Wan 2.2 Text-to-Video with Lightning LoRA
Type a prompt, get a short video. This basic Wan 2.2 text-to-video workflow uses the high-noise and low-noise models with fast lightning LoRAs.
- L40S
- Text to video
- Free template
How it works
This is a 17-node workflow built from core ComfyUI nodes. Wan 2.2 text-to-video uses two models: wan2.2_t2v_high_noise_14B_fp8_scaled handles the early, noisy steps and wan2.2_t2v_low_noise_14B_fp8_scaled handles the final refinement. Each has its own Wan2.2-Lightning_T2V-v1.1-A14B-4steps LoRA (HIGH and LOW), applied with LoraLoaderModelOnly.
The prompt is encoded with umt5_xxl_fp8_e4m3fn_scaled through CLIPLoader. EmptyHunyuanLatentVideo creates a latent of 832 by 544 with 81 frames. Two KSamplerAdvanced nodes run in sequence with the euler sampler: the first covers steps 0 to 3 and returns leftover noise, and the second continues from step 3 to the end of 7 steps.
VAEDecode with wan_2.1_vae converts the latent, and CreateVideo and SaveVideo write the file.
Tips
- Adjust the latent size and frame count in
EmptyHunyuanLatentVideoto change resolution and length. More frames take more time and memory. - The default negative prompt is a long Chinese list that targets overexposure, blur and bad hands, so keep it unless you have a reason to change it.
- Because both LoRAs are lightning versions, keep steps low. Raising them gives little benefit.
- Describe subject, motion and camera in the prompt.
What you can use it for
- Short clips from a written prompt
- Fast previews of video ideas
- Testing prompts before a longer render
- A baseline Wan 2.2 workflow to build on
Run it in three steps
Open the template
Click Run on ComfyDeploy. The workflow, its custom nodes and models are set up for you on a cloud machine.
Add your inputs
Provide text prompt and negative prompt in the playground or the ComfyUI editor.
Run, share or deploy
Run it on an L40S, share it with your team as a playground, or deploy it as an API endpoint.
Frequently asked questions
What GPU does this workflow need?
It is set to run on an L40S GPU on ComfyDeploy.
How many steps does it use?
The workflow description says 4 to 7 steps with the lightning LoRAs. The saved settings use 7 total steps split between the two samplers.
How do I run it as an API?
Deploy it on ComfyDeploy and expose the prompt inputs. The graph uses only core ComfyUI nodes.
