytePlus generates images too, but is image- and video-only, it never appears as a chat option. fal.ai is the reverse: video-only, hosting several third-party video models behind one key, and it also never appears for chat.
By mode
| Mode | Providers |
|---|---|
| Chat | Qwen, OpenAI, Google Gemini, Anthropic Claude, Kimi, DeepSeek, GLM, and a custom endpoint |
| Image | Qwen, OpenAI, Google Gemini, and BytePlus |
| Video | Qwen, fal.ai, and BytePlus |
Within video, fal.ai is a hub for several underlying models under one key, including options built on Veo, Seedance, Kling, and a lip-sync model. Qwen and BytePlus generate video directly through their own models. See Generating video for how these show up as choices in the composer.
Lip-sync is applied for you, not picked from the dropdown
fal.ai's roster does include a lip-sync model, but you do not select it yourself from the plain Video mode model dropdown. Marketing Studio applies it automatically for talking-avatar formats, chaining a talking-head clip into a lip-sync pass so the mouth matches the spoken audio. The Video mode dropdown itself lists the selectable clip models (Seedance 2.0, Seedance 2.0 Fast, Veo 3.1, Veo 3.1 Fast, Kling 2.6), not the lip-sync step.
Common tasks
Short recipes for picking the right key for a goal. This page is about which provider to choose; for the mechanics of entering, testing, and switching keys, see Connecting a provider.
I want to chat and also generate images with one key
Connect OpenAI, Google Gemini, or Qwen. These are the only three providers that do both chat and native image generation, so a single key covers the assistant and image mode. Claude, Kimi, DeepSeek, and GLM are chat-only and cannot make images, so pairing one of them with an image provider means connecting two keys.
I only want video
Connect fal.ai or BytePlus. Neither one appears as a chat option, and you do not need a chat provider to use them. You pick the actual model later in the composer's Video mode: fal.ai offers Seedance 2.0, Seedance 2.0 Fast, Veo 3.1, Veo 3.1 Fast, and Kling 2.6; BytePlus offers Seedance 2.0 and Seedance 2.0 Fast (and Seedream for images). Qwen also generates video, if you would rather keep chat, image, and video together on one key.
I want the widest coverage on the fewest keys
Qwen alone covers chat, native image, and video, so a single Qwen key is the broadest one-provider footprint. Add fal.ai only if you specifically want that hub's video models built on Veo, Kling, or the Seedance variants, or add OpenAI or Claude if you prefer their chat models for the assistant.
Where to get each key
Every provider's key comes from that provider's own developer console, not from the app. In Settings, AI Settings, Provider & Model tab, a small external link sits next to each provider's name and opens the right page. Key formats differ, which is a quick sanity check that you connected the correct provider.
| Provider | Where its key lives | Key looks like |
|---|---|---|
| OpenAI | The OpenAI developer platform, API keys page | sk-... |
| Anthropic Claude | The Claude platform console | sk-ant-... |
| Google Gemini | Google AI Studio, the Gemini API console | AIza... |
| Qwen (DashScope) | Alibaba Cloud Model Studio (DashScope) | sk-... |
| Kimi (Moonshot) | The Moonshot AI platform | sk-... |
| DeepSeek | The DeepSeek platform | sk-... |
| GLM (Zhipu) | Zhipu AI's BigModel console | xxxxxxxx.xxxxxxxx |
| fal.ai | The fal.ai dashboard, Keys page | key_id:key_secret |
| BytePlus | The BytePlus ModelArk console | a provider-issued token |
You do not need to memorize these pages, the external link in Settings takes you straight there. The format column is only a hint: if a key you pasted looks nothing like the shape above, you may have copied a token for a different product from the same company.
Example models
Each provider offers a small family of models to pick from once connected. Exact model names update as providers release new versions, but the families stay stable.
| Provider | Model family |
|---|---|
| Qwen (DashScope) | Qwen chat models in max, plus, and flash tiers, plus Qwen and Wan image and video models |
| OpenAI | GPT chat models in standard and mini tiers, plus a native image model |
| Google Gemini | Gemini chat models in pro and flash tiers, plus native image models |
| Anthropic Claude | Claude Opus, Sonnet, and Haiku |
| Kimi (Moonshot) | Kimi K2 chat models, including code-focused variants |
| DeepSeek | DeepSeek V-series chat models in pro and flash tiers |
| GLM (Zhipu) | GLM chat models, including a vision-capable variant |
| Custom endpoint | Whatever model name your endpoint exposes |
| BytePlus | Seedream for images, Seedance for video |
| fal.ai | A roster of hosted video models, including ones built on Veo, Seedance, Kling, and a lip-sync model |
Vision support
Whether a provider's chat models can look at an attached image varies too, which matters for chat and the in-editor assistant.
| Provider | Vision |
|---|---|
| OpenAI, Google Gemini, Anthropic Claude, Custom endpoint | Every model |
| Qwen, Kimi, GLM | Only specific models |
| DeepSeek | Not supported |
Attach an image while a text-only model is active on a provider that has a vision-capable option, and the app switches to that model for the message automatically.
Notes
- Chat providers power the assistant and skills.
- Image and video providers power image generation and video generation.
- A custom endpoint works for any provider that speaks the OpenAI-compatible chat completions format. It is chat-only. See Connecting a provider.
- Ask for an image while a chat-only provider is active, and the assistant falls back to a saved key for Qwen, OpenAI, or Google if you have one connected, since those are the only providers that generate images natively.
You can connect several providers at once and switch which one is active for chat from Settings, so you are never tied to a single model.
Costs are your provider's
You are billed by each provider for what you use, at that provider's own rates. Video in particular can be costly, so check the cost preview before you generate.
Troubleshooting
Choosing a provider here is separate from wiring it up. For anything about entering, testing, or fixing a key, see Connecting a provider, which covers what each Test connection result means (including why fal.ai and BytePlus only confirm the key saved rather than live-testing it), why reasoning effort resets on a new browser or device, and how to switch your active chat provider without losing your other keys.
Common picking mistakes:
- Expecting images from a chat-only provider. Claude, Kimi, DeepSeek, GLM, and a custom endpoint cannot generate images. If image mode has nothing to run on, connect OpenAI, Google, or Qwen.
- Looking for a chat toggle on fal.ai or BytePlus. There isn't one: they are generation engines that show a badge instead of a model dropdown, and the model is chosen at generation time in Video mode or Marketing Studio.
- Attaching an image to a text-only chat model. On a provider that has a vision model, the app switches to it automatically for that message. On DeepSeek, which has no vision model, it cannot, so use a vision-capable provider for image questions.