sdxs-2b alpha version - Checkpoint for Other

sdxs-2b - alpha version

NSFW

A free, compact, and fast text-to-image model developed by AiArtLab for training on consumer-grade GPUs.

CosmosTransformer3D: 2b parameters

Qwen3.5: 0.8b parameters

Qwen vae: 16ch8x

Resolution: Default 768x1152, trained from 576 to 1152 with step 64

Limitations: trained on small datasets ~2.5kk, focused on art / illustrations / anime and photo

Captions: danbooru, natural (short/medium), trained with 250 max toks

Comfyui pipeline

Description

XS Size, Excess Quality

Comments (2)

Seii1Sep 8, 2026

CivitAI

I wonder if this would be better than anima

Cybel_CCTS

Author

Sep 8, 2026

The two models share the same CosmosTransformer3D backbone and both have 2B parameters, but there are major differences in positioning, training data, licensing and native features.

SDXS-2B supports a broad spectrum of art styles rather than being anime-only.

It accepts Danbooru tags plus short/long natural language prompts with a 250-token maximum. Its built-in automatic prompt expansion is beginner-friendly: simple Danbooru tags get automatically expanded to generate high-quality imagery.

SDXS drawbacks: It remains in preview. Its training dataset is smaller than Anima’s, so it delivers less refined results for anime characters and subtle artistic details.

Checkpoint

Other

by Cybel_CCTS

Download (Beta) View on CivitAI

sdxs-2b

base model

Details

Downloads

Platform

CivitAI

Platform Status

Available

Created

9/8/2026

Updated

9/27/2026

Deleted

Files

sdxs2b_alphaVersion.safetensors

Size:

3.64 GB

SHA256:

fadf4d9a714ac24a8ec97e4e70027df8e28593d94265d5bf329937727f199efb

Mirrors

HuggingFace (1 mirrors)

diffusion_pytorch_model.bf16.safetensors

CivitAI (1 mirrors)

sdxs2b_alphaVersion.safetensors

sdxs-2b_vae.safetensors

Size:

242.05 MB

SHA256:

0c8bc8b758c649abef9ea407b95408389a3b2f610d0d10fcb054fe171d0a8344