chenxwh / meissonic

Revitalizing Masked Generative Transformers for Efficient High-Resolution Text-to-Image Synthesis

Cold

Public
35 runs
L40S
GitHub
Weights
Paper
License

Run with an API

Playground API Examples README Versions

Input

prompt

string

Shift + Return to add a new line

a photo of an astronaut riding a horse on marsa photo of an astronaut riding a horse on mars

Input prompt

Default: "a photo of an astronaut riding a horse on mars"

negative_prompt

string

Shift + Return to add a new line

worst quality, low quality, low res, blurry, distortion, watermark, logo, signature, text, jpeg artifacts, signature, sketch, duplicate, ugly, identifying markworst quality, low quality, low res, blurry, distortion, watermark, logo, signature, text, jpeg artifacts, signature, sketch, duplicate, ugly, identifying mark

Specify things to not see in the output

Default: "worst quality, low quality, low res, blurry, distortion, watermark, logo, signature, text, jpeg artifacts, signature, sketch, duplicate, ugly, identifying mark"

num_inference_steps

integer

(minimum: 1, maximum: 100)

Number of denoising steps

Default: 64

guidance_scale

number

(minimum: 0, maximum: 20)

Scale for classifier-free guidance

Default: 9

seed

integer

Random seed. Leave blank to randomize the seed

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

npx create-replicate --model=chenxwh/meissonic

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

const output = await replicate.run(
  "chenxwh/meissonic:92c16966ef0dda80cf0afb11544371bfb2944025c47c4036615fcfd2f7515a21",
  {
    input: {
      prompt: "a photo of an astronaut riding a horse on mars",
      guidance_scale: 9,
      negative_prompt: "worst quality, low quality, low res, blurry, distortion, watermark, logo, signature, text, jpeg artifacts, signature, sketch, duplicate, ugly, identifying mark",
      num_inference_steps: 64
    }
  }
);
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 chenxwh/meissonic using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

output = replicate.run(
    "chenxwh/meissonic:92c16966ef0dda80cf0afb11544371bfb2944025c47c4036615fcfd2f7515a21",
    input={
        "prompt": "a photo of an astronaut riding a horse on mars",
        "guidance_scale": 9,
        "negative_prompt": "worst quality, low quality, low res, blurry, distortion, watermark, logo, signature, text, jpeg artifacts, signature, sketch, duplicate, ugly, identifying mark",
        "num_inference_steps": 64
    }
)
print(output)

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 chenxwh/meissonic 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": "92c16966ef0dda80cf0afb11544371bfb2944025c47c4036615fcfd2f7515a21",
    "input": {
      "prompt": "a photo of an astronaut riding a horse on mars",
      "guidance_scale": 9,
      "negative_prompt": "worst quality, low quality, low res, blurry, distortion, watermark, logo, signature, text, jpeg artifacts, signature, sketch, duplicate, ugly, identifying mark",
      "num_inference_steps": 64
    }
  }' \
  https://api.replicate.com/v1/predictions

