Frank (Haofan) Wang @Haofan_Wang Twitter profile | Pikagi

Pikagi

Frank (Haofan) Wang

@Haofan_Wang

1,082

Followers

330

Following

69

Media

382

Statuses

Researcher @InstantX . Ex @CarnegieMellon . I'm a NPC.

Beijjng

https://t.co/2T6WdifqTz

Joined February 2017

Don't wanna be here? Send us removal request.

Pinned Tweet

@Haofan_Wang

Frank (Haofan) Wang

1 month

Thanks @_akhaliq for sharing! TL;DR: InstantStyle is a framework that employs straightforward yet potent techniques for achieving effective disentanglement of style and content from reference images. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

@_akhaliq

AK

1 month

InstantStyle Free Lunch towards Style-Preserving in Text-to-Image Generation Tuning-free diffusion-based models have demonstrated significant potential in the realm of image personalization and customization. However, despite this notable progress, current models continue to

Tweet media one

1

31

149

6

37

191

Last Seen Profiles

@AbanteNews

@TippePvP

@2Late4Stop

@BradyKoehler3

@kogeinu

@Carissaiel

@Ignites

@sovyvmazutu

@BadKissingen

@Ryn_writes

@fedepizarro03

@urosnoetic

@OrcaForce

@BeechamRochne

@paoconaguacate

@ParentClub

@Elizabethbtc017

@Tre_Byrd_20

@augeeto

@lherealchieff

@Hamathyst

@cody_hirano

@travis_craven7

@RafaSBF_

@SnKumaa

@DardarNato16146

@RCHumaneSociety

@vaughnjenne1983

@xHatake2k

@alvin__yuan

@AnkaraGayVideos

@miguelangels

@hunterk17_

@BoxersBaseball

@EricDickerson

@AnkaraGayVideos

@Haofan_Wang

Frank (Haofan) Wang

4 months

Thanks @_akhaliq for sharing our work! TL;DR: We introduce InstantID as the state-of-the-art tuning-free method to achieve ID-Preserving generation with only single image. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

@_akhaliq

AK

4 months

InstantID: Zero-shot Identity-Preserving Generation in Seconds paper page: model supports identity-preserving generation in high fidelity with only single reference image in any style

Tweet media one

6

67

328

8

84

364

@Haofan_Wang

Frank (Haofan) Wang

4 months

InstantID + LCM-LoRA by @SimianLuo

Tweet media one

3

24

154

@Haofan_Wang

Frank (Haofan) Wang

4 months

We have released pre-trained checkpoints, inference codes and a online Gradio demo. Thanks for all your waits! Code: Demo:

Tweet card media

InstantID - a Hugging Face Space by InstantX

@Haofan_Wang

Frank (Haofan) Wang

4 months

Thanks @_akhaliq for sharing our work! TL;DR: We introduce InstantID as the state-of-the-art tuning-free method to achieve ID-Preserving generation with only single image. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

8

84

364

5

36

145

@Haofan_Wang

Frank (Haofan) Wang

5 months

InstantID Looks much better than IP-Adapter, would be very helpful for preserving ID consistency.

Tweet media one

4

23

103

@Haofan_Wang

Frank (Haofan) Wang

3 months

InstantID works with ControlNet Pose and LCM now, you will see it on huggingface spaces soon! Credit to @radamar

@radamar

Radamés Ajna

3 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

152

810

1

22

101

@Haofan_Wang

Frank (Haofan) Wang

3 months

Now, you have SDXL-lightning on InstantID! For this example (faded film style), I think LCM-LoRA suffer less from style degradation (not a rigorous comparison).

Tweet media one

5

10

84

@Haofan_Wang

Frank (Haofan) Wang

4 months

I'm so excited to share that our InstantID is ranked #1 on huggingface spaces trending. Proud of our InstantX team❤️! @QixunWang @baymin_coding @qinke87 @antonch44379293

Tweet media one

@Haofan_Wang

Frank (Haofan) Wang

4 months

Thanks @_akhaliq for sharing our work! TL;DR: We introduce InstantID as the state-of-the-art tuning-free method to achieve ID-Preserving generation with only single image. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

8

84

364

6

11

80

@Haofan_Wang

Frank (Haofan) Wang

3 months

Thanks to the community, InstantID will be supported in WebUI soon.

