# Generate nodes Generate nodes call a model to make text, images, video or music. They cost credits each time they run. Last updated: 2026-10-09 **Key facts** - Spacelab's generate nodes, Any LLM, Image Model, Video, Video (references), Lyria 3, Text to speech and Lipsync, call a model and cost credits each time they run. - Spacelab generate nodes take their prompt from a wire, so a canvas starts with a Prompt node, and a setting the chosen model doesn't support is hidden. - A Spacelab Image Model takes up to 10 reference images, or fewer if the chosen model takes fewer, and with Meta Muse it makes 1 to 10 images per run. - Spacelab's Text to speech node reads a script of up to 1,000 characters, about a minute of speech, in a stock ElevenLabs voice. - Spacelab's Lipsync node uses Sync Lipsync 2 and takes audio of up to 30 seconds. Web page: https://sloosh.ai/docs/nodes-generate Every Generate node takes its prompt from a wire, so start with a Prompt node. The settings on a node change with the model you pick: a setting the model doesn't support is hidden. The Models section of the add menu lists each model by name. Each entry says what the node does, then gives the same lines in the same order: Best for, In, Out, Settings, Cost, Limits, Example and Tip. A line that doesn't apply is left out. To find a node on the canvas, search Add node by name. ## At a glance - [Any LLM](https://sloosh.ai/docs/nodes-generate.md#any-llm): writes text, reads images, fills in structured fields. - [Image Model](https://sloosh.ai/docs/nodes-generate.md#image-model): makes images from a prompt and references. - [Video](https://sloosh.ai/docs/nodes-generate.md#video): makes a clip, optionally from a first and last frame. - [Video (references)](https://sloosh.ai/docs/nodes-generate.md#video-references): makes a clip that keeps a product or person the same. - [Lyria 3](https://sloosh.ai/docs/nodes-generate.md#lyria-3): makes music. - [Text to speech](https://sloosh.ai/docs/nodes-generate.md#text-to-speech): reads a script aloud. - [Lipsync](https://sloosh.ai/docs/nodes-generate.md#lipsync): makes a face on video say an audio track. ## Any LLM Node: Any LLM (`text-gen`) - Inputs: `prompt` (Prompt, text), `system_prompt` (System prompt, setting), `reference_images__0` (Image 1, image), `documents__0` (Document 1, document) - Outputs: `text` (Text, text), `json` (Json, text) A text model. It writes, rewrites, describes images and fills in structured fields. - Best for: captions, product copy, or turning a photo into a written prompt for another model. - In: Prompt (required), System prompt, up to 10 images for models that can see, and up to 3 documents from a Document node. - Out: Text, plus JSON when you set Structured output. - Settings: Model, System prompt, and under advanced, Temperature, Max tokens and Structured output. - Cost: credits by length of the prompt and answer, by model. Structured output takes one field per line as `name: type`, for example `headline: string` and `tags: string[]`. The node then returns those fields as JSON: on the JSON port, and as JSON text on Text. - Example: product photo + Prompt (write a 20-word caption for this product) → Any LLM → Text into a Prompt variable. ## Image Model Node: Image Model (`image-gen`) - Inputs: `prompt` (Prompt, text), `aspect_ratio` (Aspect ratio, setting), `reference_images__0` (Image 1, image) - Outputs: `image` (Image, image), `images` (Images, image) Makes an image from a prompt and, optionally, reference images. - In: Prompt (required), Aspect ratio, and up to 10 reference images, or fewer if the model takes fewer. - Out: Image (the first one) and Images (all of them). - Settings: Model, Aspect ratio, Images per run (Meta Muse only, 1 to 10) and Output format; under advanced, Image size (1K, 2K or 4K where the model has them), Quality (where the model has it), Seed, Enable eval and Input labels. - Cost: credits per image, by model and image size. - Example: product photo + Prompt → Image Model (Nano Banana Pro, 4:5) → Upscaler. - Tip: try the same prompt on two models before you build a space on one. Enable eval checks each image against your inputs (and your prompt, if you name it in Input labels) and regenerates a rejected one, up to 3 attempts in one run. Input labels give each wired input a short name, such as Garment or Model, so both the model and the check can tell them apart. Prompt tips per model: [Prompting GPT Image](https://sloosh.ai/docs/prompting-gpt-image.md) and [Prompting Nano Banana](https://sloosh.ai/docs/prompting-nano-banana.md). ## Video Node: Video (`video-gen`) - Inputs: `prompt` (Prompt, text), `first_frame` (First frame, image), `last_frame` (Last frame, image), `aspect_ratio` (Aspect ratio, setting), `duration_seconds` (Duration (s), setting) - Outputs: `video` (Result, video) Makes a clip from a prompt, optionally starting and ending on images you give it. - Best for: motion from a still. When a product or person has to stay the same through the clip, use Video (references). - In: Prompt (required), First frame, Last frame, Aspect ratio, Duration. - Out: Video. - Settings: Model, Aspect ratio (16:9 by default), Resolution, Duration, Generate audio, and per model, Seed, Negative prompt and others. - Cost: credits