Sergei Belousov

AI can write apps and tests. But who actually opens the app and taps the buttons? I built a QA agent that tests UI promises, not code. It taps through the app on a simulator, reads screenshots, and catches bugs unit tests can't see. Blogpost: bes-dev.com/posts/teaching…

Open-sourcing my local alternative to @rork. Been using it to ship production apps. Etnamute — AI mobile developer on top of @claude_code . Idea → interview → market research → build with QA → App Store deploy. No framework, no extra subscription. github.com/bes-dev/etnamu…

Finally scratched an old itch — AI sticker packs from your photos. Works with Telegram/WhatsApp. 👉 apps.apple.com/us/app/stckzy-…

@Eliasfiz Russian

We're releasing Orpheus' most requested feature soon: 🌎Multi-language capabilities🌎! Drop your language requests below and we'll add them to the model!

I created gpt-scraper to automate data extraction from sites with unique structures — just describe what data you need, and the AI-powered library writes the scraper code for you. With the new o1-mini model, it’s faster and more accurate than ever! github.com/bes-dev/gpt-sc…

My first contribution to open-source protein design: a new way to combine ProteinMPNN and AlphaFold for designing sequences for a given structure colab.research.google.com/github/sokrypt… More context and info below 👇

@meta__artem We experimented with onnxruntime (github.com/bes-dev/stable…), but this backend has a lot of limitations, and I don't recommend to use it

@bes_dev Hi, can you explain to me how I can use the onnx version of stable diffusion. I see several files on hugging face and it confuses me, in onnx runtime tutorials we load one file with the model.

The new version of our engine for Stable Diffusion optimized to CPU is ready. Now it supports Image-To-Image and Inpainting too. Also, we release the initial version of pretty GUI. Join to the future of media art here: github.com/bes-dev/stable…… #art #deeplearning

@akiyamasho_dev It was a part of our first results in optimization of the stable diffusion to CPU: github.com/bes-dev/stable…

just discovered an ONNX version of stable-diffusion from @bes_dev !! huggingface.co/bes-dev/stable… it’s only been days and stable-diffusion demos, tutorials, tools, etc. have been mutiplying like mad. OSS is really something.

We published our early results to run state-of-the-art Text-To-Image models on consumer CPU. You can found project here: github.com/bes-dev/stable… #syntheticmedai #stablediffusion #ai #openvino

@JackdeS11 Nice work 💪

I uploaded a simple sample using MobileStyleGAN on iOS to GitHub. github.com/john-rocky/Cor… The Converted CoreML model have been uploaded to CoreML-Models. github.com/john-rocky/Cor…

Our new library for text-driven bounding box analysis was published. Also we provide examples to integrate it with popular object detector like YOLOV5 and MaskRCNN. github.com/bes-dev/pytorc… #OpenSource #Python #DeepLearning

Our new tiny zero-shot text-driven object detector named ClipRCNN was released: github.com/bes-dev/pytorc… #opensource #deeplearning #computervision

Our new library to make CLIP guided loss for Text-To-Image generation was published on github: github.com/bes-dev/pytorc… #opensource #python #deeplearning

Our new library to make differentiable Haar Wavelets Transform was published on github: github.com/bes-dev/haar_p… #OpenSource #Python #DeepLearning #PyTorch

Just a one-weekend project 🤗 We propose a new architecture of theDecoder network (based on VQVAE's decoder) for DALL-E that allows to generate images 512x512 instead of 256x256 as in baseline DALL-E model: github.com/bes-dev/vqvae_… #deeplearning #dalle #gans #opensource

@ak92501 @julien_c @huggingface @Gradio Great news! And also, today was released a new version of MobileStyleGAN for FFHQ with almost x2 quality improvement (FID decreased from 12.38 to 7.75) twitter.com/bes_dev/status…

MobileStyleGAN coming to @huggingface Spaces using @Gradio

The new version of our library for fast image generation of people who do not exist was released. x2 Better quality of generated images is here (= python library: github.com/bes-dev/random… training tool: github.com/bes-dev/Mobile… demo video: youtube.com/watch?v=Rewpnz…

Now, high-resolution style-based generative models run on mobile devices! On-device demo of our MobileStyleGAN is here: youtube.com/watch?v=CBebSI… Follow the project: github.com/bes-dev/Mobile…