Sogni: Learn logo

Creating Videos

Sogni Studio generates video from text, images, audio, or other videos. Pick a workflow based on what you've got to start from — a written idea, a still you already love, an audio clip, or a reference video whose motion you want to borrow.

#Workflows

Sogni Studio auto-picks the right workflow based on the inputs you provide:

Workflow When to use it Inputs
Text-to-Video (T2V) Generate motion from a written prompt alone. Prompt
Image-to-Video (I2V) Animate a still image. Optionally pin start and end frames for continuity. Start frame (+ optional end frame) + prompt
Sound-to-Video (S2V) Drive motion from an audio track — music, dialogue, ambient sound. Reference image + audio
Audio-to-Video (A2V) Generate video from audio alone, no reference image. Audio
Motion Transfer Apply another video's motion to your character. Reference image + source video
AnimateReplace Keep a video's motion and scene, swap the subject. Reference image + source video

See Convert Image to Video for the quickest path (a single image, no reference), or Video Create Mode for the full control panel with keyframe-level overrides.

#Models

Video runs on the Sogni Supernet and uses video-specialized models. Current lineup:

  • LTX-2.3 — image-to-video; fast, high motion fidelity.
  • LTX-2.3 Dev — text-to-video; latest generation, longer coherent shots.
  • Wan 2.2 — first-party Sogni video model, still available.

Distilled variants of each are available for faster turnaround at slight quality cost. Pick a model preset in Video Create mode; Studio matches the workflow to what each model supports.

#After generation: stitching clips together

Generated clips can be composed into longer pieces using the Clip Mixer, a desktop-only timeline editor with FPS, aspect-ratio, and re-encode controls. See Clip Mixer.

#Privacy

Generated files are saved only on your Mac in the gallery folder you choose. Supernet rendering processes the job and purges inputs after a brief retention period — your videos don't get added to any public dataset.

#See also