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

Output

{
  "completed_at": "2024-10-20T19:31:33.131320Z",
  "created_at": "2024-10-20T19:28:43.630000Z",
  "data_removed": false,
  "error": null,
  "id": "xmnt06t0dsrgj0cjndkv7try1c",
  "input": {
    "prompt": "a photo of an astronaut riding a horse on mars",
    "guidance_scale": 9,
    "negative_prompt": "worst quality, low quality, low res, blurry, distortion, watermark, logo, signature, text, jpeg artifacts, signature, sketch, duplicate, ugly, identifying mark",
    "num_inference_steps": 64
  },
  "logs": "Using seed: 43882\n  0%|          | 0/64 [00:00<?, ?it/s]\n  2%|▏         | 1/64 [00:00<00:31,  2.02it/s]\n  3%|▎         | 2/64 [00:00<00:21,  2.94it/s]\n  5%|▍         | 3/64 [00:00<00:17,  3.43it/s]\n  6%|▋         | 4/64 [00:01<00:16,  3.72it/s]\n  8%|▊         | 5/64 [00:01<00:15,  3.90it/s]\n  9%|▉         | 6/64 [00:01<00:14,  4.02it/s]\n 11%|█         | 7/64 [00:01<00:13,  4.10it/s]\n 12%|█▎        | 8/64 [00:02<00:13,  4.16it/s]\n 14%|█▍        | 9/64 [00:02<00:13,  4.19it/s]\n 16%|█▌        | 10/64 [00:02<00:12,  4.22it/s]\n 17%|█▋        | 11/64 [00:02<00:12,  4.24it/s]\n 19%|█▉        | 12/64 [00:03<00:12,  4.25it/s]\n 20%|██        | 13/64 [00:03<00:11,  4.25it/s]\n 22%|██▏       | 14/64 [00:03<00:11,  4.18it/s]\n 23%|██▎       | 15/64 [00:03<00:11,  4.21it/s]\n 25%|██▌       | 16/64 [00:04<00:11,  4.23it/s]\n 27%|██▋       | 17/64 [00:04<00:11,  4.24it/s]\n 28%|██▊       | 18/64 [00:04<00:10,  4.25it/s]\n 30%|██▉       | 19/64 [00:04<00:10,  4.26it/s]\n 31%|███▏      | 20/64 [00:04<00:10,  4.26it/s]\n 33%|███▎      | 21/64 [00:05<00:10,  4.26it/s]\n 34%|███▍      | 22/64 [00:05<00:09,  4.26it/s]\n 36%|███▌      | 23/64 [00:05<00:09,  4.26it/s]\n 38%|███▊      | 24/64 [00:05<00:09,  4.26it/s]\n 39%|███▉      | 25/64 [00:06<00:09,  4.25it/s]\n 41%|████      | 26/64 [00:06<00:08,  4.26it/s]\n 42%|████▏     | 27/64 [00:06<00:08,  4.26it/s]\n 44%|████▍     | 28/64 [00:06<00:08,  4.26it/s]\n 45%|████▌     | 29/64 [00:07<00:08,  4.26it/s]\n 47%|████▋     | 30/64 [00:07<00:07,  4.26it/s]\n 48%|████▊     | 31/64 [00:07<00:09,  3.42it/s]\n 50%|█████     | 32/64 [00:08<00:09,  3.40it/s]\n 52%|█████▏    | 33/64 [00:08<00:08,  3.62it/s]\n 53%|█████▎    | 34/64 [00:08<00:07,  3.79it/s]\n 55%|█████▍    | 35/64 [00:08<00:07,  3.91it/s]\n 56%|█████▋    | 36/64 [00:08<00:06,  4.01it/s]\n 58%|█████▊    | 37/64 [00:09<00:06,  4.07it/s]\n 59%|█████▉    | 38/64 [00:09<00:06,  4.12it/s]\n 61%|██████    | 39/64 [00:09<00:06,  4.16it/s]\n 62%|██████▎   | 40/64 [00:09<00:05,  4.19it/s]\n 64%|██████▍   | 41/64 [00:10<00:05,  4.21it/s]\n 66%|██████▌   | 42/64 [00:10<00:05,  4.22it/s]\n 67%|██████▋   | 43/64 [00:10<00:04,  4.23it/s]\n 69%|██████▉   | 44/64 [00:10<00:04,  4.24it/s]\n 70%|███████   | 45/64 [00:11<00:04,  4.24it/s]\n 72%|███████▏  | 46/64 [00:11<00:04,  4.24it/s]\n 73%|███████▎  | 47/64 [00:11<00:04,  4.24it/s]\n 75%|███████▌  | 48/64 [00:11<00:03,  4.22it/s]\n 77%|███████▋  | 49/64 [00:12<00:03,  4.23it/s]\n 78%|███████▊  | 50/64 [00:12<00:03,  4.24it/s]\n 80%|███████▉  | 51/64 [00:12<00:03,  4.24it/s]\n 81%|████████▏ | 52/64 [00:12<00:02,  4.24it/s]\n 83%|████████▎ | 53/64 [00:12<00:02,  4.24it/s]\n 84%|████████▍ | 54/64 [00:13<00:02,  4.24it/s]\n 86%|████████▌ | 55/64 [00:13<00:02,  4.25it/s]\n 88%|████████▊ | 56/64 [00:13<00:01,  4.25it/s]\n 89%|████████▉ | 57/64 [00:13<00:01,  4.25it/s]\n 91%|█████████ | 58/64 [00:14<00:01,  4.25it/s]\n 92%|█████████▏| 59/64 [00:14<00:01,  4.24it/s]\n 94%|█████████▍| 60/64 [00:14<00:00,  4.24it/s]\n 95%|█████████▌| 61/64 [00:14<00:00,  4.24it/s]\n 97%|█████████▋| 62/64 [00:15<00:00,  4.23it/s]\n 98%|█████████▊| 63/64 [00:15<00:00,  4.24it/s]\n100%|██████████| 64/64 [00:15<00:00,  4.25it/s]\n100%|██████████| 64/64 [00:15<00:00,  4.11it/s]",
  "metrics": {
    "predict_time": 18.370107867,
    "total_time": 169.50132
  },
  "output": "https://replicate.delivery/pbxt/tWEHQdPsnuqPOZrcbbZmLDqYZTrFZIappz8k0Wgp0l1kMM6E/out.png",
  "started_at": "2024-10-20T19:31:14.761212Z",
  "status": "succeeded",
  "urls": {
    "get": "https://api.replicate.com/v1/predictions/xmnt06t0dsrgj0cjndkv7try1c",
    "cancel": "https://api.replicate.com/v1/predictions/xmnt06t0dsrgj0cjndkv7try1c/cancel"
  },
  "version": "92c16966ef0dda80cf0afb11544371bfb2944025c47c4036615fcfd2f7515a21"
}

