afiaka87/retrieval-augmented-diffusion 🔢📝🖼️✓❓ → ❓

▶️ 38.4K runs 📅 Aug 2022 ⚙️ Cog 0.3.13 🔗 GitHub 📄 Paper ⚖️ License
image-to-image retrieval-augmented text-to-image

About

Generate 768px images from text using CompVis `retrieval-augmented-diffusion`

Example Output

Output

[object Object]

Performance Metrics

5.76s Prediction Time
5.94s Total Time
All Input Parameters
{
  "seed": -1,
  "scale": 5,
  "steps": 50,
  "width": 768,
  "height": 768,
  "prompts": "an astronaut riding a horse in photorealistic style, digital art",
  "use_database": true,
  "database_name": "laion-aesthetic",
  "num_database_results": "20"
}
Input Parameters
seed Type: integerDefault: -1
Seed for the random number generator. Set to -1 to use a random seed.
scale Type: numberDefault: 5
Classifier-free unconditional scale for the sampler.
steps Type: integerDefault: 50
How many steps to run the model for. Using more will make generation take longer. 50 tends to work well.
width Type: integerDefault: 768
Desired width of generated images. Values beside 768 are not supported, likely to cause artifacts.
height Type: integerDefault: 768
Desired width of generated images. Values beside 768 are likely to cause zooming issues.
prompts Type: stringDefault:
(batched) Use up to 8 prompts by separating with a `|` character.
ddim_eta Type: numberDefault: 0
The eta parameter for ddim sampling.
image_prompt Type: string
(overrides `prompts`) Use an image as the prompt to generate variations of an existing image.
use_database Type: booleanDefault: true
Whether to use the database for the semantic search.
database_name Default: laion-aesthetic
Which database to use for the semantic search. Different databases have different capabilities.
ddim_sampling Type: booleanDefault: false
Use ddim sampling instead of the faster plms sampling.
negative_prompt Type: stringDefault:
(experimental) Use this prompt as a negative prompt for the sampler.
num_database_results Type: integerDefault: 10Range: 1 - 20
The number of search results from the retrieval backend to guide the generation with.
number_of_variations Type: integerDefault: 4Range: 1 - 8
Number of variations to generate when using only one `text_prompt`, or an `image_prompt`.
Output Schema

Output

Type: array

Example Execution Logs
Prompts: ['an astronaut riding a horse in photorealistic style, digital art']
Using random seed 4267824009
Seed: -1
CLIP Text Embed: torch.Size([1, 1, 768])
Using plms sampling
Data shape for PLMS sampling is (1, 16, 48, 48)
Running PLMS Sampling with 50 timesteps

PLMS Sampler:   0%|          | 0/50 [00:00<?, ?it/s]
PLMS Sampler:   2%|▏         | 1/50 [00:00<00:08,  6.06it/s]
PLMS Sampler:   6%|▌         | 3/50 [00:00<00:04,  9.92it/s]
PLMS Sampler:  10%|█         | 5/50 [00:00<00:03, 11.42it/s]
PLMS Sampler:  14%|█▍        | 7/50 [00:00<00:03, 12.13it/s]
PLMS Sampler:  18%|█▊        | 9/50 [00:00<00:03, 12.59it/s]
PLMS Sampler:  22%|██▏       | 11/50 [00:00<00:03, 12.90it/s]
PLMS Sampler:  26%|██▌       | 13/50 [00:01<00:02, 12.42it/s]
PLMS Sampler:  30%|███       | 15/50 [00:01<00:02, 12.21it/s]
PLMS Sampler:  34%|███▍      | 17/50 [00:01<00:02, 12.24it/s]
PLMS Sampler:  38%|███▊      | 19/50 [00:01<00:02, 11.95it/s]
PLMS Sampler:  42%|████▏     | 21/50 [00:01<00:02, 11.68it/s]
PLMS Sampler:  46%|████▌     | 23/50 [00:01<00:02, 11.48it/s]
PLMS Sampler:  50%|█████     | 25/50 [00:02<00:02, 11.36it/s]
PLMS Sampler:  54%|█████▍    | 27/50 [00:02<00:02, 11.37it/s]
PLMS Sampler:  58%|█████▊    | 29/50 [00:02<00:01, 11.60it/s]
PLMS Sampler:  62%|██████▏   | 31/50 [00:02<00:01, 11.38it/s]
PLMS Sampler:  66%|██████▌   | 33/50 [00:02<00:01, 11.67it/s]
PLMS Sampler:  70%|███████   | 35/50 [00:02<00:01, 11.85it/s]
PLMS Sampler:  74%|███████▍  | 37/50 [00:03<00:01, 12.09it/s]
PLMS Sampler:  78%|███████▊  | 39/50 [00:03<00:00, 12.41it/s]
PLMS Sampler:  82%|████████▏ | 41/50 [00:03<00:00, 12.70it/s]
PLMS Sampler:  86%|████████▌ | 43/50 [00:03<00:00, 12.52it/s]
PLMS Sampler:  90%|█████████ | 45/50 [00:03<00:00, 12.65it/s]
PLMS Sampler:  94%|█████████▍| 47/50 [00:03<00:00, 12.83it/s]
PLMS Sampler:  98%|█████████▊| 49/50 [00:04<00:00, 12.99it/s]
PLMS Sampler: 100%|██████████| 50/50 [00:04<00:00, 12.07it/s]
Version Details
Version ID
ac7878dfd12cb445115fb250a79faf3c3028e6cfa6c94788a7c9c53b0ce5898e
Version Created
August 20, 2022
Run on Replicate →