Commission Series #2: Heloise Huthart/Heloise Hut (@heloisehut). Commissioned by ghuni25. Trained on large dataset (1100+ images). Sample images generated with minimal prompting as a demonstration of the LoRA's capabilities, using hi-res.fix and my ShotByElsaBleda LoRA.
Check out my checkpoint merges here: Aneurysm (photoreal), Acoustica (semi-real) and Anabolic (modern anime). Download my other LoRAs here
Feedback (comments/reviews/likes) are most welcome, as are tips/buzz so I can keep making these! I'd love to see the images you create, so please share on here or reach me on Discord (zombiehead91). A lot of work went into this one so I appreciate your responses.
Notice : Usage of this resource is ONLY permitted for educational, personal and non-commercial use. Please observe the permissions listed on this page - do NOT share this resource anywhere else without my knowledge, especially on other sites for commercial use. Please be respectful of the subject and CivitAI terms of use, and avoid all NSFW content. No warranty is provided, and the user bears responsibility for any/all use of this resource. Happy to discuss this further on Discord if unclear
About the subject:
American Instagram fashion model & personal trainer who grew up in Thailand, and launched her modeling career in 2017, focusing on bikini, swimwear, lingerie, apparel, and sports fashion. Her modeling content has helped her earn 550,000 followers and she would eventually go on to be signed by Scout Models. She has worked with many well-known brands such as Boutinela, Inamorata, Victoria’s Secret, San Lorenzo Bikinis, and Nohea Yogini.
Trigger word: ZH_HeloiseH,
Other useful triggers: 1girl, solo, realistic, lips, long hair, blonde hair, blue eyes, looking at viewer, (see tags in LoRA for full list).
Suggested settings
ADetailer, hi-res.fix use is recommended.
Suggested settings : 0.9 to 1, Clip Skip: 1, ADetailer: eyes only
Method: DPM++(3M or 2M) SDE Karras, Steps:30-50, CFG : 6-7, 512x768 + high res fix with 2x upscale
About the training data:
This LoRA was trained on 1100+ curated images (with a LOT of work put in to curate)
Please do not share this LoRA anywhere else without my knowledge, or for any commercial use.
Description
ZH_HeloiseH_v1 - trained on 1100+ images
FAQ
Comments (4)
NICE LORA Milada Moore PLZ！！！
hey there, sorry - not familiar with her. Do you have a dataset compiled? i usually need a large amount of images.
Eventhough I believe i'm using the suggested settings, results aren't quite the same on ComfyUI. Any suggestions? I guess you only use automatic111? (considering learning automatic for the sake of trying to get similar results), on comfyUI they're close but not quite the same, feel yours are kinda better
Hey there, yes I'm using automatic1111. Never used ComfyUI before but could you elaborate on how the results are different? Maybe lower the strength to 0.7-0.9, or adjust the CFG (a higher CFG usually means the gens are closer to the training data)