This took a bunch of effort generating realistic images from veo based on screenshots.
Her primary scar is often missing unless you ask for it, and very hard to make it the correct color. I used smaller 1:1 AR images for training and model struggles with anatomy with higher LoRA weights (>80) OR large images (above ~1600). I have had a lot of success with LoRA weights in the 60-80 range with highresfix enabled (960x1600 to 1216x2048, 95% second pass weight, CFG 1.5-2.5, 8-10 steps).
I use DrawThings and this has been my general config:
{
"upscaler": null,
"cfg_zero_init_steps": 4,
"width": 1216,
"seed_mode": "Scale Alike",
"compression_artifacts": "disabled",
"target_height": 2048,
"batch_count": 1,
"target_width": 1216,
"refiner_model": null,
"negative_prompt_for_image_prior": true,
"guidance_scale": 1.2000000476837158,
"hires_fix": true,
"hires_fix_height": 1600,
"zero_negative_prompt": false,
"sharpness": 0,
"loras": [{
"file": "naoe_v1_lora_f16.ckpt",
"weight": 0.74,
"mode": "all"
}],
"upscaler_scale": 0,
"tea_cache_threshold": 0.20000000298023224,
"tea_cache": false,
"seed": 1820756551,
"model": "z_image_turbo_1.0_q8p.ckpt",
"compression_artifacts_quality": 43.099998474121094,
"shift": 8,
"hires_fix_strength": 0.949999988079071,
"causal_inference_pad": 0,
"decoding_tile_height": 640,
"negative_original_height": 1024,
"crop_top": 0,
"num_frames": 49,
"tea_cache_max_skip_steps": 3,
"original_height": 2048,
"hires_fix_width": 960,
"causal_inference": 0,
"original_width": 1216,
"cfg_zero_star": false,
"negative_original_width": 576,
"clip_l_text": null,
"negative_aesthetic_score": 2.5,
"resolution_dependent_shift": false,
"tiled_diffusion": false,
"tea_cache_end": -1,
"refiner_start": 0.10700000077486038,
"speed_up_with_guidance_embed": true,
"aesthetic_score": 6,
"open_clip_g_text": null,
"motion_scale": 127,
"tea_cache_start": 5,
"guidance_embed": 3.5,
"mask_blur": 1.5,
"start_frame_guidance": 1,
"prompt": "The subject, NAOE, is a young woman...",
"image_prior_steps": 5,
"batch_size": 4,
"separate_open_clip_g": false,
"controls": [],
"stochastic_sampling_gamma": 0.3,
"diffusion_tile_overlap": 128,
"t5_text_encoder_decoding": true,
"tiled_decoding": false,
"diffusion_tile_width": 1024,
"stage_2_guidance": 1,
"height": 2048,
"decoding_tile_overlap": 128,
"preserve_original_after_inpaint": true,
"guiding_frame_noise": 0.019999999552965164,
"negative_prompt": "",
"mask_blur_outset": 0,
"image_guidance": 1.5,
"diffusion_tile_height": 1024,
"clip_skip": 1,
"separate_t5": false,
"sampler": "DDIM Trailing",
"separate_clip_l": false,
"t5_text": null,
"clip_weight": 1,
"decoding_tile_width": 640,
"fps": 5,
"crop_left": 0,
"stage_2_shift": 1,
"strength": 1,
"steps": 9
}