Generated in

18.4 seconds

Tweak it ShareReport

Using seed: 43882
  0%|          | 0/64 [00:00<?, ?it/s]
  2%|▏         | 1/64 [00:00<00:31,  2.02it/s]
  3%|▎         | 2/64 [00:00<00:21,  2.94it/s]
  5%|▍         | 3/64 [00:00<00:17,  3.43it/s]
  6%|▋         | 4/64 [00:01<00:16,  3.72it/s]
  8%|▊         | 5/64 [00:01<00:15,  3.90it/s]
  9%|▉         | 6/64 [00:01<00:14,  4.02it/s]
 11%|█         | 7/64 [00:01<00:13,  4.10it/s]
 12%|█▎        | 8/64 [00:02<00:13,  4.16it/s]
 14%|█▍        | 9/64 [00:02<00:13,  4.19it/s]
 16%|█▌        | 10/64 [00:02<00:12,  4.22it/s]
 17%|█▋        | 11/64 [00:02<00:12,  4.24it/s]
 19%|█▉        | 12/64 [00:03<00:12,  4.25it/s]
 20%|██        | 13/64 [00:03<00:11,  4.25it/s]
 22%|██▏       | 14/64 [00:03<00:11,  4.18it/s]
 23%|██▎       | 15/64 [00:03<00:11,  4.21it/s]
 25%|██▌       | 16/64 [00:04<00:11,  4.23it/s]
 27%|██▋       | 17/64 [00:04<00:11,  4.24it/s]
 28%|██▊       | 18/64 [00:04<00:10,  4.25it/s]
 30%|██▉       | 19/64 [00:04<00:10,  4.26it/s]
 31%|███▏      | 20/64 [00:04<00:10,  4.26it/s]
 33%|███▎      | 21/64 [00:05<00:10,  4.26it/s]
 34%|███▍      | 22/64 [00:05<00:09,  4.26it/s]
 36%|███▌      | 23/64 [00:05<00:09,  4.26it/s]
 38%|███▊      | 24/64 [00:05<00:09,  4.26it/s]
 39%|███▉      | 25/64 [00:06<00:09,  4.25it/s]
 41%|████      | 26/64 [00:06<00:08,  4.26it/s]
 42%|████▏     | 27/64 [00:06<00:08,  4.26it/s]
 44%|████▍     | 28/64 [00:06<00:08,  4.26it/s]
 45%|████▌     | 29/64 [00:07<00:08,  4.26it/s]
 47%|████▋     | 30/64 [00:07<00:07,  4.26it/s]
 48%|████▊     | 31/64 [00:07<00:09,  3.42it/s]
 50%|█████     | 32/64 [00:08<00:09,  3.40it/s]
 52%|█████▏    | 33/64 [00:08<00:08,  3.62it/s]
 53%|█████▎    | 34/64 [00:08<00:07,  3.79it/s]
 55%|█████▍    | 35/64 [00:08<00:07,  3.91it/s]
 56%|█████▋    | 36/64 [00:08<00:06,  4.01it/s]
 58%|█████▊    | 37/64 [00:09<00:06,  4.07it/s]
 59%|█████▉    | 38/64 [00:09<00:06,  4.12it/s]
 61%|██████    | 39/64 [00:09<00:06,  4.16it/s]
 62%|██████▎   | 40/64 [00:09<00:05,  4.19it/s]
 64%|██████▍   | 41/64 [00:10<00:05,  4.21it/s]
 66%|██████▌   | 42/64 [00:10<00:05,  4.22it/s]
 67%|██████▋   | 43/64 [00:10<00:04,  4.23it/s]
 69%|██████▉   | 44/64 [00:10<00:04,  4.24it/s]
 70%|███████   | 45/64 [00:11<00:04,  4.24it/s]
 72%|███████▏  | 46/64 [00:11<00:04,  4.24it/s]
 73%|███████▎  | 47/64 [00:11<00:04,  4.24it/s]
 75%|███████▌  | 48/64 [00:11<00:03,  4.22it/s]
 77%|███████▋  | 49/64 [00:12<00:03,  4.23it/s]
 78%|███████▊  | 50/64 [00:12<00:03,  4.24it/s]
 80%|███████▉  | 51/64 [00:12<00:03,  4.24it/s]
 81%|████████▏ | 52/64 [00:12<00:02,  4.24it/s]
 83%|████████▎ | 53/64 [00:12<00:02,  4.24it/s]
 84%|████████▍ | 54/64 [00:13<00:02,  4.24it/s]
 86%|████████▌ | 55/64 [00:13<00:02,  4.25it/s]
 88%|████████▊ | 56/64 [00:13<00:01,  4.25it/s]
 89%|████████▉ | 57/64 [00:13<00:01,  4.25it/s]
 91%|█████████ | 58/64 [00:14<00:01,  4.25it/s]
 92%|█████████▏| 59/64 [00:14<00:01,  4.24it/s]
 94%|█████████▍| 60/64 [00:14<00:00,  4.24it/s]
 95%|█████████▌| 61/64 [00:14<00:00,  4.24it/s]
 97%|█████████▋| 62/64 [00:15<00:00,  4.23it/s]
 98%|█████████▊| 63/64 [00:15<00:00,  4.24it/s]
100%|██████████| 64/64 [00:15<00:00,  4.25it/s]
100%|██████████| 64/64 [00:15<00:00,  4.11it/s]

