Replicate - GenAI, LLM, 14 models Plugin for Bubble, by Panja

[Vincent_0x20100]

Hello everyone !

I’m Vincent

I have been using Bubble for few years now and I always made private plugin to meet my needs.
But today I’m glad and proud to present you my very first plugin available on the Bubble Marketplace !

Since AI is growing fast I have been using Replicate to find cool model to use with Bubble.
To make it easier between my apps I created a plugin compiling multiple model for an easy use.

In this plugin you will find 14 different models :
• Stable Diffusion 1.5 (text-to-image)
• SDXL (text-to-image, image-to-image, image-inpainting)
• Llama 2 70b (LLM, chat like GPT)
• GFPGAN (image upscaler & face restoration)
• blip (read, caption or answer a question about an image)
• Codeformer (another image upscaler & face restoration)
• Faceswap
• Whisper (transcription & subtitles)
• Wav2Lips (synchronisation between an audio and the lips of a person in a video)
• Deoldify (restore and color old photo)
• RemoveBG (Remove the background of an image)
• AudioGen (Generate audio from a prompt)
• Llama 2 13b (LLM, chat like GPT, supports streaming)
• Llava 13b (LLM with vision, chat like GPT, supports streaming)

Llama 2 13b and Llava 13b both support streaming !
It means that you can start to show the output before it completely finishes.

Llama 2 13b is a LLM, kinda like ChatGPT,
and Llava is also a LLM but with vision !

There is a special Element to enable the streaming.

All models support Webhooks so it doesn’t block the workflow.
For this to work you’ll need to create backend API workflows.

If you are new to backend API workflows you can find many tutorials on Youtube about it!

If the webhook field is let blank the workflow will be blocked until the end of the generation, however it timeout after 30s. (Image generation usually takes longer and won’t work this way).

There is a demo here with limited credit please try one or two generations to let everyone try the plugin : https://vima-plugins-demo.bubbleapps.io/version-test/replicate

Here is the page of the plugin : Replicate - GenAI, LLM, 9 models Plugin | Bubble

Tell me what you think, and if you need any help to set it up !

Thank you

[Vincent_0x20100]

[New Model Added]
Just added Instant-ID model in the plugin
It works almost like SDXL generation

It allows you to create avatars from just one picture
It also optionally integrates LCM for fast output

“Close-up photography, A Jedi with red saber, beautiful landscape, photo grain, fujifilm, 8K, ultra-realistic”

Now the plugin has 15 models !!!

Replicate - GenAI, LLM, 15 models Plugin for Bubble, by Panja

[Vincent_0x20100]

An example of use for Wav2Lips :

1/ First you need in your database at least a Datatype as follow :

the “jobid” field will allow us to find and complete the generation with our backend workflow (webhook)

2/ Now add a form on your page, you will need 1 file upload for each :

• the original video
• the new audio for the video

Then create a workflow on click on the button

3/ In your workflow you need to add the “Replicate - Wav2Lip - LipSync” action

And fill the fields :

don’t forget “https:” if you get the URL of the file from the bubble file uploaders
You need to use a webhook for this model, the generation is long
Fill the Webhook URL field with your backend workflow URL

4/ The 2nd action of your workflow is to create a new Wav2Lip in your database with the Prediction ID of the previous action

5/ Finally go to your backend workflow and add an action to change your Wav2Lip object in the database

You need to find your Wav2Lip object using the ID of the prediction.

you need basic Bubble knowledge to initiate the backend workflow in order for the plugin to work, and you’ll need a paying plan of Bubble.

[rccanlas]

Can replicate handle multiple concurrent requests?
Rate limits for commercial app