| license: creativeml-openrail-m | |
| Hypernetwork trained on some Midjourney V4 portraits. | |
| Euler a and DPM++ samplers work best. CFG scale at 7 and low step count (<50) work well. | |
| Prompts should include words like "portrait", "octane render" and "highly detailed" for best results. | |
| Avoid using face restoration like GFPGAN or CodeFormer if possible |