Radamés Ajna @radamar Twitter profile

Last Seen Profiles

@Almisehal

@GorglessVL

@ChantelJeffries

@DianeHall120182

@ZubayraZubeyre

@kgeoffray

@redspatulaio

@MrJustice38377

@soyunargento

@Rara_122147

@fillooooooo

@WillemHazenberg

@christiancashot

@AdelJamal528522

@aasf44

@rogertugas

@borofergie

@HarryBee_Yhu

@WINkorg777

@VictoriaVa66900

@RasharkinGAC

@Mad_Hildebrandt

@iamliamlove

@non_tyoko_love

@Richard_Scorer

@green_ninjaa

@M_Ivenkov

@afghan_witness

@Gendolpf_ru

@RailUKForums

@ChackingkingM

@NotARealCourt

@BregmanMarina

@BluesPolski

@ShiseidoParlour

@Smoltog

Radamés Ajna

@radamar

11 months

The source code for DragGAN has been released! Huge thanks to community contributor LeoXing1996 for the @gradio demo try it now:

23

749

3K

Radamés Ajna

@radamar

11 months

Every day, there's a new exciting paper demo on the hub, it's hard to keep up! New AI video-to-video translation

31

492

3K

Radamés Ajna

@radamar

6 months

Thanks to the Latent Consistency Model (LCM), we're nearing real-time image diffusion. I've made a simple MJPEG server for generation stream using diffusers img2img pipeline. It's really fun to play with it. Can't wait for the ControlNet version. try it:

43

345

2K

Radamés Ajna

@radamar

6 months

I've made it possible to share you're screen on the real-time Latent Consistency Model demo, thanks to Screen Capture Web API! No need for a custom drawing tool, use your favorite one 🤩. Plus, enjoy musk rickrolling demo:

24

192

1K

Radamés Ajna

@radamar

11 months

Create sentence embeddings in your browser with transformers.js! My guide walks you through generating embeddings and applying UMAP dimension reduction + more - all in JavaScript, no server needed

15

207

1K

Radamés Ajna

@radamar

11 months

Great news! The @huggingface hub now has the first QR code AI art generator. You only need the QR Code content and a text-to-image prompt idea, or you can upload your image! Check it out!

34

172

908

Radamés Ajna

@radamar

6 months

Just added ControlNet Canny to the near real-time Latent Consistency Model demo. It's much better than just img2img! Any updates to the UI parameters and prompts happen instantly. Video here at 2x speed Demo:

33

171

906

Radamés Ajna

@radamar

4 months

InstantID works with ControlNet Pose and LCM, and it might actually work with any ControlNet. The trade-off of using multiple ControlNets results in a slight loss of facial detail.

15

153

811

Radamés Ajna

@radamar

5 months

In a couple of weeks, we went from LCM LoRa getting about ~5fps to now ~17fps with the latest SD-Turbo distilled model. Thanks @StabilityAI the quality with SD-Turbo is incredible! Video is at normal speed. see Musk and Gates with curly hair 😅 demo:

17

135

811

Radamés Ajna

@radamar

10 months

Here is the DragGAN Face Inversion @Gradio demo. You can upload your image and experiment with some wild edits. Please be patient, as the inversion training process takes approximately 2 minutes 😞

6

163

753

Radamés Ajna

@radamar

5 months

Here's the demo "Enhance This"! It's a surreal image magnifier that creates a high-res version by imagining new details, using the SDXL base model. Thanks to @RuoyiDu 's DemoFusion research. It takes a ~minute to generate a 2024x2024 image.

