alexgenovese/sdxl-custom-model 🔢📝 → 🖼️

▶️ 1.3K runs 📅 Oct 2023 ⚙️ Cog 0.8.6
lora sdxl text-to-image

About

Custom improvements like a custom callback to enhance the inference | It's a WIP and it may causes some wrong outputs

Example Output

Prompt:

"a full body realistic photo of beautiful woman 40 y.o, walking in Madison Square Garden, smile, blue eyes, short hair, dark makeup, hyperdetailed photography, soft light, ((full body)), ((3/4 view))"

Output

Example output

Performance Metrics

32.34s Prediction Time
115.71s Total Time
All Input Parameters
{
  "width": 1024,
  "height": 1024,
  "prompt": "a full body realistic photo of beautiful woman 40 y.o, walking in Madison Square Garden, smile, blue eyes, short hair, dark makeup, hyperdetailed photography, soft light, ((full body)), ((3/4 view))",
  "refiner": true,
  "denoising": 0.8,
  "seed_number": 12345,
  "guidance_scale": 8,
  "num_inference_steps": 35
}
Input Parameters
seed Type: integerDefault: 12345
Seed number
width Type: integerDefault: 1024
Width
height Type: integerDefault: 1024
Height
prompt Type: stringDefault:
Prompt text
lora_url Type: string
Link to lora model trained (TAR) or Safetensors
optimizer Type: numberDefault: 0.8
Define the optimizer value
custom_model Type: string
guidance_scale Type: numberDefault: 8
CFG
negative_prompt Type: stringDefault:
Negative prompt
num_inference_steps Type: integerDefault: 30
Number of inference step
Output Schema

