afiaka87/pyglide – Run with an API on Replicate

afiaka87 / pyglide

The predecessor to DALLE-2, GLIDE (filtered) with faster PRK/PLMS sampling.

Cold

Public
18.6K runs
T4
GitHub
Paper
License

Run with an API

Playground API Examples README Versions

Input

prompt

*string

Shift + Return to add a new line

detailed oil painting of a pembroke welsh corgidetailed oil painting of a pembroke welsh corgi

Text prompt to use. Keep it simple/literal and avoid using poetic language (unlike CLIP).

batch_size

integer

(minimum: 1, maximum: 8)

Batch size. Number of generations to predict

Default: 3

side_x

integer

Must be multiple of 8. Going above 64 is not recommended. Actual image will be 4x larger.

Default: 64

side_y

integer

Must be multiple of 8. Going above 64 is not recommended. Actual image will be 4x larger.

Default: 64

upsample_stage

boolean

If true, uses both the base and upsample models. If false, only the (finetuned) base model is used. This is useful for testing the upsampler, which is not finetuned.

Default: false

guidance_scale

number

Classifier-free guidance scale. Higher values move further away from unconditional outputs. Lower values move closer to unconditional outputs. Negative values guide towards semantically opposite classes. 4-16 is a reasonable range.

Default: 4

upsample_temp

string

Upsample temperature. Consider lowering to ~0.997 for blurry images with fewer artifacts.

Default: "0.998"

timestep_respacing

string

Number of timesteps to use for base model PLMS sampling. Usually don't need more than 50.

Default: "35"

sr_timestep_respacing

string

Number of timesteps to use for upsample model PLMS sampling. Usually don't need more than 20.

Default: "17"

seed

integer

Seed for reproducibility

Default: 0

Run this model in Node.js with one line of code:

npx create-replicate --model=afiaka87/pyglide

or set up a project from scratch

Install Replicate’s Node.js client library

npm install replicate

Set the REPLICATE_API_TOKEN environment variable

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Import and set up the client

import Replicate from "replicate";

const replicate = new Replicate({
  auth: process.env.REPLICATE_API_TOKEN,
});

Run afiaka87/pyglide using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

const output = await replicate.run(
  "afiaka87/pyglide:d0b38636cb245fbec3e5a7a26356fa55bc79ea1032f12e5ea821173f171be0b5",
  {
    input: {
      seed: 234,
      prompt: "detailed oil painting of a pembroke welsh corgi",
      side_x: 64,
      side_y: 64,
      batch_size: 3,
      upsample_temp: "1.0",
      guidance_scale: 4,
      upsample_stage: true,
      timestep_respacing: "35",
      sr_timestep_respacing: "27"
    }
  }
);
console.log(output);

To learn more, take a look at the guide on getting started with Node.js.

Install Replicate’s Python client library

pip install replicate

Set the REPLICATE_API_TOKEN environment variable

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Import the client

import replicate

Run afiaka87/pyglide using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

output = replicate.run(
    "afiaka87/pyglide:d0b38636cb245fbec3e5a7a26356fa55bc79ea1032f12e5ea821173f171be0b5",
    input={
        "seed": 234,
        "prompt": "detailed oil painting of a pembroke welsh corgi",
        "side_x": 64,
        "side_y": 64,
        "batch_size": 3,
        "upsample_temp": "1.0",
        "guidance_scale": 4,
        "upsample_stage": True,
        "timestep_respacing": "35",
        "sr_timestep_respacing": "27"
    }
)

# The afiaka87/pyglide model can stream output as it's running.
# The predict method returns an iterator, and you can iterate over that output.
for item in output:
    # https://replicate.com/afiaka87/pyglide/api#output-schema
    print(item)

To learn more, take a look at the guide on getting started with Python.