17

130

655

Radamés Ajna

@radamar

6 months

I've also put together a text-to-image version so you can get a sense of what it's like to live-prompt a model, thanks to Latent Consistency Model speed. btw I'm not good at prompting. Video is at 4x, as my typing isn't fast. here:

18

131

640

Radamés Ajna

@radamar

1 year

Check out a new ControlNet face model on the hub by @JCatrambone & @DarthMarkov , trained on the LAION-Face dataset it works with multiple faces. I've updated the live conditioning @gradio component, try the official demo here

13

127

626

Radamés Ajna

@radamar

1 year

I made a custom @gradio component to generate face landmarks conditioning image for the ControlNet Uncanny Faces model try the live demo here

15

111

624

Radamés Ajna

@radamar

1 year

Our community member, George Fe, trained ControlNet with stable diffusion using face landmarks conditioning! 🤯 Give it a try

22

116

577

Radamés Ajna

@radamar

5 months

Quick test with SDXL Turbo, another amazing, super-fast diffusion model. It works right out of the box with @diffuserslib unofficial demo txt2img and img2img:

13

113

559

Radamés Ajna

@radamar

8 months

Try the Meta Segment Anything Model (SAM) right in your browser! It performs both embedding and point prompting inferences, all powered by the Rust Candle Framework compiled to Wasm Space: Source: #webml

7

132

526

Radamés Ajna

@radamar

1 year

Since you all liked the face landmarks component, I made a custom @Gradio component for live pose estimation generating a conditioning image for the ControlNet openpose model. It's really fun to play with try the live demo here

11

107

511

Radamés Ajna

@radamar

10 months

Another ML gem on the @huggingface hub, generative dance moves demo: paper:

2

104

470

Radamés Ajna

@radamar

2 years

Here's an experimental drawing tool I made to interact with the Text2Human generative model capable of generating humans with clothes and textures. Playing with it and pretending to be a fashion designer is fun. by @Jiang_Yuming et al. @huggingface

6

81

429

Radamés Ajna

@radamar

1 year

The new Stable Diffusion version 2 by @StabilityAI comes with depth to image model capable of preserving more structures from the original image 🤯

11

52

426

Radamés Ajna

@radamar

1 year

Loving the hype for Drag Your GAN! As we wait for its official code release, check out a cool @gradio demo I made for its sibling project, UserControllableLT User-Controllable Latent Transformer #DragGAN demo:

12

106

411

Radamés Ajna

@radamar

8 months

Here are ten 42M TinyStories models running simultaneously in the browser. @karpathy 's Llama2.c code has been ported to Rust using the Candle framework and compiled to Wasm.

4

76

358

Radamés Ajna

@radamar

9 months

ControlNet with LoRa thanks @StabilityAI The weights are on our ComfyUI demo, try it now Space: model:

4

57

355

Radamés Ajna

@radamar

11 days

Testing InstantSplat with a few frames from Oppenheimer, super fast 3D Gaussian Splatting

10

47

353

Radamés Ajna

@radamar

11 months

Word-level timestamps now available on Transformers, I've updated an old demo project for word-level video trimming, using the @gradio HighlightedText component as input. It's fun for short videos. try it now:

12

77

313

Radamés Ajna

@radamar

4 months

Just discovered OSX Automator: attach a keyboard shortcut to run a Python script on highlighted text, instantly swapping with LLM magic! Now proofing grammar on-demand with Mistral 8x7B via @huggingface or Mistral 7B with @ollama LocalLLM

9

33

315

Radamés Ajna

@radamar

8 months

You can run language models in the browser using Candle (Rust) and WASM. Here are some examples Llama2.c

2

56

304

Radamés Ajna

@radamar

4 months

Now you can train LoRA Diffusion DPO models using @diffuserslib , thanks to @RisingSayak . Check out this thread for SD2.1 results, and watch LoRA's real-time impact with SD-Turbo.

7

63

285

Radamés Ajna

@radamar

1 month

I made a simple Gradio component that allows real-time prompt weighting. It's amazing to see the results update instantly!

10

36

286

Radamés Ajna

@radamar

9 months

New ML gem on the hub: LDM3D by @intel . This diffusion model generates image & depth from text prompts. Using a custom @gradio 6dof three.js component you can generate immersive 360-degree views from prompts demo: model:

5

93

280

Radamés Ajna

@radamar

5 months

Working on a open-source @Gradio demo for super resolution magnify app, wdyt?

30

28

276

Radamés Ajna

@radamar

11 months

I've been having fun with the Meta MusicGen music generation model - it's mind-blowing! I tweaked the demo to allow: mic input, melody trim, song continuation, and sharing on community discussions Check the @gradio demo here 🔊 Bach's Toccata And Fugue,…

4

53

249

Radamés Ajna

@radamar

2 years

Since @Gradio 3.0 was released last week, I've built a proof of concept for a video editor where you edit the video by editing the text. Powered by the newest @Gradio Blocks API and @huggingface automatic speech recognition pipeline.

