I am admittedly kind of a noob at this, but if you enjoy my humble work, any donations are greatly appreciated: https://ko-fi.com/insult_to_ninjary
Your generosity helps me stay motivated to keep improving and to keep trying new things!
Prompting:
***If you end up with a totally fried output image (extremely deformed or just noise), try adding or removing a few tokens to/from your prompt until you get a normal output. This is a very rare occurrence, but I did experience it once or twice during my testing. It may be related to very occasional bad luck when using weighted prompts.
Positive Prompt:
No special positive prompt is totally necessary, but to maximize realism, I recommend adding something like the below to the beginning of your prompt. You don't have to use these prompt strings exactly, just something in the same ballpark that pushes the model towards a more realistic, photographic style.
Professional aesthetic:
RAW photo, DSLR, 35mm film grain, photorealistic, hyperrealistic, candid photograph, shot on Sony A7III, 85mm f/1.8, shallow depth of field,
Amateur aesthetic:
snapshot photo, candid snapshot, amateur photography, uneven exposure, overexposed, slightly blurry, slightly tilted angle, awkward framing, low quality jpeg,
Negative Prompt:
Again, no special negative prompting is really necessary, but you can use something like the below to push the model further away from non-realistic art styles. Like above, you don't need to use this exact prompting, just something in the same spirit.
anime, illustration, illustrated, cartoon, animated, animation
At a Glance:
• Style: Realistic
• Works with LoRAs?: Technically yes, but if it actually looks good may be hit or miss
• VAE: sdxl_vae is baked (but you can override if desired)
• Suggested parameters: DPM++ 3M SDE, Karras, 30 steps, 832x1216/1216x832, CFG 3 (but feel free to experiment!)
• Hires. fix: 1.5x, 30 steps, use a quality upscaler (like 4x-UltraSharpV2), denoising strength 0.5 - 0.3
• Ext