Set the REPLICATE_API_TOKEN environment variable

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Run afiaka87/pyglide using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

curl -s -X POST \
  -H "Authorization: Bearer $REPLICATE_API_TOKEN" \
  -H "Content-Type: application/json" \
  -H "Prefer: wait" \
  -d $'{
    "version": "d0b38636cb245fbec3e5a7a26356fa55bc79ea1032f12e5ea821173f171be0b5",
    "input": {
      "seed": 234,
      "prompt": "detailed oil painting of a pembroke welsh corgi",
      "side_x": 64,
      "side_y": 64,
      "batch_size": 3,
      "upsample_temp": "1.0",
      "guidance_scale": 4,
      "upsample_stage": true,
      "timestep_respacing": "35",
      "sr_timestep_respacing": "27"
    }
  }' \
  https://api.replicate.com/v1/predictions

To learn more, take a look at Replicate’s HTTP API reference docs.

Install Cog

brew install cog

If you don’t have Homebrew, there are other installation options available.

Pull and run afiaka87/pyglide using Cog (this will download the full model and run it in your local environment):

cog predict r8.im/afiaka87/pyglide@sha256:d0b38636cb245fbec3e5a7a26356fa55bc79ea1032f12e5ea821173f171be0b5 \
  -i 'seed=234' \
  -i 'prompt="detailed oil painting of a pembroke welsh corgi"' \
  -i 'side_x=64' \
  -i 'side_y=64' \
  -i 'batch_size=3' \
  -i 'upsample_temp="1.0"' \
  -i 'guidance_scale=4' \
  -i 'upsample_stage=true' \
  -i 'timestep_respacing="35"' \
  -i 'sr_timestep_respacing="27"'

To learn more, take a look at the Cog documentation.

Pull and run afiaka87/pyglide using Docker (this will download the full model and run it in your local environment):

docker run -d -p 5000:5000 --gpus=all r8.im/afiaka87/pyglide@sha256:d0b38636cb245fbec3e5a7a26356fa55bc79ea1032f12e5ea821173f171be0b5
curl -s -X POST \
  -H "Content-Type: application/json" \
  -d $'{
    "input": {
      "seed": 234,
      "prompt": "detailed oil painting of a pembroke welsh corgi",
      "side_x": 64,
      "side_y": 64,
      "batch_size": 3,
      "upsample_temp": "1.0",
      "guidance_scale": 4,
      "upsample_stage": true,
      "timestep_respacing": "35",
      "sr_timestep_respacing": "27"
    }
  }' \
  http://localhost:5000/predictions