10

58

239

Radamés Ajna

@radamar

10 months

Here's my initial attempt running NerfStudio as @huggingface Space template. Everything's within the same container, both the trainer and the viewer. You can now use our GPUs to train your NeRFs. ps, there are still a few steps you'll need edit on the Dockerfile. I wish the UI…

3

43

239

Radamés Ajna

@radamar

2 years

We've been working on a multiplayer experiment for #StableDiffusion powered by @huggingface GPU Space and multiplayer API by @liveblocks , inspired by @lkwq007 SD infinity project coming soon -

5

46

232

Radamés Ajna

@radamar

2 years

Here is the link for the #StableDiffusion multiplayer experiment. If the frame is empty, we run text2img. Otherwise, we inpaint/outpaint empty areas. You can zoom, draw a custom mask or override painted areas. Looking forward to seeing what you create.

12

48

234

Radamés Ajna

@radamar

11 months

This is not a QR Code

8

21

235

Radamés Ajna

@radamar

23 days

If you need a Llama3 break, here is the InstantStyle + 2step SDXL-Lightning demo for you. Generation time from ~20s to ~9s

8

50

219

Radamés Ajna

@radamar

2 years

Here is a quick @huggingface Spaces demo for the original PIFu project. From a single image, it can generate a 3D model with colors!! While it's an old method (2 years old 😂) it's still very very impressive

1

48

211

Radamés Ajna

@radamar

2 months

Testing new pix2pix-Turbo in real-time, very interesting GAN architecture that leverages SD-Turbo model. Here I'm using edge2image LoRA single-step inference 🤯

11

48

209

Radamés Ajna

@radamar

4 months

Check out this video editing demo. it's super fast now! demo:

3

32

206

Radamés Ajna

@radamar

10 months

I've got ComfyUI up and running on Spaces with a custom Dockerfile! Go ahead, duplicate it and run it on your own GPU

11

41

194

Radamés Ajna

@radamar

2 years

My latest experiment: Since now @huggingface transformers include a zero-shot depth estimation model and @Gradio has a new 3d model viewer, why not convert the depth map to a 3d object? Try it yourself, it works really well with selfies

4

36

194

Radamés Ajna

@radamar

7 months

You can now add multiple mask points and background points in our Rust Segment Anything demo. All inferences run locally on the web browser with Wasm.

2

34

189

Radamés Ajna

@radamar

1 year

I just tried the instruct InstructPix2Pix video and it's 🔥🔥🔥🔥

Sylvain Filoni

@fffiloni

1 year

And there it is 🔥 — The @gradio demo on @huggingface for Pix2Pix Video #pix2pix #pix2pixvideo ENJOY ! 🤗 SHARE YOUR RESULTS 👌 And follow for updates 😉 — Link:

13

67

302

6

30

180

Radamés Ajna

@radamar

10 months

Just 1.5 months after DragGan's publication, they've published a Diffusion Models approach for DragGan 😱

AK

@_akhaliq

10 months

DragonDiffusion: Enabling Drag-style Manipulation on Diffusion Models paper page: Despite the ability of existing large-scale text-to-image (T2I) models to generate high-quality images from detailed textual descriptions, they often lack the ability to…

1

59

269

5

27

176

Radamés Ajna

@radamar

8 months

The Microsoft Phi-1.5 model achieves state-of-the-art performance with just 1.3 billion parameters. Check it out running on a browser at ~2.1 tok/sec

3

27

171

Radamés Ajna

@radamar

5 months

@doganuraldesign thanks for the funny idea, here's my attempt using an open-source version, cc @ClementDelangue demo:

8

29

170

Radamés Ajna

@radamar

2 years

Today I'm excited to share that I have joined the @huggingface product team as a Frontend Engineer. I'm thrilled to be surrounded by top-notch researchers and developers and to share my passion for Data Visualization and Data Science.

