pruna / wan-t2v

## Overview WAN T2V is a text-to-video model from Pruna. It takes a detailed scene description and synthesizes a short video clip as a sequence of generated frames. It then packages the result as a playable MP4, which makes it easy to drop into edits or prototypes. The main value is fast iteration on motion, camera ideas, and style without filming. ## What you can build - 5-second concept clips for storyboards and pitch decks - Vertical social posts (9:16) from a single written shot description - Landscape b-roll ideas (16:9) for ads, trailers, and mood reels - Style explorations like cinematic, anime, macro, or watercolor looks - Quick motion tests for “camera moves” like pan, dolly, and drone shots ## Inputs - A required written description of the scene, including action, motion cues, camera movement, and visual style. - A required output resolution choice of 480p or 720p. - An optional aspect ratio choice of 16:9 (landscape) or 9:16 (vertical). For 16:9, outputs map to 832×480 or 1280×720. For 9:16, outputs map to 480×832 or 720×1280. - An optional toggle to enable frame interpolation for smoother motion. When enabled, the output video is upsampled to 30 fps. - An optional toggle for a fast mode. Turn it off when you want higher visual quality. - An optional toggle to auto-optimize the prompt by translating it to Chinese before generation. - An optional motion shift control (a decimal number from 1.0 to 20.0). Higher values tend to increase visible motion. - An optional seed value (an integer from 0 to 99,999,999) to help reproduce a result. Set it to 0 to randomize. - An optional toggle to skip the safety checker on generated content. ## Outputs - A single MP4 video clip. - The clip matches the selected resolution and aspect ratio. - If frame interpolation is enabled, the MP4 is delivered at 30 fps for smoother playback. ## Limitations - The model generates short clips (about 5 seconds). It isn’t meant for long scenes. - Output size is limited to 480p or 720p, with only 16:9 or 9:16 framing. - Text-only conditioning can miss exact details like logos, specific text, or consistent identity across frames. - Fast mode can reduce fine detail, increase artifacts, or soften textures. - Results depend heavily on prompt quality. Vague prompts often produce unclear motion or unintended scene changes. - Seeds can improve repeatability, but changes to resolution, aspect ratio, or motion settings can still change outcomes. ## Safety & compliance - The model can run with a safety checker. If you disable it, you take responsibility for the output. - Don’t generate sexual content, nudity, or graphic violence. - Don’t generate content involving minors in sexual contexts. - Respect privacy and likeness rights. Get permission before depicting real people.