Which model can be used to produce Lego style videos?

There seems to be a usable LoRA for this:

The most practical open-source-ish route I found is probably ComfyUI + Wan2.1 14B T2V + a Lego-style LoRA , rather than looking for a hosted HF Space that already does it.

The relevant LoRA is:

• Remade-AI/Lego

It is described as a “Lego Style LoRA for Wan2.1 14B T2V”. The model card says it is trained on the Wan2.1 14B T2V base model, and it includes:

• the LoRA file: lego_35_epochs.safetensors
• a ComfyUI workflow: wan_txt2vid_lora_workflow.json
• a trigger phrase
• example prompts
• suggested settings

The trigger phrase is:

l3g0_5ty13 Lego animation style

The model card’s recommended starting settings are:

LoRA Strength: 1.0
Embedded Guidance Scale: 6.0
Flow Shift: 5.0

So I would start from the workflow on the LoRA page itself, then use the general Wan2.1 ComfyUI setup notes from the ComfyUI examples.

Useful links:

• Remade-AI/Lego — Lego-style Wan2.1 14B T2V LoRA
• Official ComfyUI Wan2.1 examples
• ComfyUI docs: Wan2.1 Video Examples
• Comfy-Org/Wan_2.1_ComfyUI_repackaged
• Kijai/ComfyUI-WanVideoWrapper

A rough ComfyUI path would be:

1. Update ComfyUI.
2. Set up Wan2.1 T2V in ComfyUI.
3. Use the workflow from the Remade-AI/Lego model page.
4. Put the Wan diffusion model, text encoder, VAE, and LoRA in the expected ComfyUI model folders.
5. Start prompts with the trigger phrase: l3g0_5ty13 Lego animation style.

The ComfyUI Wan examples say the Wan diffusion models go here:

ComfyUI/models/diffusion_models/

The common Wan2.1 supporting files are usually placed like this:

ComfyUI/models/text_encoders/
ComfyUI/models/vae/
ComfyUI/models/clip_vision/

For the LoRA, use the LoRA path expected by your workflow / custom nodes. In a normal ComfyUI LoRA setup, this is usually something like:

ComfyUI/models/loras/

If using Kijai’s WanVideoWrapper workflow, follow the node/workflow expectations there, because the Remade model card says their ComfyUI workflow is based on a modified version of Kijai’s Wan Video Wrapper workflow with a Wan LoRA node connected to the base model.

A starting prompt could be:

l3g0_5ty13 Lego animation style, a plastic toy-brick minifigure walking through a miniature brick-built city, glossy plastic surfaces, stop-motion animation look, cinematic lighting, shallow depth of field, smooth camera movement

Another example:

l3g0_5ty13 Lego animation style, a close-up shot of a toy-brick pirate minifigure with a black tricorn hat, eye patch, and red coat, holding a tiny silver cutlass, with a colorful brick-built ocean scene in the background, glossy plastic material, cinematic lighting, shallow depth of field

I would treat this mainly as a text-to-video route, because the LoRA is specifically for Wan2.1 14B T2V. It may or may not behave correctly if you try to use it in an image-to-video workflow without adaptation.

Also, I did not find a public HF Space that appears to run this exact LoRA out of the box. There are Wan-related Spaces, but the more reliable recommendation here seems to be ComfyUI using the LoRA workflow from the model card.

One caveat: for public/commercial use, it may be safer to describe the intended look as “toy-brick”, “plastic minifigure”, “brick-built”, or “stop-motion toy animation” rather than relying only on the LEGO trademark in user-facing text.