5

11

159

Radamés Ajna

@radamar

4 months

The InstantID demo is live on @huggingface amazing work by @Haofan_Wang ! Note: It seems you can change the base model to any SDXL. I tested it with sdxl-turbo for 4 steps, and here is the result. demo:

3

25

157

Radamés Ajna

@radamar

18 days

Another distillation technique by ByteDance Hyper-SD I love unified LORA multi step support. I made a demo with InstantStyle + ControlNet, all compatible with diffusers

5

38

161

Radamés Ajna

@radamar

8 months

Here's @tairov llama2.c @Modular_AI mojo 🔥 running on docker Spaces with a simple @Gradio UI. You can duplicate and run you Mojo code demo Space:

4

38

156

Radamés Ajna

@radamar

1 year

Our Community Inference API now supports Image-to-Image models! Here are some examples of how to use it with our JavaScript library - huggingface.js. Any model with the image-to-image task tag can work. Of course Controlnet models lllyasviel/control_v11f1p_sd15_depth

3

25

145

Radamés Ajna

@radamar

7 months

BLIP Image Captioning is now running on the browser with Wasm powered by Candle. WIP more performance improvements demo:

3

21

147

Radamés Ajna

@radamar

7 months

Trying out the new @gradio /lite, running the entire Gradio app on a browser with no servers. With some internal tweaks, you can run external js code with it. Here's a Gradio UI demo for the Candle Segment Anything model

5

35

142

Radamés Ajna

@radamar

2 months

. @AndrewYNg casually making announcements on the hub.

2

17

142

Radamés Ajna

@radamar

5 months

Thanks to @SebastienBubeck for uploading the weights to the hub! Here's what we can build with it: Phi-2 quantized, running in the browser ~3 tok/s, ~1.57GB artifact. **Video is sped up** demo:

6

29

135

Radamés Ajna

@radamar

11 months

Rerender Update: You can use various SD base models, such as Analog Diffusion & Stable Diffusion PaperCut, as below. Unbatched processing takes ~10min for 6s of video. While it struggles with fast/large motions, the results are remarkable. I can't wait to see cool videos to come!

5

20

132

Radamés Ajna

@radamar

9 months

Have you tried using llama.cpp with a GPU? Check out this Space Docker template for Llama-2-7B-Chat-GGML. Easy to duplicate and switch models. Inference is lightning fast! Space:

2

41

131

Radamés Ajna

@radamar

2 years

While @Gradio team is working towards the 3.0 release, I've been stress-testing the new Blocks low-level API. This time I've stitched together a zero-shot depth estimation from an image to an autostereogram (Magic Eye) on @huggingface Spaces

6

19

132

Radamés Ajna

@radamar

3 months

@charliebholtz This is my favorite failure case 😅 I'm looking forward to creating more glitch art with this

3

1

104

Radamés Ajna

@radamar

3 months

Now, with MultiControlnet it adds more details from the reference image 🤯

8

25

98

Radamés Ajna

@radamar

2 years

For the Interactive Map Day 25 #30DayMapChallenge , I've collaborated with @recifs , @mootari on @observablehq to build this fun game! Using #TensorflowJS and #facemesh , you have to use your nose to match places around the globe. Happy Thanksgiving Map Day!

3

18

100

Radamés Ajna

@radamar

1 month

Wow! Quantized Moondream2 is now running on the browser with Candle Rust WIP, slow but it works! cc @vikhyatk @santiagomedr @lmazare

4

14

95

Radamés Ajna

@radamar

2 months

Finally played with LayerDiffuse Latent Transparency, and it's a lot of fun! You can blend from background, foreground, or just a transparent image. Made a Gradio demo

1

14

90

Radamés Ajna

@radamar

2 years

Inspired by @mattdesl and @dribnet . I made with @victormustar a live demo for you to generate color palettes from #stabledifussion hosted on @huggingface Spaces.