Tweet card media

:sparkles: Instant ID by huchenlei · Pull Request #2580 · Mikubill/sd-webui-controlnet

Instant ID project https://github.com/InstantID/InstantID Instant ID uses a combination of ControlNet and IP-Adapter to control the facial features in the diffusion process. One unique design for I...

1

10

73

@Haofan_Wang

Frank (Haofan) Wang

4 months

InsantID is finally out! Check paper for more details. The code and pre-trained models will be released within this month. Paper: Project Page: Code:

Tweet media one

2

23

72

@Haofan_Wang

Frank (Haofan) Wang

3 months

MultiControlNet will be supported in InstantID very soon.

@radamar

Radamés Ajna

3 months

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

Tweet media one

8

25

98

1

9

71

@Haofan_Wang

Frank (Haofan) Wang

2 months

OpenDiT is a great work by @oahzxl @zzk_zhao @lzm_mlsys from @NUSingapore , which is an Easy, Fast and Memory-Efficent System for DiT Training and Inference. This year will belong to DiT, you can't miss it if you are on generative boat.

Tweet card media

GitHub - NUS-HPC-AI-Lab/OpenDiT: OpenDiT: An Easy, Fast and Memory-Efficient System for DiT...

OpenDiT: An Easy, Fast and Memory-Efficient System for DiT Training and Inference - NUS-HPC-AI-Lab/OpenDiT

2

9

58

@Haofan_Wang

Frank (Haofan) Wang

3 months

InstantID is running on A100 GPUs now, supported by OpenXLab.

@Haofan_Wang

Frank (Haofan) Wang

4 months

Thanks @_akhaliq for sharing our work! TL;DR: We introduce InstantID as the state-of-the-art tuning-free method to achieve ID-Preserving generation with only single image. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

8

84

364

0

10

54

@Haofan_Wang

Frank (Haofan) Wang

2 months

I'm excited to share that our InstantID has been deployed on @FEDML_AI which provides the generative AI platform and foundation models to enable developers and enterprises to build and commercialize generative AI applications. Find more info at .

Tweet media one

1

11

52

@Haofan_Wang

Frank (Haofan) Wang

25 days

InstantStyle are on HuggingFace Trending now! As the successor to InstantID.

Tweet media one

@Haofan_Wang

Frank (Haofan) Wang

1 month

Thanks @_akhaliq for sharing! TL;DR: InstantStyle is a framework that employs straightforward yet potent techniques for achieving effective disentanglement of style and content from reference images. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

6

37

191

3

9

51

@Haofan_Wang

Frank (Haofan) Wang

1 month

Image-based stylization has been supported in InstantStyle. Found more information at …. We will further combine InstantID for face stylization once stars reach 1K on GitHub.

Tweet media one

@Haofan_Wang

Frank (Haofan) Wang

1 month

Thanks @_akhaliq for sharing! TL;DR: InstantStyle is a framework that employs straightforward yet potent techniques for achieving effective disentanglement of style and content from reference images. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

6

37

191

2

9

47

@Haofan_Wang

Frank (Haofan) Wang

16 days

Integrated by our teammate and HuggingFace Team, InstantStyle is now natively supported in diffusers❤️

Tweet media one

@Haofan_Wang

Frank (Haofan) Wang

1 month

Thanks @_akhaliq for sharing! TL;DR: InstantStyle is a framework that employs straightforward yet potent techniques for achieving effective disentanglement of style and content from reference images. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

6

37

191

4

10

48

@Haofan_Wang

Frank (Haofan) Wang

10 months

Very lucky citation number, let's go for 1000 this year! 2 papers got accepted to ICCV 2023, see you in Paris.

Tweet media one

2

0

45

@Haofan_Wang

Frank (Haofan) Wang

30 days

InstantStyle is officially on @Gradio now!

Tweet card media

InstantStyle - a Hugging Face Space by InstantX

1

11

44

@Haofan_Wang

Frank (Haofan) Wang

2 months

A recent comparison about Distillation methods (4 steps with CFG=0). I have used these methods in my daily workflow, and I love SDXL-Lightning most for its good tradeoff between style degradation and image quality. Not sure whether TCD achieves the worst result in my test.