per second of video, by model and resolution. The node's Run button shows the price for the model and settings you picked. - Limits: Kling models need a first frame. - Example: product photo → First frame, Prompt (slow push in, steam rising from the cup) → Video. ## Video (references) Node: Video (references) (`video-reference`) - Inputs: `prompt` (Prompt, text), `reference_images__0` (Reference image 1, image), `reference_videos__0` (Reference video 1, video), `reference_audios__0` (Reference audio 1, audio), `aspect_ratio` (Aspect ratio, setting), `duration_seconds` (Duration (s), setting) - Outputs: `video` (Result, video) Makes a clip that follows reference images, and for some models reference videos and audio. - Best for: a clip where a product or person has to stay the same from start to end. - In: Prompt (required), reference images, reference videos, reference audio. The video and audio ports appear only when the model accepts them. - Out: Video. - Settings: the same as Video. - Cost: credits per second of video, by model and resolution. The node's Run button shows the price for the model and settings you picked. - Limits: Gemini Omni takes one reference video of up to 10 seconds, up to 6 images, and makes 16:9 or 9:16 only. - Example: Characters (your model) + product photo + Prompt → Video (references). Wire the clip and a Lyria 3 or Text to speech track into Merge audio + video. To point at one reference in the prompt, write its tag: `@Image 1` is the first reference image, `@Video 1` the first reference video, `@Audio 1` the first clip. Seedance reads them as exact pointers. Type `@` in the Prompt node to pick them; see [Tag with @](https://sloosh.ai/docs/tagging.md#in-spacelab). Prompt tips for Omni, including its image tags: [Prompting Gemini Omni](https://sloosh.ai/docs/prompting-gemini-omni.md). ## Lyria 3 Node: Lyria 3 (`lyria-3`) - Inputs: `prompt` (Prompt, text), `reference_images__0` (Reference image 1, image) - Outputs: `audio` (Audio, audio) Makes music from a prompt. - In: Prompt (required), up to 10 reference images to set the mood. - Out: Audio (MP3). - Settings: Model. Clip makes 30 seconds, Pro makes a full song. - Cost: credits per track. - Example: Prompt (warm lo-fi beat, no vocals) → Lyria 3 → Merge audio + video. ## Text to speech Node: Text to speech (`text-to-speech`) - Inputs: `text` (Script, text) - Outputs: `audio` (Audio, audio) Reads a script aloud in a stock ElevenLabs voice. - Best for: voiceovers and spoken lines for ads and product videos. - In: Script (required). Wire a Text Input, Prompt or Any LLM into it. - Out: Audio (MP3). - Settings: Voice. - Cost: credits per run. - Limits: a script can be up to 1,000 characters, about a minute of speech. An empty or longer script fails the run. - Example: Text Input (script) → Text to speech → Merge audio + video. ## Lipsync Node: Lipsync (`lipsync`) - Inputs: `video` (Video, video), `audio` (Audio, audio) - Outputs: `video` (Video, video) Makes the person in a video say an audio track, with Sync Lipsync 2. - In: Video with a visible face, Audio (both required). - Out: Video. - Settings: Model. lipsync-2-pro keeps more detail on close-ups and costs more. - Cost: credits per run, by model. - Limits: audio can be up to 30 seconds; longer audio fails the run. The result is as long as the shorter input, so use a video at least as long as the audio. - Example: Video Input (talking head) + Text to speech → Lipsync. ## Frequently asked questions ### Which Spacelab node keeps a product the same through a video? Use Video (references) in Spacelab: it makes a clip that follows reference images, and for some models reference videos and audio, so a product or person stays the same from start to end. The plain Video node suits motion from a still, with an optional first and last frame. ### Can Spacelab add music to a video? Yes. Lyria 3 makes music from a prompt in Spacelab: Clip makes 30 seconds and Pro makes a full song, as an MP3. Wire the track and your video into Merge audio + video, which replaces the video's own sound and keeps the video's length. ### How do I get structured JSON from a text model in Spacelab? Set Structured output on Spacelab's Any LLM node, one field per line as `name: type`, for example `headline: string` and `tags: string[]`. Any LLM then returns those fields as JSON on its JSON port, and as JSON text on its Text port. ### Can Spacelab check and redo a bad AI image automatically? Yes. Turn on Enable eval under a Spacelab Image Model's advanced settings. It checks each image against your inputs, and your prompt if you name it in Input labels, and regenerates a rejected image, up to 3 attempts in one run. ### Why does a Kling video in Spacelab need an image? Kling models in Spacelab's Video node need a first frame, so wire an image into the First frame port before you run. For other models, the Video node's First frame and Last frame ports are optional, and the clip can come from the prompt alone. ### How do I make a talking-head video in Spacelab? Wire a Video Input with a visible face and a Text to speech node into Lipsync, which makes the person say the audio with Sync Lipsync 2. The Lipsync result is as long as the shorter input, so use a video at least as long as the audio.