Skip to main content
Generators make a new file from a prompt. Image and video generation can also take a file and start from it — a reference picture, or the first frame of a clip. Voice, music and sound effects take no file and must be the first step. Each model is listed as its own skill page. Leave model off to use the default.
Signed-in Workbench with the default image model as step 1 and Make 1 picture

Make an image. Default model is Nano Banana 2 at 20 credits. Shape and size sit on the step.

Make an image

Slug: generate. Default 20 credits / image, on Nano Banana 2. A bigger size reprices per model — Nano Banana 2 runs 30 credits at 2048px, and Gemini 3 Pro Image 75 credits at 4K. GPT Image 2 is drawn at 1024px and priced by quality instead: 2 credits low, 14 medium, 53 high. It can also leave the background transparent (background: transparent). GPT Image 2 and Flux Schnell work from a prompt only; the other models also take a reference picture.

Text to video

Slug: text-to-video. Billed per second of the clip. On Seedance 2.5, resolution: draft makes a 480p draft at 56 credits a second; put Finish the draft (finish-draft, 291 a second) under it, or run it on the draft within seven days, for the same clip at 1080p. Default model is Veo 3.1 Lite at 13 credits / second (720p). A dearer resolution reprices per second. Lengths run from three to thirty seconds; each model offers only the lengths it makes — MiniMax H3 five to fifteen, Seedance 2.5 four to fifteen.

Effects

effect picks a named shot with the camera work already written: crash-zoom, dolly-zoom, orbit, bullet-time, fpv-drone, earth-zoom-out, disintegrate, on-fire, chrome, melt, explode, freeze, levitate, lightning, or none (the default). With an effect and a picture, prompt may be left out; anything sent becomes the scene the effect happens in. Without a picture, prompt is still required, to say who or what it happens to. An effect costs what the model and length cost — it changes the words the model is sent, not the price. Animate a photo (animate) runs on Veo 3.1 Lite from a still and uses the same per-second rates: 13 credits / second at 720p, 20 at 1080p.

Text to voice

Slug: text-to-voice. Default 12 credits / minute on Gemini 3.1 Flash TTS. MiniMax Speech 2.8 HD takes an emotion for the read, and pitch, timbre and intensity under More settings.
Signed-in Workbench with Gemini 3.1 Flash TTS as step 1 and Make 1 track for 12 credits

Text to voice takes no file and must be the first step.

Design a voice

Slug: voice-design. 770 credits a voice. Describe the voice in prompt; it is made on MiniMax, reads line in it (or introduces itself), and is saved to your account as voice-name. Then pick it in Text to voice: model minimax-speech-2.8-hd, and the voice under My voice (my-voice). MiniMax deletes a designed voice nobody reads with for a week, so the run reads with it once straight away, which keeps it. It designs a new voice from words; it doesn’t clone a recording.

Make music

Slug: music. Billed in the provider’s unit — Stable Audio and Lyria bill per track, ElevenLabs Music bills per minute. Lyria sets its own length and takes no length. length is a setting and, on a per-minute model, is billed on: a three-minute bed bills three minutes, not the one-minute floor.

Make a sound effect

Slug: sound-effects. Default 3 credits / sound on ElevenLabs Sound Effects, at five seconds. ElevenLabs bills the seconds, so its price follows length: 1 credit for 2 seconds, 5 for 10, 11 for 22. CassetteAI is 3 credits at any length. loop (ElevenLabs only) writes a sound that loops seamlessly. Takes no file and must be the first step.

From the API

Send a prompt, and optionally a model. Image and video generation also take a file or url.
See Run a skill. Prompts are capped at 2,000 characters.