Output

{
  "completed_at": "2022-02-16T21:13:19.703037Z",
  "created_at": "2022-02-16T21:12:43.381788Z",
  "data_removed": false,
  "error": null,
  "id": "fxjuxfbjnrgvppyr466ug2psre",
  "input": {
    "seed": 234,
    "prompt": "detailed oil painting of a pembroke welsh corgi",
    "side_x": "64",
    "side_y": "64",
    "batch_size": "3",
    "upsample_temp": "1.0",
    "guidance_scale": 4,
    "upsample_stage": true,
    "timestep_respacing": "35",
    "sr_timestep_respacing": "27"
  },
  "logs": "\n  0%|          | 0/33 [00:00<?, ?it/s]\n  3%|▎         | 1/33 [00:00<00:24,  1.31it/s]\n  6%|▌         | 2/33 [00:01<00:22,  1.38it/s]\n  9%|▉         | 3/33 [00:02<00:21,  1.40it/s]\n 12%|█▏        | 4/33 [00:02<00:14,  1.99it/s]\n 15%|█▌        | 5/33 [00:02<00:10,  2.59it/s]\n 18%|█▊        | 6/33 [00:02<00:08,  3.18it/s]\n 21%|██        | 7/33 [00:02<00:07,  3.71it/s]\n 24%|██▍       | 8/33 [00:03<00:05,  4.17it/s]\n 27%|██▋       | 9/33 [00:03<00:05,  4.54it/s]\n 30%|███       | 10/33 [00:03<00:04,  4.82it/s]\n 33%|███▎      | 11/33 [00:03<00:04,  5.06it/s]\n 36%|███▋      | 12/33 [00:03<00:04,  5.23it/s]\n 39%|███▉      | 13/33 [00:03<00:03,  5.36it/s]\n 42%|████▏     | 14/33 [00:04<00:03,  5.44it/s]\n 45%|████▌     | 15/33 [00:04<00:03,  5.50it/s]\n 48%|████▊     | 16/33 [00:04<00:03,  5.54it/s]\n 52%|█████▏    | 17/33 [00:04<00:02,  5.56it/s]\n 55%|█████▍    | 18/33 [00:04<00:02,  5.59it/s]\n 58%|█████▊    | 19/33 [00:04<00:02,  5.60it/s]\n 61%|██████    | 20/33 [00:05<00:02,  5.61it/s]\n 64%|██████▎   | 21/33 [00:05<00:02,  5.63it/s]\n 67%|██████▋   | 22/33 [00:05<00:01,  5.63it/s]\n 70%|██████▉   | 23/33 [00:05<00:01,  5.63it/s]\n 73%|███████▎  | 24/33 [00:05<00:01,  5.62it/s]\n 76%|███████▌  | 25/33 [00:06<00:01,  5.64it/s]\n 79%|███████▉  | 26/33 [00:06<00:01,  5.63it/s]\n 82%|████████▏ | 27/33 [00:06<00:01,  5.63it/s]\n 85%|████████▍ | 28/33 [00:06<00:00,  5.63it/s]\n 88%|████████▊ | 29/33 [00:06<00:00,  5.63it/s]\n 91%|█████████ | 30/33 [00:06<00:00,  5.63it/s]\n 94%|█████████▍| 31/33 [00:07<00:00,  5.62it/s]\n 97%|█████████▋| 32/33 [00:07<00:00,  5.63it/s]\n100%|██████████| 33/33 [00:07<00:00,  5.62it/s]\n100%|██████████| 33/33 [00:07<00:00,  4.41it/s]\n\n  0%|          | 0/25 [00:00<?, ?it/s]\n  4%|▍         | 1/25 [00:01<00:45,  1.91s/it]\n  8%|▊         | 2/25 [00:03<00:43,  1.91s/it]\n 12%|█▏        | 3/25 [00:05<00:42,  1.91s/it]\n 16%|█▌        | 4/25 [00:06<00:28,  1.35s/it]\n 20%|██        | 5/25 [00:06<00:20,  1.03s/it]\n 24%|██▍       | 6/25 [00:07<00:16,  1.18it/s]\n 28%|██▊       | 7/25 [00:07<00:13,  1.38it/s]\n 32%|███▏      | 8/25 [00:08<00:10,  1.55it/s]\n 36%|███▌      | 9/25 [00:08<00:09,  1.69it/s]\n 40%|████      | 10/25 [00:09<00:08,  1.79it/s]\n 44%|████▍     | 11/25 [00:09<00:07,  1.87it/s]\n 48%|████▊     | 12/25 [00:10<00:06,  1.93it/s]\n 52%|█████▏    | 13/25 [00:10<00:06,  1.98it/s]\n 56%|█████▌    | 14/25 [00:11<00:05,  2.01it/s]\n 60%|██████    | 15/25 [00:11<00:04,  2.03it/s]\n 64%|██████▍   | 16/25 [00:11<00:04,  2.05it/s]\n 68%|██████▊   | 17/25 [00:12<00:03,  2.06it/s]\n 72%|███████▏  | 18/25 [00:12<00:03,  2.06it/s]\n 76%|███████▌  | 19/25 [00:13<00:02,  2.06it/s]\n 80%|████████  | 20/25 [00:13<00:02,  2.07it/s]\n 84%|████████▍ | 21/25 [00:14<00:01,  2.07it/s]\n 88%|████████▊ | 22/25 [00:14<00:01,  2.07it/s]\n 92%|█████████▏| 23/25 [00:15<00:00,  2.07it/s]\n 96%|█████████▌| 24/25 [00:15<00:00,  2.07it/s]\n100%|██████████| 25/25 [00:16<00:00,  2.07it/s]\n100%|██████████| 25/25 [00:16<00:00,  1.53it/s]",
  "metrics": {
    "predict_time": 36.103013,
    "total_time": 36.321249
  },
  "output": [
    {
      "file": "https://replicate.delivery/mgxm/fcddf21c-6531-4295-8a29-813674150d98/upsample_predictions.png"
    }
  ],
  "started_at": "2022-02-16T21:12:43.600024Z",
  "status": "succeeded",
  "urls": {
    "get": "https://api.replicate.com/v1/predictions/fxjuxfbjnrgvppyr466ug2psre",
    "cancel": "https://api.replicate.com/v1/predictions/fxjuxfbjnrgvppyr466ug2psre/cancel"
  },
  "version": "9218d6e9a0f5f147f53f31ab6a2d562e5e88a3dab5c2f74fce9e067b868b9ca4"
}