Tweet media one

@Haofan_Wang

Frank (Haofan) Wang

3 months

Now, you have SDXL-lightning on InstantID! For this example (faded film style), I think LCM-LoRA suffer less from style degradation (not a rigorous comparison).

Tweet media one

5

10

84

1

4

43

@Haofan_Wang

Frank (Haofan) Wang

4 months

Tuning, see you soon.

Tweet media one

@Gradio

Gradio

4 months

Coming Soon... to a Gradio Space for the community! #InstantID 😉

Tweet media one

3

15

104

3

10

39

@Haofan_Wang

Frank (Haofan) Wang

1 month

InstantStyle is supported in AnyV2V by @vinesmsuic for video stylization. Excellent performance. InstantStyle: AnyV2V:

Tweet card media

GitHub - TIGER-AI-Lab/AnyV2V: Perform Video Editing with only one image. Now with gradio demo

Perform Video Editing with only one image. Now with gradio demo - TIGER-AI-Lab/AnyV2V

@vinesmsuic

Max Ku @ICLR2024

1 month

AnyV2V + InstantStyle

0

4

11

0

10

39

@Haofan_Wang

Frank (Haofan) Wang

28 days

Thank you for making this happen!

@cocktailpeanut

cocktail peanut

@cocktailpeanut

28 days

Run InstantStyle Locally with 1 Click InstantStyle lets you generate images with a style of ANY other image, instantly. No LoRA required. Both text-to-image/image-to-image. I wrote a 1 click launcher for the gradio app from @Haofan_Wang (The author of InstantStyle/InstantId!).

9

31

114

1

5

38

@Haofan_Wang

Frank (Haofan) Wang

3 months

Now, InstantID is on WebUI now. Try it and share your results with us!

Tweet card media

[1.1.439] Instant ID · Mikubill sd-webui-controlnet · Discussion #2589

Instant ID project https://github.com/InstantID/InstantID Instant ID uses a combination of ControlNet and IP-Adapter to control the facial features in the diffusion process. One unique design for I...

@Haofan_Wang

Frank (Haofan) Wang

3 months

Thanks to the community, InstantID will be supported in WebUI soon.

1

10

73

1

4

32

@Haofan_Wang

Frank (Haofan) Wang

3 months

We clarify that is quite misleading but not authorized and has never contacted us for official cooperation. Please pay attention to your personal privacy. We currently only have a project page, Github page, and a huggingface spaces demo. #InstantID

@Haofan_Wang

Frank (Haofan) Wang

4 months

Thanks @_akhaliq for sharing our work! TL;DR: We introduce InstantID as the state-of-the-art tuning-free method to achieve ID-Preserving generation with only single image. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

8

84

364

1

4

34

@Haofan_Wang

Frank (Haofan) Wang

1 month

We've gotten a lot of love❤️ from our users, and now you can support InstantID by buying us a cup of coffee via GitHub Sponsor (). More interesting projects are on the way.

Tweet card media

Buy InstantX a Coffee. ko-fi.com/instantx

Become a supporter of InstantX today! ❤️ Ko-fi lets you support the creators you love with no fees on donations.

@Haofan_Wang

Frank (Haofan) Wang

4 months

Thanks @_akhaliq for sharing our work! TL;DR: We introduce InstantID as the state-of-the-art tuning-free method to achieve ID-Preserving generation with only single image. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

8

84

364

5

5

31

@Haofan_Wang

Frank (Haofan) Wang

1 month

Analyzing the contribution of each atomic layer is incomplete because different layers may influence each other. This is quite obvious in SD1.5, and only a few block mixtures will show obvious semantics. But like SDXL, these representation layers are all located near mid_block.

Tweet media one

@Haofan_Wang

Frank (Haofan) Wang

1 month

Thanks @_akhaliq for sharing! TL;DR: InstantStyle is a framework that employs straightforward yet potent techniques for achieving effective disentanglement of style and content from reference images. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

6

37

191

1

6

30

@Haofan_Wang

Frank (Haofan) Wang

4 months

@ylecun 4) LeCun Universe

Tweet media one

Tweet media two

Tweet media three

1

1

26

@Haofan_Wang

Frank (Haofan) Wang

3 months

New features (LCM + Multi-ControlNets) are on Huggingface Spaces now. All thanks to @radamar .

