Heliosoph/dreamshaper-hyper-onnx

DreamShaper + Hyper-SD (4-step) — ONNX

ONNX export of Lykon/DreamShaper with the ByteDance/Hyper-SD 4-step LoRA fused into the UNet. SD 1.5 architecture, 512×512 native, designed to run with the Euler scheduler at CFG = 1 in 4 inference steps.

DreamShaper is Lykon's stylized SFW fine-tune — leans more illustrative / fantasy than AbsoluteReality, which is more photorealistic. Pick this one for D&D-style art, character portraits with painterly textures, and concept-art-leaning prompts.

This is a converted artifact, not a new model. All training credit belongs to Lykon (DreamShaper) and ByteDance (Hyper-SD).

What this repo contains

model_index.json
feature_extractor/
scheduler/
text_encoder/
tokenizer/
unet/                   # DreamShaper UNet + Hyper-SD-15 4-step LoRA fused in
vae_decoder/
vae_encoder/

unet/model.onnx is paired with unet/model.onnx_data (external-weights file).

How it was produced

1. 1.Load Lykon/DreamShaper via diffusers (bundled VAE).
2. 2.Load ByteDance/Hyper-SD/Hyper-SD15-4steps-lora.safetensors via peft, fuse_lora() it into the UNet.
3. 3.Save the fused pipeline to a temp directory.
4. 4.optimum-cli export onnx --model <temp> <output>.

Toolchain: optimum 1.24.0, diffusers 0.31.0, transformers 4.45.2, torch 2.4.x (CUDA 12.4). Full conversion script: `scripts/export-dreamshaper-hyper.ps1`.

Inference notes

License

CreativeML OpenRAIL-M (inherited from SD 1.5 + DreamShaper) + the Hyper-SD LoRA's OpenRAIL-M. Both license files are included in this repo. By using this model you accept those terms.

Citation

@misc{lykon-dreamshaper,
  author = {Lykon},
  title  = {DreamShaper},
  howpublished = {\url{https://huggingface.co/Lykon/DreamShaper}}
}
@article{ren2024hypersd,
  title   = {Hyper-SD: Trajectory Segmented Consistency Model for Efficient Image Synthesis},
  author  = {Ren, Yuxi and others},
  journal = {arXiv preprint arXiv:2404.13686},
  year    = {2024}
}