Generated in

36.1 seconds

Tweak itReport


  0%|          | 0/33 [00:00<?, ?it/s]
  3%|▎         | 1/33 [00:00<00:24,  1.31it/s]
  6%|▌         | 2/33 [00:01<00:22,  1.38it/s]
  9%|▉         | 3/33 [00:02<00:21,  1.40it/s]
 12%|█▏        | 4/33 [00:02<00:14,  1.99it/s]
 15%|█▌        | 5/33 [00:02<00:10,  2.59it/s]
 18%|█▊        | 6/33 [00:02<00:08,  3.18it/s]
 21%|██        | 7/33 [00:02<00:07,  3.71it/s]
 24%|██▍       | 8/33 [00:03<00:05,  4.17it/s]
 27%|██▋       | 9/33 [00:03<00:05,  4.54it/s]
 30%|███       | 10/33 [00:03<00:04,  4.82it/s]
 33%|███▎      | 11/33 [00:03<00:04,  5.06it/s]
 36%|███▋      | 12/33 [00:03<00:04,  5.23it/s]
 39%|███▉      | 13/33 [00:03<00:03,  5.36it/s]
 42%|████▏     | 14/33 [00:04<00:03,  5.44it/s]
 45%|████▌     | 15/33 [00:04<00:03,  5.50it/s]
 48%|████▊     | 16/33 [00:04<00:03,  5.54it/s]
 52%|█████▏    | 17/33 [00:04<00:02,  5.56it/s]
 55%|█████▍    | 18/33 [00:04<00:02,  5.59it/s]
 58%|█████▊    | 19/33 [00:04<00:02,  5.60it/s]
 61%|██████    | 20/33 [00:05<00:02,  5.61it/s]
 64%|██████▎   | 21/33 [00:05<00:02,  5.63it/s]
 67%|██████▋   | 22/33 [00:05<00:01,  5.63it/s]
 70%|██████▉   | 23/33 [00:05<00:01,  5.63it/s]
 73%|███████▎  | 24/33 [00:05<00:01,  5.62it/s]
 76%|███████▌  | 25/33 [00:06<00:01,  5.64it/s]
 79%|███████▉  | 26/33 [00:06<00:01,  5.63it/s]
 82%|████████▏ | 27/33 [00:06<00:01,  5.63it/s]
 85%|████████▍ | 28/33 [00:06<00:00,  5.63it/s]
 88%|████████▊ | 29/33 [00:06<00:00,  5.63it/s]
 91%|█████████ | 30/33 [00:06<00:00,  5.63it/s]
 94%|█████████▍| 31/33 [00:07<00:00,  5.62it/s]
 97%|█████████▋| 32/33 [00:07<00:00,  5.63it/s]
100%|██████████| 33/33 [00:07<00:00,  5.62it/s]
100%|██████████| 33/33 [00:07<00:00,  4.41it/s]

