Vimoo-1.0-Lite-ComfyUI

Vimoo-1.0-Lite-ComfyUI contains the ComfyUI weights for Vimoo-1.0-Lite TI2V, one ≈9.5B model for both text-to-video (T2V) and image-to-video (I2V), and for Vimoo-1.0-Lite TI2V Distilled, a distilled LoRA adapter for the same model. It also includes the text encoders, vision encoder and VAE. Use it with ComfyUI-Vimoo, which provides the nodes and workflow templates.

See the Vimoo 1.0 repository for the model family, online API, system overview, architecture and evaluation.

🔥 News

• [2026-10] Release Vimoo 1.0: T2V / I2V models, Diffusers inference examples and ComfyUI workflows.

🎬 Showcase

🚀 Quick Start

Requirements

• ComfyUI: 0.34.0 or later.
• Python: 3.10 or later; install dependencies in the environment that launches ComfyUI.
• PyTorch: 2.9.0 with CUDA 13.0 (cu130), torchvision 0.24.0 and torchaudio 2.9.0. Use an NVIDIA driver compatible with CUDA 13.0.
• Hardware: an NVIDIA CUDA GPU.
• Attention: all supplied workflows default to PyTorch SDPA (attention=pytorch), with no additional attention package. Install FA2, FA3, FA4 or SageAttention only when selecting it in Vimoo Performance Options; see performance options.

Use separate environments for ComfyUI and the Diffusers examples; their Transformers dependencies differ. ComfyUI manages its Torch installation. Vimoo-1.0-Lite TI2V templates use the im2col_bounded VAE backend with PyTorch 2.9.0 or later; see VAE decoding for its workspace settings.

Install the nodes

Install ComfyUI, then run these commands from its root directory:

python3 -m pip install "torch==2.9.0+cu130" "torchvision==0.24.0+cu130" "torchaudio==2.9.0+cu130" --index-url https://download.pytorch.org/whl/cu130
git clone https://github.com/Vimoo-AI/ComfyUI-Vimoo custom_nodes/ComfyUI-Vimoo
python3 -m pip install -r custom_nodes/ComfyUI-Vimoo/requirements.txt

Restart ComfyUI and refresh the browser to load the Vimoo nodes and templates.

Download weights

Run these commands from the ComfyUI root directory. Install the Hugging Face CLI with python3 -m pip install -U huggingface_hub if hf is not available. Download the Lite checkpoint with the shared text encoders, SigLIP vision encoder and VAE:

hf download Vimoo-AI/Vimoo-1.0-Lite-ComfyUI \
  diffusion_models/vimoo_v1_lite_bf16.safetensors \
  diffusion_models/vimoo_v1_lite_bf16.safetensors.vimoo.json \
  text_encoders/vimoo_v1_qwen2_5_vl_7b_bf16.safetensors \
  text_encoders/vimoo_v1_glyph_byt5_fp32.safetensors \
  clip_vision/vimoo_v1_siglip_so400m_fp16.safetensors \
  vae/vimoo_v1_wan22_ti2v_vae_fp32.safetensors \
  --local-dir models

For Vimoo-1.0-Lite TI2V Distilled, also download its adapter:

hf download Vimoo-AI/Vimoo-1.0-Lite-ComfyUI \
  loras/vimoo_v1_ti2v_distilled_lora_fp32.safetensors \
  loras/vimoo_v1_ti2v_distilled_lora_fp32.safetensors.json \
  --local-dir models

The downloaded files use this layout under ComfyUI/models/:

ComfyUI/models/
├── diffusion_models/
│   ├── vimoo_v1_lite_bf16.safetensors
│   └── vimoo_v1_lite_bf16.safetensors.vimoo.json
├── text_encoders/
│   ├── vimoo_v1_qwen2_5_vl_7b_bf16.safetensors
│   └── vimoo_v1_glyph_byt5_fp32.safetensors
├── clip_vision/
│   └── vimoo_v1_siglip_so400m_fp16.safetensors
├── vae/
│   └── vimoo_v1_wan22_ti2v_vae_fp32.safetensors
└── loras/
    ├── vimoo_v1_ti2v_distilled_lora_fp32.safetensors
    └── vimoo_v1_ti2v_distilled_lora_fp32.safetensors.json

Keep each .vimoo.json and .json file next to its weights. The nodes read the supported tasks, the matching base checkpoint and the sampling settings from them.