Output

Type: stringFormat: uri

Example Execution Logs
Starting fp16 torch.float32 cuda
Fused all loras in UNet
Applied correction to prompts positive a full body realistic photo of beautiful woman 40 y.o, walking in Madison Square Garden, smile, blue eyes, short hair, dark makeup, hyperdetailed photography, soft light, (full body)1.21, (3/4 view)1.21, detailed eyes, (masterpiece,best quality,ultra_detailed,highres,absurdres)1.20, 4k, professional, intricate details even to the smallest particle, extreme detail of the environment, sharp portrait, well lit, (ultra realistic, realistic)1.90, beautiful shadows, bright, photo quality, masterpiece, 8k, ultra-detailed, highly detailed, sharp focus, award winning, octane render unreal engine, volumetrics dtx, high definition, bokeh, (fashion model)1.50, (realistic, photorealistic)2.00, (fashion-forward)1.30
Applied correction to negative prompts  ,(worst quality, low quality, normal quality, lowres, low details, oversaturated, undersaturated, overexposed, underexposed, grayscale, bw, bad photo, bad photography, bad art)1.40, (watermark, signature, text font, username, error, logo, words, letters, digits, autograph, trademark, name)1.20, (blur, blurry, grainy)1.10, morbid, ugly, asymmetrical, mutated malformed, mutilated, poorly lit, bad shadow, draft, cropped, out of frame, cut off, censored, jpeg artifacts, out of focus, glitch, duplicate, (airbrushed, cartoon, anime, semi-realistic, cgi, render, blender, digital art, manga, amateur)1.30, (3D ,3D Game, 3D Game Scene, 3D Character)1.10, (bad hands, bad anatomy, bad body, bad face, bad teeth, bad arms, bad legs, deformities)1.30, peopleneg, unaestheticXL_Jug6
Applied optimization
Created compel for prompts
Starting inference...
timestep: 946 max: 3.99 min: -4.34 mean: -0.01
  0%|          | 0/28 [00:00<?, ?it/s]
timestep: 919 max: 4.03 min: -4.43 mean: 0.0
  4%|▎         | 1/28 [00:01<00:35,  1.30s/it]
timestep: 892 max: 4.04 min: -4.39 mean: -0.01
  7%|▋         | 2/28 [00:02<00:25,  1.01it/s]
timestep: 865 max: 4.05 min: -4.46 mean: -0.01
 11%|█         | 3/28 [00:02<00:23,  1.06it/s]
timestep: 838 max: 4.04 min: -4.48 mean: -0.02
 14%|█▍        | 4/28 [00:03<00:20,  1.15it/s]
timestep: 811 max: 4.04 min: -4.51 mean: -0.02
 18%|█▊        | 5/28 [00:04<00:19,  1.20it/s]
timestep: 784 max: 4.04 min: -4.53 mean: -0.02
 21%|██▏       | 6/28 [00:05<00:17,  1.24it/s]
timestep: 757 max: 4.03 min: -4.55mean: -0.02
 25%|██▌       | 7/28 [00:06<00:16,  1.26it/s]
timestep: 730 max: 4.03 min: -4.56 mean: -0.03
 29%|██▊       | 8/28 [00:06<00:15,  1.27it/s]
timestep: 703 max: 4.0 min: -4.55 mean: -0.03
 32%|███▏      | 9/28 [00:07<00:14,  1.28it/s]
timestep: 676 max: 3.97 min: -4.54 mean: -0.03
 36%|███▌      | 10/28 [00:08<00:13,  1.29it/s]
timestep: 649 max: 3.96 min: -4.54 mean: -0.03
 39%|███▉      | 11/28 [00:09<00:13,  1.29it/s]
timestep: 622 max: 3.94 min: -4.53 mean: -0.03
 43%|████▎     | 12/28 [00:09<00:12,  1.30it/s]
timestep: 595 max: 3.91 min: -4.51 mean: -0.03
 46%|████▋     | 13/28 [00:10<00:11,  1.30it/s]
timestep: 568 max: 3.88 min: -4.5 mean: -0.03
 50%|█████     | 14/28 [00:11<00:10,  1.30it/s]
timestep: 541 max: 3.84 min: -4.48 mean: -0.03
 54%|█████▎    | 15/28 [00:12<00:09,  1.30it/s]
timestep: 514 max: 3.8 min: -4.45 mean: -0.03
 57%|█████▋    | 16/28 [00:12<00:09,  1.30it/s]
timestep: 487 max: 3.75 min: -4.42 mean: -0.03
 61%|██████    | 17/28 [00:13<00:08,  1.30it/s]
timestep: 460 max: 3.7 min: -4.37 mean: -0.03
 64%|██████▍   | 18/28 [00:14<00:07,  1.31it/s]
timestep: 433 max: 3.64 min: -4.32 mean: -0.03
 68%|██████▊   | 19/28 [00:15<00:06,  1.31it/s]
timestep: 406 max: 3.57 min: -4.27 mean: -0.03
 71%|███████▏  | 20/28 [00:15<00:06,  1.31it/s]
timestep: 379 max: 3.5 min: -4.2 mean: -0.03
 75%|███████▌  | 21/28 [00:16<00:05,  1.31it/s]
timestep: 352 max: 3.43 min: -4.13 mean: -0.03
 79%|███████▊  | 22/28 [00:17<00:04,  1.31it/s]
timestep: 325 max: 3.36 min: -4.05 mean: -0.03
 82%|████████▏ | 23/28 [00:18<00:03,  1.31it/s]
timestep: 298 max: 3.3 min: -3.96 mean: -0.03
 86%|████████▌ | 24/28 [00:19<00:03,  1.31it/s]
timestep: 271 max: 3.29 min: -3.87 mean: -0.03
 89%|████████▉ | 25/28 [00:19<00:02,  1.31it/s]
timestep: 244 max: 3.27 min: -3.8mean: -0.02
 93%|█████████▎| 26/28 [00:20<00:01,  1.31it/s]
timestep: 217 max: 3.26 min: -3.74 mean: -0.02
 96%|█████████▋| 27/28 [00:21<00:00,  1.31it/s]
100%|██████████| 28/28 [00:22<00:00,  1.31it/s]
100%|██████████| 28/28 [00:22<00:00,  1.27it/s]
timestep: 197 max: 3.13 min: -3.54 mean: -0.02
  0%|          | 0/8 [00:00<?, ?it/s]
timestep: 169 max: 3.01 min: -3.35 mean: -0.02
 12%|█▎        | 1/8 [00:00<00:04,  1.58it/s]
timestep: 141 max: 2.9 min: -3.17 mean: -0.02
 25%|██▌       | 2/8 [00:01<00:03,  1.59it/s]
timestep: 113 max: 2.8 min: -2.98 mean: -0.02
 38%|███▊      | 3/8 [00:01<00:03,  1.59it/s]
timestep: 85 max: 2.7 min: -2.79 mean: -0.03
 50%|█████     | 4/8 [00:02<00:02,  1.59it/s]
timestep: 57 max: 2.58 min: -2.58 mean: -0.03
 62%|██████▎   | 5/8 [00:03<00:01,  1.59it/s]
timestep: 29 max: 2.4 min: -2.42 mean: -0.03
 75%|███████▌  | 6/8 [00:03<00:01,  1.59it/s]
timestep: 1 max: 2.45 min: -2.42 mean: -0.03
 88%|████████▊ | 7/8 [00:04<00:00,  1.59it/s]
100%|██████████| 8/8 [00:05<00:00,  1.59it/s]
100%|██████████| 8/8 [00:05<00:00,  1.59it/s]
Inference took:  31.832138061523438 34902
Version Details
Version ID
f92c17c92b1e8bf200d56ca237c397b3450f96de039dfcb6a4d033a9c54d782c
Version Created
November 14, 2023
Run on Replicate →