Type | |
Stats | 4,176 |
Reviews | |
Published | May 30, 2023 |
Base Model | |
Training | Steps: 2 Epochs: 4,000 |
Trigger Words | vintage art illustration |
Hash | AutoV2 526BC3E643 |
This art style was based on vintage art illustrations of posters and advertisements from the 1950s.
Sample images for training:
Trained this on 20 images with 4,000 steps and 2 epochs.
Recommended samplers: Euler a, DPM++ 2M Karras, and DPM++ SDE Karras
CFG Scale: ~5.5-7
LoRA weight: ~0.8-1
Checkpoints: I still think Deliberate is the best choice since I tested all of my images on this checkpoint and it gives me the most accurate images to my prompts as much as possible.
Denoising strength (if you're doing hires. fix): ~0.55-0.7
I generate the most accurate results with prompts that start with this:
illustration, vintage art
Negative prompt that works well:
deformed, bad anatomy, disfigured, poorly drawn face, mutation, mutated, extra limb, ugly, disgusting, poorly drawn hands, missing limb, floating limbs, disconnected limbs, malformed hands, blurry, ((((mutated hands and fingers)))), duplicate images, duplicate, (deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation, EasyNegative, ng_deepnegative_v1_75t
EasyNegative and DeepNegative are very recommended.
Suggestions and feedback are always welcome! Enjoy!