4

15

88

Radamés Ajna

@radamar

2 years

and another one to that transform the depth map into voxels 🧊🧊🧊

2

14

85

Radamés Ajna

@radamar

1 year

I'm very excited that you can now use JS to interact with Hugging Face! I created this interactive @observablehq notebook to explore all the supported tasks. Thanks to @coyotte508 @linesofcodedev

Hello Huggingface.js Inference

@huggingface/hub: Interact with huggingface.co to create or delete repos and commit / download files @huggingface/inference: Use the Inference API to make calls to Machine Learning models This...

observablehq.com

1

19

83

Radamés Ajna

@radamar

3 months

Amazing! this one is another candidate to run on the device demo:

Alexander Visheratin

@visheratin

3 months

VLMs have a resolution problem, which prevents them from finding small details in large images. In this @huggingface community post, I discuss the ways to solve it and describe the details of MC-LLaVA architecture:

5

38

151

3

16

79

Radamés Ajna

@radamar

5 months

Very cool, @FlowiseAI running with @supabase and mistralai/Mixtral-8x7B-Instruct-v0.1 endpoint from @huggingface here's a Space template for you

2

17

79

Radamés Ajna

@radamar

1 year

I've tried integrating StyleGAN image inversion from pixel2style2pixel, but ended up creating a bizarre face generator😱 Even though the inversion works moderately well, the transformations in the latent space with UserControllableLT didn't work as expected.…

3

9

74

Radamés Ajna

@radamar

3 months

quick test with SDXL-Lightning real-time typing with Gradio demo:

5

16

76

Radamés Ajna

@radamar

1 year

Expressive Text-to-Image Generation with Rich Text coming to our hub, integrated to @Gradio

3

14

75

Radamés Ajna

@radamar

28 days

If you need to search for @huggingface models, datasets, or spaces in your @Gradio app, we now have a hub quick search custom component. Here's an example with mergekit config generator.

2

9

72

Radamés Ajna

@radamar

5 months

@_akhaliq this is the technique behind this 🤯

あき先生 / Aki

@cumulo_autumn

5 months

StreamDiffusion、ほぼ100fpsで画像生成出来るようになりました！！ sd-turbo, 512x512, batch size 1, txt2imgだと10msで1枚画像が生成出来ます！多分これが一番速いと思います

53

1K

5K

0

12

70

Radamés Ajna

@radamar

11 months

Full source code is here

GitHub - XingangPan/DragGAN: Official Code for DragGAN (SIGGRAPH 2023)

Official Code for DragGAN (SIGGRAPH 2023). Contribute to XingangPan/DragGAN development by creating an account on GitHub.

github.com

2

15

67

Radamés Ajna

@radamar

4 months

there is an open PR and we'll update the Spaces soon

InstantID - a Hugging Face Space by InstantX

huggingface.co

3

7

67

Radamés Ajna

@radamar

11 months

quoting @_akhaliq for more context

AK

@_akhaliq

11 months

AI video to video translation demo: Rerender A Video: Zero-Shot Text-Guided Video-to-Video Translation paper proposes a novel zero-shot text-guided video-to-video translation framework to adapt image models to videos. The framework includes two parts:…

8

111

476

3

8

64

Radamés Ajna

@radamar

6 months

Here is the source. I'm sure there is room for performance improvements and image consistency

GitHub - radames/Real-Time-Latent-Consistency-Model: App showcasing multiple real-time diffusion...

App showcasing multiple real-time diffusion models pipelines with Diffusers - radames/Real-Time-Latent-Consistency-Model

github.com

2

9

63

Radamés Ajna

@radamar

2 years

that's crazy! here's a very very long downhill street in street in San Francisco #StableDiffusion

AK

@_akhaliq

2 years

Optimized Stable Diffusion update, 1600x1600 and 1920x1088 on 8 gb vram colab: github: reddit thread:

4

91

573

0

3

59

Radamés Ajna

@radamar

1 year

Exciting tool for #JavaScript devs from @Gradio . The new gradio/client lets you use any Gradio app as an API, providing an easier way to build custom front-end components. Check out this interactive tutorial on @observablehq

Hello Gradio Client JS

Gradio is an open-source Python library designed to create web applications and demos for machine learning and data science. It offers a simple and straightforward way to build and distribute a...