@radamar

Radamés Ajna

3 months

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

Tweet media one

8

25

98

0

6

24

@Haofan_Wang

Frank (Haofan) Wang

1 month

InstantStyle + Face-to-All will be very interesting!

@multimodalart

apolinario (multimodal.art)

1 month

Introducing Face-to-All👨‍🎤, a diffusers 🧨 workflow inspired by @fofrAI amazing Face-to-Many ComfyUI workflow Input a face, any style LoRA and get a stylized portrait Colab with code: Thanks @Haofan_Wang for merging our img2img pipeline to InstantID!

Tweet media one

7

30

132

0

1

24

@Haofan_Wang

Frank (Haofan) Wang

1 month

InstantStyle is naturally supported in ComfyUI developed by @cubiq . Follow our updates at

Tweet card media

GitHub - InstantStyle/InstantStyle: InstantStyle: Free Lunch towards Style-Preserving in Text-to-...

InstantStyle: Free Lunch towards Style-Preserving in Text-to-Image Generation 🔥 - InstantStyle/InstantStyle

@Haofan_Wang

Frank (Haofan) Wang

1 month

Thanks @_akhaliq for sharing! TL;DR: InstantStyle is a framework that employs straightforward yet potent techniques for achieving effective disentanglement of style and content from reference images. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

6

37

191

0

7

23

@Haofan_Wang

Frank (Haofan) Wang

30 days

We are also on GitHub Trending today.

Tweet media one

@Haofan_Wang

Frank (Haofan) Wang

1 month

Our InstantStyle has been on top Trending Research at . Don't miss it.

Tweet media one

0

1

8

0

2

22

@Haofan_Wang

Frank (Haofan) Wang

1 month

Interesting work! Cobra extends Mamba to Multi-Modal LLM for efficient inference. Try on huggingface

Tweet card media

Cobra - a Hugging Face Space by han1997

@han_zhao97

Han Zhao

1 month

Our Huggingface demo is online! Welcome to try it out! The code and weight are also updated. Demo: ; Project page: ; Weight: ; Github: ; ArXiv:

Tweet media one

0

6

32

0

5

21

@Haofan_Wang

Frank (Haofan) Wang

1 month

Tweet media one

5

1

21

@Haofan_Wang

Frank (Haofan) Wang

3 months

We are co-organizing a Spring Festival event with @huggingface from 2.7 to 2.25 on @xiaohongshu . Post your image with Spring Festival costumes, and win our official gifts. Happy Chinese 🐲 New Year! Happy Lunar New Year!

@Haofan_Wang

Frank (Haofan) Wang

4 months

Thanks @_akhaliq for sharing our work! TL;DR: We introduce InstantID as the state-of-the-art tuning-free method to achieve ID-Preserving generation with only single image. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

8

84

364

0

4

20

@Haofan_Wang

Frank (Haofan) Wang

2 months

It's sad to know all plugins (LoRAs, ControlNets, Adapters, InstantID, etc) for playground-2.5 model need to be re-trained although it's also SDXL structure. A good news is that I have made a PR to well support playground-2.5 in diffusers.

Support latents_mean and latents_std by haofanwang · Pull Request #7132 · huggingface/diffusers

As title. This PR supports latents_mean and latents_std in the decoding stage of VAE as done in pipeline_stable_diffusion_xl.py to make the pipeline compatible to models like playgroundai/playgroun...

@Suhail

Suhail

2 months

@Haofan_Wang @SimianLuo The community would need to train it and help so we can focus on the foundation model bits.

0

0

1

1

2

20

@Haofan_Wang

Frank (Haofan) Wang

20 days

Thanks for @DanielCohenOr1 for mentioning our InstantID at China 3DV.

Tweet media one

0

0

17

@Haofan_Wang

Frank (Haofan) Wang

4 months

Thanks @hysts12321 for supporting webcam and uploading images from mobile devices on Spaces.

@Gradio

Gradio

4 months

🔥InstantID demo is now out on Spaces. Thanks @Haofan_Wang et al, for building a brilliant Gradio demo for the community🙌 Check out the path-breaking demo now! Here is an example of a Marvel superhero, 🦸‍♂️ @ylecun , generated using InstatID within seconds!

Tweet media one

6

20

135

0

2

16

@Haofan_Wang

Frank (Haofan) Wang

2 months

Renting an A100 (80GB) costs about $1,000 per month, which means you'd have to have 100 subscribers willing to pay $10 per month to break even. This is just the cost of GPUs.