Generate video

1. Open the workflow for your model from ComfyUI's template browser, or drag its JSON from the models and workflows table onto the canvas.
2. Select the checkpoint in Vimoo Model Loader. For a distilled workflow, check its adapter inside the generation subgraph. Keep the saved node connections and defaults.
3. Edit the prompt. For I2V, upload a reference image in the input node; it is scaled and center-cropped to the output size.
4. Click Queue. Videos are saved to ComfyUI/output/.

Vimoo-1.0-Lite workflows select t2v or i2v in the input node. Vimoo-1.0-Lite TI2V shares its checkpoint across both tasks, and Vimoo-1.0-Lite TI2V Distilled also shares its adapter.

📋 Models and Workflows

Model	Checkpoint	LoRA	Workflow
Lite TI2V	vimoo_v1_lite_bf16.safetensors	None	vimoo_v1_ti2v_lite.json
Lite TI2V Distilled	vimoo_v1_lite_bf16.safetensors	vimoo_v1_ti2v_distilled_lora_fp32.safetensors	vimoo_v1_ti2v_lite_distilled.json

The root recipe.json and inference_defaults.json apply to Vimoo-1.0-Lite TI2V only; Vimoo-1.0-Lite TI2V Distilled reads its settings from the adapter .json file.

🎛️ Workflow Defaults

The templates provide model loading, performance, generation and video output nodes; I2V adds an image input. Expand the generation subgraph to select supporting encoder, VAE and adapter files. Vimoo-1.0-Lite TI2V exposes steps and text CFG; Vimoo-1.0-Lite TI2V Distilled keeps its fixed sampling settings.

resolution=auto fits the oriented I2V image ratio to a 1280 × 720 pixel budget, rounding both axes down to multiples of 16. A square image produces 960 × 960; a portrait 9:16 image produces 720 × 1280. Choose manual to enter width and height. In T2V mode, auto uses 1280 × 720. The initial frame count is 121 at 24 fps.

Vimoo-1.0-Lite TI2V and Vimoo-1.0-Lite TI2V Distilled require BF16 compute; select dtype=bf16 in Vimoo Model Loader. FP16 and FP32 compute overrides are rejected for these profiles.

The Distilled workflow uses CFG 1 and only the positive prompt; keep its saved sampling grid and matching adapter metadata. Keep the generation FPS and the video output FPS equal.

Memory, attention and VAE decoding options are described in the ComfyUI-Vimoo README.

📝 Prompting Guide

Use English prompts. The current open-source models have been validated only with English prompts. Generation quality and prompt adherence in other languages have not been validated. We recommend translating prompts into English before generation.

Write clear, detailed prompts. Specific descriptions of the subject, scene, action sequence, and camera movement can help produce results that better match your intent than short, vague prompts. See the Prompt Guide for structure and examples, and focus on relevant, consistent visual details.

The Prompt Guide describes automatic prompt enhancement in the hosted service. Local examples use your input prompt directly; follow the guide's writing structure and examples when preparing prompts.

📜 License

Vimoo model weights and official adapters in this repository are provided under the Apache License 2.0.

• Commercial use, modification, and redistribution are permitted under the terms of the license.
• Redistribution must include a copy of the license, retain applicable copyright and attribution notices, and identify modified files.
• Attribution and third-party component notices are provided in THIRD_PARTY_NOTICES.md. Third-party components remain subject to their own licenses.