observablehq.com

2

18

58

Radamés Ajna

@radamar

2 years

It seems like #dallemini also speaks other languages

2

6

56

Radamés Ajna

@radamar

15 days

I've been using ZeroGPU on my @huggingface projects and it's very promising 40GB of vRAM workloads on A100s

Victor M

@victormustar

15 days

ZeroGPU is free distributed GPUs in HF Spaces 🔥 ⬇️ will give access to 100 new people in the next hours

42

35

248

2

8

56

Radamés Ajna

@radamar

1 month

I've plugged it into the SDXL Lightning Space

Real-Time Text-to-Image SDXL Lightning - a Hugging Face Space by radames

huggingface.co

4

6

51

Radamés Ajna

@radamar

1 year

Bookmark this javascript code to make the minimal Stable Diffusion UI thanks to @huggingface inference API

1

8

51

Radamés Ajna

@radamar

1 year

a @gradio GPT-4 experiment that transforms a prompt and media files into an FFMPEG command. thanks to @victormustar

0

10

50

Radamés Ajna

@radamar

1 year

The latest 🔥 text-to-image model on the scene, DeepFloyd IF @deepfloydai . Imagining if @huggingface were a cookie company.

2

7

47

Radamés Ajna

@radamar

8 months

Thanks to @kolibril13 for the inspiration, here is the @Gradio version running Blender render engine with the bpy pip package Space: Code:

1

11

49

Radamés Ajna

@radamar

2 years

ok, now the Spaces is running on a GPU container, it takes less than 10s to generate a model, you can also download the glTF 3D file. Thanks to @ak92501 I'm using now u-2-net background removal

2

11

47

Radamés Ajna

@radamar

5 months

featuring new @Gradio custom component

pngwn

@evilpingwin

5 months

I've just published a custom `ImageSlider` component for @Gradio Easily compare two images with a nice slidey UI.

6

19

117

3

5

44

Radamés Ajna

@radamar

1 year

Try this new and impressive TTS @gradio demo by @elevenlabsio , showcasing impressive multilingual capabilities. "Bonjour, mon ami ¿Cómo estás hoy? Como vai? Ich hoffe, es geht dir gut. La vita è bella, non è vero?. Ας χαμογελάσουμε στη ζωή."

3

11

43

Radamés Ajna

@radamar

4 months

here is the code

GitHub - radames/LLM-automator: Create keyboard shortcuts for an LLM using OpenAI GPT, Ollama,...

Create keyboard shortcuts for an LLM using OpenAI GPT, Ollama, HuggingFace with Automator on macOS. - radames/LLM-automator

github.com

0

2

46

Radamés Ajna

@radamar

2 months

Very interesting image model trained on legally licensed data @diffuserslib compatible demo:

Ron Mokady

@MokadyRon

2 months

Excited to share new @huggingface demos for @bria_ai_ text-to-image 🚨 As always our models do not violate copyrights, trained on 100% legal data 🌟 BRIA HD offers full HD resolution (1920X1080) for high-quality textures 🚀 👉

10

32

167

1

8

42

Radamés Ajna

@radamar

1 year

Now there's an easy way to share and load rich text prompts in the Expressive Text-to-Image Generation demo

2

12

40

Radamés Ajna

@radamar

10 months

addictive game running purely in the browser. thanks for sharing the process @xenovacom

Xenova

@xenovacom

10 months

Introducing Doodle Dash, an ML-powered web game that runs completely in your browser, thanks to Transformers.js! 🤯 You have 60 seconds to draw as many words as you can, while a neural network tries to guess what you're drawing in real time! Play here:

5

29

103

0

6

40

Radamés Ajna

@radamar

11 months

Absolutely mind-blowing editing technique 🤯

Linoy Tsaban🎗️

@linoy_tsaban

11 months

Introducing the DDPM inversion Space 🤗- a new Space for real image editing 🖼️ Based on the very cool edit friendly DDPM inversion method by @inbarhub This technique somehow got under the radar and should get more attention🔥 So how does it work? 🧵1/7

4

48

161

2

9

37

Radamés Ajna

@radamar

9 months

Another Space template for you:🧘‍♀️Fooocus by @lvminzhang minimal and magical UI for so you can focus on prompting and generating. Duplicate and use on your own GPU. live demo: code: