> Creatomate AI reference — topic page "AI generation providers"; the index of all pages is https://creatomate.com/llms.txt ## AI generation providers The user's own provider keys power AI generation, configured at **Project Settings → Integration**. Defer provider setup until after the first stock-asset render — a missing key is one of the most common render failures, and it interrupts the first win. - **ElevenLabs speech** (recommended default): on an audio element, `provider: "elevenlabs model_id=eleven_multilingual_v2 voice_id=XrExE9yKIg1WjnnlVkGX"` with `source` set to the words to speak (that `voice_id` is the stock default voice; swap it for the user's chosen voice). Includes word-level timestamps, so subtitles generate without a transcription step. - **OpenAI speech**: `provider: "openai model=tts-1-hd voice='alloy'"` with `source` set to the words to speak. - **Stability AI images**: on an image element, `provider: "stabilityai model=sic"` with `source` set to the text prompt. `model` is required: `sic` (Stable Image Core, the editor's default), `sd3` or `sd3-turbo` (Stable Diffusion 3). Options: `aspect_ratio=` one of `16:9`, `1:1` (default), `21:9`, `2:3`, `3:2`, `4:5`, `5:4`, `9:16`, `9:21`; with `sic` also `style_preset=` one of `3d-model`, `analog-film`, `anime`, `cinematic`, `comic-book`, `digital-art`, `enhance`, `fantasy-art`, `isometric`, `line-art`, `low-poly`, `modeling-compound`, `neon-punk`, `origami`, `photographic`, `pixel-art`, `tile-texture`. - **OpenAI images**: on an image element, `provider: "openai model=gpt-image-2"` with `source` set to the text prompt. `model` is required: `gpt-image-2` (the editor's default), `gpt-image-1.5`, `gpt-image-1` or `gpt-image-1-mini`; `dall-e-2` and `dall-e-3` are discontinued and fail the render. Options: `size=WIDTHxHEIGHT` (default `1024x1024`; `gpt-image-2` also takes `auto`, `1536x1024`, `1024x1536`, `2048x2048`, `2048x1152`, `1152x2048`, `3840x2160`, `2160x3840`, or any size whose edges are multiples of 16, at most 3840 px, within 3:1 of each other and 655,360 to 8,294,400 pixels in total; the other models take `1024x1024`, `1536x1024` or `1024x1536`) and `quality=low|medium|high` (default `medium`; `auto` on `gpt-image-2` only). Example: `"provider": "openai model=gpt-image-2 size=1024x1536 quality=high"`. If a render fails with a missing-key error, the fix is Project Settings → Integration — or keep momentum with a stock voiceover / an audio file URL (see "Stock assets" (https://creatomate.com/llms/stock-assets.md)) and configure the provider afterward.