  0%|          | 0/25 [00:00<?, ?it/s]
  4%|▍         | 1/25 [00:01<00:45,  1.91s/it]
  8%|▊         | 2/25 [00:03<00:43,  1.91s/it]
 12%|█▏        | 3/25 [00:05<00:42,  1.91s/it]
 16%|█▌        | 4/25 [00:06<00:28,  1.35s/it]
 20%|██        | 5/25 [00:06<00:20,  1.03s/it]
 24%|██▍       | 6/25 [00:07<00:16,  1.18it/s]
 28%|██▊       | 7/25 [00:07<00:13,  1.38it/s]
 32%|███▏      | 8/25 [00:08<00:10,  1.55it/s]
 36%|███▌      | 9/25 [00:08<00:09,  1.69it/s]
 40%|████      | 10/25 [00:09<00:08,  1.79it/s]
 44%|████▍     | 11/25 [00:09<00:07,  1.87it/s]
 48%|████▊     | 12/25 [00:10<00:06,  1.93it/s]
 52%|█████▏    | 13/25 [00:10<00:06,  1.98it/s]
 56%|█████▌    | 14/25 [00:11<00:05,  2.01it/s]
 60%|██████    | 15/25 [00:11<00:04,  2.03it/s]
 64%|██████▍   | 16/25 [00:11<00:04,  2.05it/s]
 68%|██████▊   | 17/25 [00:12<00:03,  2.06it/s]
 72%|███████▏  | 18/25 [00:12<00:03,  2.06it/s]
 76%|███████▌  | 19/25 [00:13<00:02,  2.06it/s]
 80%|████████  | 20/25 [00:13<00:02,  2.07it/s]
 84%|████████▍ | 21/25 [00:14<00:01,  2.07it/s]
 88%|████████▊ | 22/25 [00:14<00:01,  2.07it/s]
 92%|█████████▏| 23/25 [00:15<00:00,  2.07it/s]
 96%|█████████▌| 24/25 [00:15<00:00,  2.07it/s]
100%|██████████| 25/25 [00:16<00:00,  2.07it/s]
100%|██████████| 25/25 [00:16<00:00,  1.53it/s]

This example was created by a different version, afiaka87/pyglide:9218d6e9.

Examples

View more examples

Run time and cost

This model costs approximately $0.042 to run on Replicate, or 23 runs per $1, but this varies depending on your inputs. It is also open source and you can run it on your own computer with Docker.

This model runs on Nvidia T4 GPU hardware. Predictions typically complete within 4 minutes. The predict time for this model varies significantly based on the inputs.

Readme

pyglide

OpenAI filtered all humans out of the training set for this model. This means a lot of prompts simply can’t work. Still a fun model though and incredibly fast!

Thanks to “Pseudo Numerical Methods for Diffusion Models on Manifolds”, results are now even faster and more accurate. Credit goes to Katherine Crowson for implementing this for GLIDE specifically (https://github.com/crowsonkb/glide-text2im).

an analog clock hanging on a blue wall

a lonely robot in the middle of the field a lonely robot on hanging out on a cliff

a goose made of paper. paper goose. a goose rendered in minecraft. minecraft goose.

Installation

First clone this repository:

git clone https://github.com/afiaka87/text-glided-diffusion.git
cd text-glided-diffusion

You also need to install glide-text2im from openai’s repository.

python3 -m venv .venv
source .venv/bin/activate
(.venv) python -m pip install -r requirements.txt
(.venv) git clone https://github.com/openai/glide-text2im.git
(.venv) cd glide-text2im/
(.venv) python -m pip install -e .
(.venv) cd ../

Usage

time python tgd.py --prompt "the beach at sunset"
Selected device: cuda:0.
1. Creating model and diffusion.
1. Done.
2. Running base GLIDE text2im model.
2. Base model generations complete. Check glide_outputs/base/the_beach_at_sunset/the_beach_at_sunset.png for generations.
3. Loading GLIDE upsampling diffusion model.
3. Done.
4. Running GLIDE upsampling from 64x64 to 256x256.
4. Done. Check glide_outputs/sr/the_beach_at_sunset/the_beach_at_sunset.png for generations.

real    1m4.775s
user    1m9.648s
sys     0m8.894s

Detailed Usage

usage: tgd.py [-h] --prompt PROMPT [--batch_size BATCH_SIZE] [--guidance_scale GUIDANCE_SCALE] [--base_x BASE_X] [--base_y BASE_Y] [--respace RESPACE] [--prefix PREFIX] [--upsample_temp UPSAMPLE_TEMP]

optional arguments:
  -h, --help            show this help message and exit
  --prompt PROMPT       a caption to visualize
  --batch_size BATCH_SIZE
  --guidance_scale GUIDANCE_SCALE
  --base_x BASE_X       width of base gen. has to be multiple of 16
  --base_y BASE_Y       width of base gen. has to be multiple of 16
  --respace RESPACE     Number of timesteps to use for generation. Lower is faster but less accurate.
  --prefix PREFIX       Output dir for generations. Will be created if it doesn't exist with subfolders for base and upsampled.
  --upsample_temp       0.0 to 1.0. 1.0 can introduce artifacts, lower can introduce blurriness.