mohsin-riad/upscaler-ultra 🖼️🔢📝❓✓ → 🖼️

▶️ 2.4K runs 📅 May 2024 ⚙️ Cog 0.8.0-beta11 🔗 GitHub

image-restoration image-upscaling

About

Upscale | Enhancer | Ultra-Resolution | Restoration |

Example Output

Prompt:

"masterpiece, best quality, highres, lora:more_details:0.5 lora:SDXLrender_v2.0:1"

Output

Performance Metrics

13.09s Prediction Time

47.86s Total Time

All Input Parameters

{
  "seed": 1337,
  "image": "https://replicate.delivery/pbxt/N5tDPCJFw7xAXnrnD1B66b1m30spvfbvY7qHEPb0nAjNFSHl/image.png",
  "prompt": "masterpiece, best quality, highres, <lora:more_details:0.5> <lora:SDXLrender_v2.0:1>",
  "dynamic": 6,
  "handfix": "disabled",
  "pattern": false,
  "sharpen": 0,
  "sd_model": "juggernaut_reborn.safetensors [338b85bc4f]",
  "scheduler": "DPM++ 3M SDE Karras",
  "creativity": 0.35,
  "lora_links": "",
  "downscaling": false,
  "resemblance": 0.6,
  "scale_factor": 2,
  "tiling_width": 112,
  "output_format": "png",
  "tiling_height": 144,
  "custom_sd_model": "",
  "negative_prompt": "(worst quality, low quality, normal quality:2) JuggernautNegative-neg",
  "num_inference_steps": 18,
  "downscaling_resolution": 768
}

Input Parameters

mask Type: string: Mask image to mark areas that should be preserved during upscaling
seed Type: integerDefault: 1337: Random seed. Leave blank to randomize the seed
image (required) Type: string: input image
prompt Type: stringDefault: masterpiece, best quality, highres, <lora:more_details:0.5> <lora:SDXLrender_v2.0:1>: Prompt
dynamic Type: numberDefault: 6Range: 1 - 50: HDR, try from 3 - 9
handfix Default: disabled: Use clarity to fix hands in the image
pattern Type: booleanDefault: false: Upscale a pattern with seamless tiling
sharpen Type: numberDefault: 0Range: 0 - 10: Sharpen the image after upscaling. The higher the value, the more sharpening is applied. 0 for no sharpening
sd_model Default: juggernaut_reborn.safetensors [338b85bc4f]: Stable Diffusion model checkpoint
scheduler Default: DPM++ 3M SDE Karras: scheduler
creativity Type: numberDefault: 0.35Range: 0 - 1: Creativity, try from 0.3 - 0.9
lora_links Type: stringDefault:: Link to a lora file you want to use in your upscaling. Multiple links possible, seperated by comma
downscaling Type: booleanDefault: false: Downscale the image before upscaling. Can improve quality and speed for images with high resolution but lower quality
resemblance Type: numberDefault: 0.6Range: 0 - 3: Resemblance, try from 0.3 - 1.6
scale_factor Type: numberDefault: 2: Scale factor
tiling_width Default: 112: Fractality, set lower tile width for a high Fractality
output_format Default: png: Format of the output images
tiling_height Default: 144: Fractality, set lower tile height for a high Fractality
custom_sd_model Type: stringDefault:
negative_prompt Type: stringDefault: (worst quality, low quality, normal quality:2) JuggernautNegative-neg: Negative Prompt
num_inference_steps Type: integerDefault: 18Range: 1 - 100: Number of denoising steps
downscaling_resolution Type: integerDefault: 768: Downscaling resolution

