Bunimation

Just a guy on a mission to build the best uncensored AI animation platform. Obviously depicting adults only.

[Bunimation @BunimationAi]

Open Beta coming soon to https://t.co/XSNdlKpVob. Follow for now. #bunimation #nsfw #civitai

[Bunimation @BunimationAi]

@satellitedown You mentioned its only for 5090 right now. Could we get this running on an RTX 6000 with minimal changes? You'd be able to run an entire office or a mega-swarm off one gpu with these speeds. 🤤

[Bunimation @BunimationAi]

@JonathanEngle @TungaBayrak @DesignArena I went to the website and was impressed, then annoyed again because API pricing is gated behind a phone call 🤣

[Bunimation @BunimationAi]

Generate images and make full comic books in the browser. Let the community enjoy your stories or Export to all the most common publishing formats. #bunimation releasing to open beta soon. https://t.co/XSNdlKqtdJ

[Bunimation @BunimationAi]

@Akhila_988 You don't know what you've just done! This is amazing. On behalf of everyone else who is too excited to say anything, Thank you!

[Bunimation @BunimationAi]

@Perpetualmaniac If a bot can create a swarm of millions, why can’t I? Oh I forgot, I need the compute and storage to replicate.
They can’t do it either.

[Bunimation @BunimationAi]

@isidentical You stand on the shoulders of the open-source community and shit talk a team releasing the weights before you did?!

They probably don’t have even a quarter the compute you gave to this process.

[Bunimation @BunimationAi]

@loktar00 what harness did you use? I notice GLM 5.3 Flash operates better in some harnesses than others. People have had great steering in Codex.

[Bunimation @BunimationAi]

@SFourdrinier @xieenze_jr The point where you make the snapshot, the cpu and memory allocation make the biggest difference.

[Bunimation @BunimationAi]

@alwayspriyesh Me and my friends are running around on your server lol. I want to build something with you hahaha

[Bunimation @BunimationAi]

You can avoid warmups by using Memory Snapshots in Modal. If you are calling the inference over Modal API (I recommend this) then you can prewarm the instance with a very small workflow call saved in your python script. When you start your container it will load linux, all the dependancies, run comfy, and start the workflow. Once the workflow completes, then save the GPU and RAM snapshot from that point. Make sure the container is configured to go to zero.

Every call after will be a 2 second warm up. You can build yourself a small workflow controller app using gradio (or a local hosted site) and program everything you want to control on the workflow. Submitting the workflow to render can call modal over terminal or api then fetch the result from the modal storage or whatever remote storage you want to have your python process send the completed file to.

Also, force request 64GB of RAM and 6 cores. Dont let modal run on default because they give you basically 1 shared core by default and the ram is elastic. This slows down inference by a noticeable amount.

[Bunimation @BunimationAi]

I built an uncensored AI animation platform. I'll post screenshot of the generator dashboard soon.

Its a full tool:

Image Generation
Naughty Banana (Image Edit by prompt)
Video Generation (Bunny VX)
Audio Generation (text to audio and video to audio)
Comic Generator (Multi-panel styles and word bubbles)
Community social system
Content rating system
Groups (coming soon)

The site is about to come out of beta. I'll post when its ready for sign-up. #AIart️️️️️️️️️️�����️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️ #nsfw #AI