3

0

16

@Haofan_Wang

Frank (Haofan) Wang

2 months

Try this amazing tool () for generating professional headshots on the fly. To be honest, I don't want to upload dozens of my personal images to a website that uses my images for training, expensive, time-costing and unsafe.

Discord - A New Way to Chat with Friends & Communities

Discord is the easiest way to communicate over voice, video, and text. Chat, hang out, and stay close with your friends and communities.

@ArtiveAI

Artive Studio

2 months

1) Professional Headshot. Obtain your new headshot with one image, in several seconds. Try the free trial on our Discord ().

Tweet media one

0

1

6

1

2

15

@Haofan_Wang

Frank (Haofan) Wang

1 month

Crowdfunding for open source projects would be a very interesting social experiment. Is it a feasible idea?

4

1

14

@Haofan_Wang

Frank (Haofan) Wang

13 days

Great work!

@HuaqiangLiu666

Qinghe Wang

@HuaqiangLiu666

13 days

🚀CharacterFactory: Sampling Consistent Characters with GANs for Diffusion Models We propose CharacterFactory, a framework that allows sampling new characters with consistent identities in the latent space of GANs for diffusion models. 🔥Project page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

3

12

42

0

2

14

@Haofan_Wang

Frank (Haofan) Wang

3 months

Thanks @SiliconFlowAI for their OneDiff integration of our InstantID! You can enjoy accelerated inference for InstantID (1.8x acceleration on RTX 4090). You can find more details at

@Haofan_Wang

Frank (Haofan) Wang

4 months

Thanks @_akhaliq for sharing our work! TL;DR: We introduce InstantID as the state-of-the-art tuning-free method to achieve ID-Preserving generation with only single image. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

8

84

364

0

3

13

@Haofan_Wang

Frank (Haofan) Wang

3 months

Thanks @MikeShou1 for hosting me at ShowLab, thanks @YangYou1991 for arranging the guest lecture, and finally thanks the team from @HPCAITech for the hospitality tonight. Now, It’s time to take a look at my experiments and OpenAI’s Sora.

Tweet media one

Tweet media two

Tweet media three

1

0

13

@Haofan_Wang

Frank (Haofan) Wang

3 months

Coding tonight, lots of ideas to try. InstantID is not the end, not even the beginning of the end, our group are working very hard on other interesting projects, more details will be released gradually on . Run, don't walk, and always get prepared.

Tweet media one

2

2

12

@Haofan_Wang

Frank (Haofan) Wang

3 months

Do we need a stronger style model under the circumstance that we already have IP-Adapter, StyleDrop, StyleAlign and etc. Style reference on Midjourney is also good. But we cannot define a style accurately, right?

4

0

11

@Haofan_Wang

Frank (Haofan) Wang

1 month

Put InstantStyle into video domain.

@CongWei1230

Cong Wei

1 month

AnyV2V can bring Any Image Editing method to the video domain at no cost! Now we have InstantStyle + AnyV2V! AnyV2V InstantStyle Thanks @Haofan_Wang @vinesmsuic for supporting!

1

14

72

0

2

12

@Haofan_Wang

Frank (Haofan) Wang

2 months

Well, InstantID can already achieve this feature. The over-saturated result of InstantID reported in this paper is not fair, we are much better actually. Anyway, time will tell.

@op7418

歸藏(guizang.ai)

2 months

腾讯这个新研究牛批啊，支持多角色多概念在一张图片中生成。以前的 ID 或者概念保持项目只能将一个人还原在图片里面，有了这个项目以后就可以多人合照了。项目还支持与原有的 ID 保持项目一起使用比如 Lora 以及InstantID。代码已经开源，大佬们可以看看插件实现了。项目介绍：…

Tweet media one

12

78

285

3

0

10

@Haofan_Wang

Frank (Haofan) Wang

28 days

Find the most valuable diffusion-based works at by @nhciao

Tweet card media

Stable Diffusion 生态论文精选 - Latent Box

@bdsqlsz

青龍聖者

28 days

SD best papers all you need.

Tweet media one

3

33

121

0

0

11

@Haofan_Wang

Frank (Haofan) Wang

1 month

Community contribution is always welcome!