Output Schema

Output

Type: array • Items Type: string • Items Format: uri

Example Execution Logs

Running prediction
Upscaling with scale_factor:  2.0
[Tiled Diffusion] upscaling image with 4x-UltraSharp...
[Tiled Diffusion] ControlNet found, support is enabled.
2025-05-29 16:40:57,767 - ControlNet - [0;32mINFO[0m - unit_separate = False, style_align = False
2025-05-29 16:40:57,767 - ControlNet - [0;32mINFO[0m - Loading model from cache: control_v11f1e_sd15_tile
2025-05-29 16:40:57,783 - ControlNet - [0;32mINFO[0m - Using preprocessor: tile_resample
2025-05-29 16:40:57,783 - ControlNet - [0;32mINFO[0m - preprocessor resolution = 1536
2025-05-29 16:40:57,886 - ControlNet - [0;32mINFO[0m - ControlNet Hooked - Time = 0.12337040901184082
MultiDiffusion hooked into 'DPM++ 3M SDE Karras' sampler, Tile size: 144x112, Tile count: 4, Batch size: 4, Tile batches: 1 (ext: ContrlNet)
[Tiled VAE]: the input size is tiny and unnecessary to tile.
MultiDiffusion Sampling:   0%|          | 0/1 [00:00<?, ?it/s]
  0%|          | 0/7 [00:00<?, ?it/s][A
Total progress:   0%|          | 0/7 [00:00<?, ?it/s][A
 14%|█▍        | 1/7 [00:01<00:07,  1.30s/it][A
Total progress:  29%|██▊       | 2/7 [00:00<00:01,  3.97it/s][A
 29%|██▊       | 2/7 [00:01<00:04,  1.21it/s][A
Total progress:  43%|████▎     | 3/7 [00:00<00:01,  2.84it/s][A
 43%|████▎     | 3/7 [00:02<00:02,  1.48it/s][A
Total progress:  57%|█████▋    | 4/7 [00:01<00:01,  2.47it/s][A
 57%|█████▋    | 4/7 [00:02<00:01,  1.66it/s][A
Total progress:  71%|███████▏  | 5/7 [00:01<00:00,  2.29it/s][A
 71%|███████▏  | 5/7 [00:03<00:01,  1.77it/s][A
Total progress:  86%|████████▌ | 6/7 [00:02<00:00,  2.20it/s][A
 86%|████████▌ | 6/7 [00:03<00:00,  1.86it/s][A
100%|██████████| 7/7 [00:04<00:00,  1.91it/s][A
100%|██████████| 7/7 [00:04<00:00,  1.65it/s]
Total progress: 100%|██████████| 7/7 [00:02<00:00,  2.14it/s][A[Tiled VAE]: input_size: torch.Size([1, 4, 192, 192]), tile_size: 128, padding: 11
[Tiled VAE]: split to 2x2 = 4 tiles. Optimal tile size 96x96, original tile size 128x128
[Tiled VAE]: Fast mode enabled, estimating group norm parameters on 128 x 128 image
[Tiled VAE]: Executing Decoder Task Queue:   0%|          | 0/492 [00:00<?, ?it/s][A[A
[Tiled VAE]: Executing Decoder Task Queue:  25%|██▌       | 124/492 [00:00<00:00, 874.78it/s][A[A
[Tiled VAE]: Executing Decoder Task Queue:  50%|█████     | 247/492 [00:00<00:00, 980.35it/s][A[A
[Tiled VAE]: Executing Decoder Task Queue:  75%|███████▌  | 370/492 [00:00<00:00, 1021.20it/s][A[A
[Tiled VAE]: Executing Decoder Task Queue: 100%|██████████| 492/492 [00:00<00:00, 1064.73it/s]
[Tiled VAE]: Done in 1.116s, max VRAM alloc 5122.526 MB
Total progress: 100%|██████████| 7/7 [00:04<00:00,  2.14it/s][A
Total progress: 100%|██████████| 7/7 [00:04<00:00,  1.60it/s]
Prediction took 11.69 seconds

Version Details

Version ID: 641915cc4f4abefcdd361438162097266f3889d71aa90727d53b70ec3ed211cf
Version Created: May 29, 2025

Run on Replicate →