Wan 2.2 Text-to-Video
Text → VideoText-to-video with Wan 2.2 — generate a short clip directly from a prompt, no source image.
wan22_t2v12 GB VRAM38 GB of weightsCalling it
FlixML has no generate button: you tell your agent what you want and it calls the API on your install. The id is what names this workflow.
POST /api/video/generate
{
"workflow": "wan22_t2v",
"prompt": "..."
}Parameters
promptstrrequiredText prompt: subject, setting, and action. Describe big physical action with strong verbs (walks, jumps, turns). Words like subtle, slowly, or gently produce a near-still clip.
negative_promptstrdefault bright colors, overexposed, static, blurred detailswidthintdefault 640heightintdefault 640lengthintdefault 81Frame count
fpsintdefault 16seedintsteps_highintdefault 2steps_lowintdefault 2total_stepsintdefault 4steps_high + steps_low; passed to KSamplerAdvanced.steps
cfg_highfloatdefault 1cfg_lowfloatdefault 1shiftfloatdefault 5samplerstrdefault eulerschedulerstrdefault simplehigh_modelstrdefault wan2.2_t2v_high_noise_14B_fp8_scaled.safetensorsHigh-noise UNet model
low_modelstrdefault wan2.2_t2v_low_noise_14B_fp8_scaled.safetensorsLow-noise UNet model
vaestrdefault Wan2.1_VAE.safetensorsclipstrdefault umt5_xxl_fp8_e4m3fn_scaled.safetensorslightx2v_high_lorastrdefault wan2.2_t2v_lightx2v_4steps_lora_v1.1_high_noise.safetensorslightx2v_low_lorastrdefault wan2.2_t2v_lightx2v_4steps_lora_v1.1_low_noise.safetensorslightx2v_high_strengthfloatdefault 1lightx2v_low_strengthfloatdefault 1Your install is the source of truth: GET /api/workflows returns these parameters as they are on your machine.
What it costs to run
UNTESTED: not yet run end-to-end with the default fp8_scaled UNets and T2V lightx2v LoRAs. The fp8 .safetensors defaults load through ComfyUI's Dynamic VRAM path, which streams weights from disk instead of holding them in RAM. GGUF files skip that path in ComfyUI-GGUF (PR #427 unmerged) and keep both experts in system RAM, so a GGUF swap needs its own RAM check.