@Gradio

Gradio

1 month

🤔Want to experience the power of InstantStyle? 🚀Code has been released, we'd welcome community contributed gradio demos! 💡Get inspired and start building your own style-preserving image generation apps with Gradio! Start here-

1

1

6

1

2

11

@Haofan_Wang

Frank (Haofan) Wang

1 month

Decoupling content and style is promising. I trained B-LoRA locally on several samples using official training setting but cannot get satisfied results (middle columns). Not sure is there something I have missed.

Tweet media one

@yarden343

Yarden Frenkel

1 month

Excited to share our new work B-LoRA🚀. With our method, you can use a simplified version of LoRA trained on SDXL to perform style transfer between images and manipulate styles based on text. Check out our website:

Tweet media one

6

36

134

1

2

11

@Haofan_Wang

Frank (Haofan) Wang

1 month

Awesome! We will support in diffusers API very soon.

@linoy_tsaban

Linoy Tsaban🎗️

1 month

Had to try InstantStyle myself, so good 🔥 I used the default settings - no tweaking ⚡️ no optimization & fully compatible with ip-adapter = 🚀

Tweet media one

3

9

68

0

0

11

@Haofan_Wang

Frank (Haofan) Wang

4 months

InstantID is on Windows right now.

@bdsqlsz

青龍聖者

4 months

After testing,InstantID>IpadapterFaceID v2 >Photomaker in datasets testing InstantID traiginng a controlnetxl model for position i made windows, 16G VRAM need Auto install and Download Model. it used SDXL model Feture: ·install windows ·load local model ·support mac mps

Tweet media one

4

16

54

0

2

10

@Haofan_Wang

Frank (Haofan) Wang

3 months

My notes on Stable Diffusion 3 based on its generated results. It will become a new baseline soon. 1) Better text understanding capability with a new text encoder, maybe T5-XXL model. (1/n)

@StabilityAI

Stability AI

3 months

Announcing Stable Diffusion 3, our most capable text-to-image model, utilizing a diffusion transformer architecture for greatly improved performance in multi-subject prompts, image quality, and spelling abilities. Today, we are opening the waitlist for early preview. This phase…

Tweet media one

269

1K

5K

1

0

10

@Haofan_Wang

Frank (Haofan) Wang

4 months

We will be on Gradio soon!

@Gradio

Gradio

4 months

Zero-shot face-adapted image generation is a rapidly developing niche research field. If you're looking to stay ahead of the curve or to simply exploring current possibilities with Gradio apps, this thread is the perfect place to start. 1⃣IPAdapter 2⃣PhotoMaker 3⃣InstantID

Tweet media one

5

32

163

1

1

9

@Haofan_Wang

Frank (Haofan) Wang

27 days

Next is learning how to build and deliver great products.

Tweet media one

0

0

9

@Haofan_Wang

Frank (Haofan) Wang

1 year

We support #ControlNet and #T2IAdapter both in #diffusers now! A few comments for recent progress: (1) The first released work has huge advantage over all other followers. (2) The completness and usability are quite important.

GitHub - haofanwang/ControlNet-for-Diffusers: Transfer the ControlNet with any basemodel in...

Transfer the ControlNet with any basemodel in diffusers🔥 - haofanwang/ControlNet-for-Diffusers

1

3

9

@Haofan_Wang

Frank (Haofan) Wang

2 months

I do love open source. But it sucks when I see some other companies adopting our work directly into their products for profit, ignoring the licenses you have.

3

0

9

@Haofan_Wang

Frank (Haofan) Wang

3 months

I'm very happy to visit MMLAB @NTU lead by @liuziwei7 today. Thanks @scy994 for hosting me. Thanks @ziqi_huang_ @JohannWang4 @Jiang_Yuming for the warm campus tour. It was a great experience and I look forward to future research collaborations.

@TheAITalksOrg

The AI Talks

3 months

Thank you @Haofan_Wang for the nice talk! Wish you a wonderful holiday trip in Singapore 🇸🇬! 🎦 Recording uploaded:

0

2

8

0

1

9

@Haofan_Wang

Frank (Haofan) Wang

1 month

Style is an underdetermined and mixed attribute that covers color, material, atmosphere, design, structure and so no. In some cases, it's even so tightly coupled to content that simply removing the content would break the style. But we can decouple part of cases at least.

