Heliosoph/epicrealism-hyper-onnx

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

ONNX export of emilianJR/epiCRealism with the ByteDance/Hyper-SD 4-step LoRA fused into the UNet. SD 1.5 architecture, 512×512 native, Euler scheduler, CFG = 1, 4 steps.

epiCRealism leans heavily toward photorealistic environments and natural lighting. Pick this one over Realistic Vision when the subject is "a place" rather than "a person."

Converted artifact, not a new model. Training credit: emilianJR (epiCRealism), ByteDance (Hyper-SD).

What this repo contains

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

How it was produced

1. Load emilianJR/epiCRealism via diffusers.
2. Load ByteDance/Hyper-SD/Hyper-SD15-4steps-lora.safetensors via peft, fuse_lora() into UNet.
3. Save fused pipeline; run optimum-cli export onnx.

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

Inference notes

Setting	Value
Scheduler	Euler
Steps	4
CFG / guidance scale	1.0
Negative prompt	Skip
Resolution	512×512 native

License

CreativeML OpenRAIL-M, inherited from SD 1.5 + epiCRealism + Hyper-SD. License files included. By using this model you accept those terms.