Wan2.2-TI2V-5B — Video Generation AI Model
Wan2.2 TI2V 5B handles both jobs from one checkpoint — a prompt alone yields a five-second clip, a prompt plus an image makes that image frame one. Native 720P at 24fps on a single 4090-class card.
What it does
Wan2.2 TI2V 5B does text-to-video and image-to-video from one checkpoint: a prompt on its own imagines the scene, and a prompt with an image makes that image the first frame. Which mode you get depends only on whether you pass an image.
Output is 720p at 24fps, landscape or portrait, up to five seconds. At five billion parameters it was built to run on a single consumer card, which is what the high-compression Wan2.2-VAE buys. Prompts work in English and in Chinese.
When to use it
- Both modes from one model, with no checkpoint to swap between
- Animating a still, with the image as the video's first frame
- 720p at 24fps, in landscape or in portrait
- Clips of two to five seconds, chosen by frame count
- Prompting in Chinese as readily as in English
Details
- CategoryVideo Generation
- Year2026
- LicenseApache-2.0
Compliance & Provenance
- ProviderAlibaba (open weights) · Specialized
- EU AI Act RiskLimited Risk
- Art. 50 TransparencyRequired — AI-generated outputs are marked
Inputs & Outputs
- TextInputstring
Text prompt for video generation
- ImageInputimageoptional
Optional first-frame image. When provided, the image becomes the first frame of the generated video and the model animates subsequent motion guided by the text prompt (image-to-video mode). When omitted, the model generates the entire video from the prompt alone (text-to-video mode). The first frame is locked to this image throughout denoising — middle/last-frame conditioning is not supported.
- VideoOutputvideo
Generated video
Parameters
- resolutionstringdefault 1280x704 (16:9)options: 1280x704 (16:9), 704x1280 (9:16)
Resolution (W×H, ratio).
Output resolution. Wan2.2-TI2V-5B supports exactly 2 sizes: '1280x704 (16:9)' (default, landscape) and '704x1280 (9:16)' (portrait). 720P is 1280x704 here, not 1280x720, because of the high-compression VAE's spatial stride. Other sizes are rejected by the model.
- num_framesintegerdefault 121 frames (~5s @ 24fps)options: 49 frames (~2s @ 24fps), 73 frames (~3s @ 24fps), +3 more
Frame count (Wan VAE multiples).
Total frames to generate. Wan VAE requires (4n+1) frames; training baseline is 121 (5s @ 24fps native). Other values are valid but quality drops as you move further from 121. Presets: '49 frames (~2s)', '73 frames (~3s)', '97 frames (~4s)', '121 frames (~5s)' (default, training baseline), '145 frames (~6s)'. Longer clips proportionally more expensive in VRAM/time.
- fpsintegerdefault 24options: 16, 24
Output FPS (24 native).
Output frame rate. 24 (default) is Wan2.2-TI2V-5B's native training fps per upstream wan_ti2v_5B.py (sample_fps=24) — recommended for natural motion. 16 just repacks the same frames into a slower-fps container, producing slower-looking playback.
- stepsintegerdefault 50range 20–60
Diffusion steps (20–60).
Number of denoising steps. Default 50 — upstream wan_ti2v_5B.py sample_steps=50. Drop to 30-40 to save time; raise to 60 only if visible noise/artifacts at default.
- guidance_scalefloatdefault 5range 1–10advanced
CFG scale.
Classifier-free guidance scale. Default 5.0 — upstream wan_ti2v_5B.py sample_guide_scale=5.0 (single value; TI2V-5B is dense, not MoE, so no second guidance scale). Higher = closer to prompt at cost of motion variety; lower = freer composition.
- negative_promptstringdefault advanced
Additional negative prompt.
Extra negative prompt text appended to Wan's default negative prompt (which already covers common quality issues). Leave empty for the default. Add specific terms to discourage e.g. 'cartoon, watermark, hands'.
- seedintegerdefault -1range -1–2147483647advanced
Random seed (-1 = random).
Random seed for reproducibility. -1 (default) selects a fresh random seed every run. Set a fixed non-negative integer to reproduce a previous video (same seed + same prompt + same params → same output).
Tags
Alternatives in Video Generation
- Cosmos3 Nano
NVIDIA world model for text-, image- and video-to-video, with optional synced audio.
- Helios Base
On-device text-to-video, up to 60s @ 24fps. 3-10 min.
- Helios Distilled
Fast distilled text-to-video, fixed 640x384. 1-5 min.
- LatentSync 1.6 (Lip Sync)
Re-sync a talking video's lips to a new voice.
- LTX-2.5
Lightricks LTX-2.5 distilled 22B: text, or text plus a first-frame image, to a video with a synchronized soundtrack generated alongside it.
- LTX-2.5 Audio-to-Video
Make a talking video from one picture and a voice.
- MiniMax-H3 Ref2VA
MiniMax-H3 Ref2VA (omni-reference) generates a 5-15s 24fps video with native stereo audio from a prompt plus a reference image and a reference clip, that clip's own soundtrack included. Video and its soundtrack come out of one denoising loop, so lip movement and sound land in sync. ~144GB bf16 weights — offloaded component-by-component, minutes-scale per clip.