1

1

8

@Haofan_Wang

Frank (Haofan) Wang

1 month

God of Thunder @leijun @Xiaomi #XiaomiSU7

Tweet media one

0

0

8

@Haofan_Wang

Frank (Haofan) Wang

2 months

I'm a big fan of IP-Adapter, which is an elegant but effective work. It has already performed pretty good on style consistency. But do you think we need a better style adapter than IP-Adapter?

4

0

8

@Haofan_Wang

Frank (Haofan) Wang

8 months

1000 citations! A new milestone.

Tweet media one

1

0

8

@Haofan_Wang

Frank (Haofan) Wang

1 month

Our InstantStyle has been on top Trending Research at . Don't miss it.

Tweet media one

@Haofan_Wang

Frank (Haofan) Wang

1 month

Thanks @_akhaliq for sharing! TL;DR: InstantStyle is a framework that employs straightforward yet potent techniques for achieving effective disentanglement of style and content from reference images. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

6

37

191

0

1

8

@Haofan_Wang

Frank (Haofan) Wang

21 days

If you have an advanced and practical model, go for closed-source product and make it profitable. If not, go for open-sourcing for popularity. The hardest thing is to find the critical point.

0

1

8

@Haofan_Wang

Frank (Haofan) Wang

4 months

Thanks @OlivioSarikas for making a Youtube video to introduce our InstantID!

Tweet card media

Depth-Anything for A1111 and InstantID

Depth-Anything for A1111 and InstantID are released. CAUTION: InstantID is based on InsightFace Technology. So no commercial use! Depth-Anything is suprising...

www.youtube.com

@Haofan_Wang

Frank (Haofan) Wang

4 months

Thanks @_akhaliq for sharing our work! TL;DR: We introduce InstantID as the state-of-the-art tuning-free method to achieve ID-Preserving generation with only single image. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

8

84

364

2

3

8

@Haofan_Wang

Frank (Haofan) Wang

2 months

The results look great. But it is not compatible with LCM-LoRA or SDXL-lightning in my local test. Any plan to support this model? We really need it in deployment. @SimianLuo

@Suhail

Suhail

2 months

1/ We are releasing Playground v2.5, our latest foundation model to create images. We tested our model across 20K+ users in a rigorous benchmark that went beyond anything we've seen to date. This model is open weights. More information in the tweets below. 👇

75

161

1K

1

0

6

@Haofan_Wang

Frank (Haofan) Wang

3 months

Hello, Singapore. DM me if you are nearby.

Tweet media one

0

0

7

@Haofan_Wang

Frank (Haofan) Wang

3 months

I plan to travel to Singapore with my family during the Chinese New Year. If you are a researcher, entrepreneur or investor and are interested in our recent work, drop me a message, I am happy to take a coffee and discuss.

1

0

7

@Haofan_Wang

Frank (Haofan) Wang

5 months

Prerequisites for generative research (1) A project page with fancy demos and potential applications. (2) Fast integration into popular library such as diffusers, with support for common backbones such as SD1.5 and SDXL. (3) Repost by AK on social media.

1

0

6

@Haofan_Wang

Frank (Haofan) Wang

8 months

About two years ago, we tried to generate a coherent comic-like story by retrieving. We never knew it could be achieved with generative models and LLM then. By

Tweet media one

0

0

5

@Haofan_Wang

Frank (Haofan) Wang

4 months

@ylecun 2) Tony LeCun in the office @ylecun

Tweet media one

1

0

6

@Haofan_Wang

Frank (Haofan) Wang

4 months

I'm excited to see that InstantID has been on Github and Paperwithcode's trending. At this moment, we are working very hard on diffusers integration and gradio demo, you will see both of them very very soon. So, don't forget to star our work.

Tweet media one

Tweet media two

0

0

6

@Haofan_Wang

Frank (Haofan) Wang

1 year

I generate some famous logos via #ControlNet , it's really amazing to see such a big progress made in AI. #AIGC

Tweet media one

Tweet media two

Tweet media three

Tweet media four

2

0

6

@Haofan_Wang

Frank (Haofan) Wang

3 months

This is exactly what I'm working on.

