← Docs Home · Model Catalog · HTTP API · Troubleshooting

Wan 2.2 T2V A14B (14B) LoRA Inference with Diffusers (AI Toolkit-trained)

API model id: wan22_14b_t2v
URL slug: wan22-14b-t2v

This page documents the reference Diffusers inference pipeline for wan22_14b_t2v (Wan 2.2 T2V A14B (14B)). It is designed for running LoRAs trained with ostris/ai-toolkit while minimizing training preview vs inference mismatch. If you are trying to reproduce AI Toolkit sample previews, treat the code linked below as the source of truth (scheduler wiring, resolution snapping, LoRA application, and conditioning).

Run in the cloud (optional): If you want to reproduce the examples on this page in a pinned runtime without local CUDA/driver setup (and reduce preview‑vs‑inference drift), run it via RunComfy’s Cloud AI Toolkit (Train + Inference). 👉 You can open it here: Cloud AI Toolkit (Train + Inference)

Quick facts

Field	Value
Pipeline	`src/pipelines/wan22_t2v.py`
Base checkpoint	`Wan-AI/Wan2.2-T2V-A14B-Diffusers`
Defaults	`sample_steps=25`, `guidance_scale=4.0`, `seed=42`
Resolution snapping	Floors width/height to a multiple of 32
Control image	No
Video	Yes (`num_frames=41`, `fps=16` by default)
LoRA scale behavior	MoE LoRA (high/low noise) loaded into `transformer` + `transformer_2`. Scale is set per transformer via `loras[].network_multiplier`.
Needs AI Toolkit	No

Reference implementation (source of truth)

MoE LoRA validation (wan22 special-case): src/api/v1/inference.py
Pipeline implementation: src/pipelines/wan22_t2v.py
Shared behaviors (snapping, seeding, LoRA base logic): src/pipelines/base.py
Request schema (parameter names): src/schemas/request.py
Model ids (enum): src/schemas/models.py
Pipeline registry (model → class): src/pipelines/__init__.py
Executor (prompt processing + caching): src/tasks/executor.py

Minimal API request

{
  "model": "wan22_14b_t2v",
  "trigger_word": "sks",
  "prompts": [
    {
      "prompt": "[trigger] a photo of a person",
      "width": 1024,
      "height": 1024,
      "seed": 42,
      "sample_steps": 25,
      "guidance_scale": 4.0,
      "neg": "",
      "num_frames": 41,
      "fps": 16
    }
  ],
  "loras": [
    {
      "path": "my_lora_job/my_high_noise.safetensors",
      "transformer": "high",
      "network_multiplier": 1.0
    },
    {
      "path": "my_lora_job/my_low_noise.safetensors",
      "transformer": "low",
      "network_multiplier": 1.0
    }
  ]
}

Pipeline behavior that matters

Uses diffusers.WanPipeline but swaps in a transformer from ai-toolkit/Wan2.2-T2V-A14B-Diffusers-bf16 to better match AI Toolkit training samples.
LoRA is MoE format: loras with transformer: "low" / "high", loaded into two transformer stacks.
Implements set_lora_scale() to update both adapters without reload.

Preview-matching notes (training preview vs inference mismatch)

If you load only one side of the MoE LoRA (high or low), the LoRA effect will not match AI Toolkit samples.
This model snaps resolution to a multiple of 32.
LoRA scale is set via loras[].network_multiplier; this pipeline exposes set_lora_scale() so changing scale between requests doesn’t require a full reload.

What to compare when debugging mismatch

Confirm the effective width/height after snapping (the server floors to the divisor).
Match sample_steps and the scheduler family (FlowMatch / UniPC / DDPM differences matter).
Match guidance_scale semantics (some pipelines map it to a different internal parameter).
Match loras[].network_multiplier and whether LoRA scale is dynamic vs fused.
If video: match num_frames and fps (and any frame-count constraints).