Examples

View more examples

Run time and cost

This model runs on Nvidia L40S GPU hardware. We don't yet have enough runs of this model to provide performance information.

Readme

Meissonic: Revitalizing Masked Generative Transformers for Efficient High-Resolution Text-to-Image Synthesis

Meissonic Banner

Meissonic Demos

🚀 Introduction

Meissonic is a non-autoregressive mask image modeling text-to-image synthesis model that can generate high-resolution images. It is designed to run on consumer graphics cards.

Key Features:
- 🖼️ High-resolution image generation (up to 1024x1024)
- 💻 Designed to run on consumer GPUs
- 🎨 Versatile applications: text-to-image, image-to-image

📚 Citation

If you find this work helpful, please consider citing:

@article{bai2024meissonic,
  title={Meissonic: Revitalizing Masked Generative Transformers for Efficient High-Resolution Text-to-Image Synthesis},
  author={Bai, Jinbin and Ye, Tian and Chow, Wei and Song, Enxin and Chen, Qing-Guo and Li, Xiangtai and Dong, Zhen and Zhu, Lei and Yan, Shuicheng},
  journal={arXiv preprint arXiv:2410.08261},
  year={2024}
}

🙏 Acknowledgements

We thank the community and contributors for their invaluable support in developing Meissonic. We thank apolinario@multimodal.art for making Meissonic Demo. We thank @NewGenAI and @飛鷹しずか@自称文系プログラマの勉強 for making YouTube toturials. We thank @pprp for making fp8 and int4 quantization. We thank @camenduru for making jupyter toturial