https://huggingface.co/cyanelis/VRChat2176/tree/main

VRChat2176 is a 2 billion parameter text-to-image model strictly designed to generate VRChat-style images.

It was primarily trained on resolutions of 1344x2176 or 2176x1344.

Installing and running

Workflow: The model is natively supported in ComfyUI. Each image above contains a workflow; you can open it in ComfyUI or drag-and-drop to get the workflow. The model files go in their respective folders inside your model directory:

• VRChat2176.safetensors goes in ComfyUI/models/diffusion_models

• qwen_3_06b_base.safetensors goes in ComfyUI/models/text_encoders

• qwen_image_vae.safetensors goes in ComfyUI/models/vae (this is the Qwen-Image VAE, you might already have it)

If you encounter an error with the workflow, try deleting the "Switch (Any)" node, and reconnect the latent correctly.

Generation settings

• Works at resolutions between 1344^2 and 2176^2 pixels.

• 30-50 steps, CFG 4-5.

• A variety of samplers work.

Prompting

The model is trained on Danbooru-style tags.

• Use lowercase for tags, and spaces instead of underscores.

• Recommended positive prefix: "masterpiece, best quality, score_7, safe, "

• Recommended negative: "worst quality, low quality, score_1, score_2, score_3, artist name"

• When using a tag that is different between Danbooru and Gelbooru, prefer the Gelbooru version.

• Prompt weighting works, but needs a weight higher than typically used for SDXL. Example: "(chibi:2)"

Safety tags

• The dataset does not contain R18 content.

Artist tags

Prefix artisthash with “dim1 dim1-”. E.g. "dim1 dim1-59151b77". You must put “dim1 dim1-” in front of the artisthash. The effect will be very weak if you don't.

• dim1 dim1-59151b77

• dim1 dim1-78416eb9

Full tag example

newest, masterpiece, best quality, score_7, thighhighs, white thighhighs, (dim1 dim1-78416eb9, 1girl, umbrella, double bun, solo, holding umbrella, chinese clothes, china dress, dress, twintails, hair bun, looking at viewer, holding, hair ornament, grey eyes, blush, white dress, floral print, open mouth, long hair, blunt bangs, white hair, pelvic curtain, grey hair, white background, bare shoulders, very long hair, breasts, simple background, tassel, small breasts, oil-paper umbrella:1)

Natural language prompting tips

• Natural language prompting is currently not supported.

License

This base model of this model is licensed under the CircleStone Labs Non-Commercial License. The model and derivatives are only usable for non-commercial purposes. Additionally, this model constitutes a "Derivative Model" of Cosmos-Predict2-2B-Text2Image, and therefore is subject to the NVIDIA Open Model License Agreement insofar as it applies to Derivative Models.

If you would like a commercial license, please email [email protected]

Note that the non-commercial restriction applies only to the Model, and not to Outputs (the generated images). You may use generated images commercially.

Examples of allowed commercial use:

• selling images

• paid commissions for images

• generating images to use as concept art or assets for a paid product (e.g. video game or visual novel)

• selling Derivative model weights, if you are operating as an individual (Section 2.c contains a carve-out for this specific use)

Examples of disallowed commercial use without a separate license:

• hosting the model behind an API and charging for access

• hosting the model on a paid online image generation platform

• embedding the model weights inside a monetized game or other product

• using the model to power some feature as part of a larger, monetized product

Built on NVIDIA Cosmos.