@hugovntr

Hugo Ventura

3 months

Midjourney just released a new feature: Style Reference It allows you to use images as references, similar to what the style tuner offered back in v5. However, the major difference between the two is that sref does not generate a "unique identifier" for your styles. → This…

Tweet media one

19

62

504

0

0

6

@Haofan_Wang

Frank (Haofan) Wang

1 month

Our work InstantStyle is lightweight and can be easily integrated into other tasks with a few lines, such as stylized video generation. Free feel to contact us for joint promotion on GitHub and Twitter👐

@Haofan_Wang

Frank (Haofan) Wang

1 month

Thanks @_akhaliq for sharing! TL;DR: InstantStyle is a framework that employs straightforward yet potent techniques for achieving effective disentanglement of style and content from reference images. Code: Project Page:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

6

37

191

0

0

6

@Haofan_Wang

Frank (Haofan) Wang

1 month

InstantStyle or FreeStyle, which name is better?

3

0

6

@Haofan_Wang

Frank (Haofan) Wang

4 months

Flying to the Mars. @elonmusk

Tweet media one

Tweet media two

Tweet media three

@Haofan_Wang

Frank (Haofan) Wang

4 months

Tuning, see you soon.

Tweet media one

3

10

39

0

0

6

@Haofan_Wang

Frank (Haofan) Wang

4 months

Thank you for providing GPUs.

@Gradio

Gradio

4 months

🔥InstantID demo is now out on Spaces. Thanks @Haofan_Wang et al, for building a brilliant Gradio demo for the community🙌 Check out the path-breaking demo now! Here is an example of a Marvel superhero, 🦸‍♂️ @ylecun , generated using InstatID within seconds!

Tweet media one

6

20

135

0

0

4

@Haofan_Wang

Frank (Haofan) Wang

4 months

Does the number of Github Stars really matter? As of now, PhotoMaker’s stars (6.1K) are greater than the sum of InstantID (1.8K) +IP-Adapter (2.8K). Can anyone tell me why?

Tweet media one

Tweet media two

Tweet media three

3

0

5

@Haofan_Wang

Frank (Haofan) Wang

3 months

Looks great! Let me think how to use it in InstantID.

Tweet media one

Tweet media two

@MokadyRon

Ron Mokady

3 months

Excited to share we just open-sourced our new background segmentation model 🥳 🚨 Check out our @gradio demo RMBG v1.4 by BRIA excels in separating foreground from background across diverse categories, surpassing current open models 🚀

13

92

394

0

0

5

@Haofan_Wang

Frank (Haofan) Wang

1 month

I visited Sanxindui museum this holiday and was really shocked by the splendid civilization at Bronze Age. Go back to work tomorrow, keep moving forward.

Tweet media one

Tweet media two

Tweet media three

Tweet media four

0

0

5

@Haofan_Wang

Frank (Haofan) Wang

19 days

昨天和振宇吃饭，确实能感觉出来创业不易

@turingou

郭宇 guoyu.eth

19 days

短暂回上海的几天去振宇家吃了顿饭，聊了很多字节的八卦故事，今年正好与我北上苏州街相隔十年，十年间，多少普通人的命运发生转变，我们都认为这些故事值得被记录，振宇夫人正在筹划一个对谈节目，嘉宾就从我们认识的身边人找起…

Tweet media one

Tweet media two

Tweet media three

Tweet media four

22

6

218

0

0

5

@Haofan_Wang

Frank (Haofan) Wang

4 months

One of our teammate @antonch44379293 will give a talk to introduce our recent InstantID at Google this month.

@Haofan_Wang

Frank (Haofan) Wang

4 months

We have released pre-trained checkpoints, inference codes and a online Gradio demo. Thanks for all your waits! Code: Demo:

5

36

145

0

0

4

@Haofan_Wang

Frank (Haofan) Wang

4 months

Another online Gradio demo. Thanks.

@camenduru

camenduru

4 months

🖼 InstantID is now running on ✨ the non-profit GPU cluster 🥳 Thanks to @QixunWang ❤ Xu Bai ❤ @Haofan_Wang ❤ Zekui Qin ❤ Anthony Chen ❤ 🌐page: 🧿demo: please try it 🐣

Tweet media one

6

32

157

0

0

5

@Haofan_Wang

Frank (Haofan) Wang

29 days

Thanks to @RisingSayak . This is a very useful feature when you have limited VRAM on each single GPU. We have added it on our InstantStyle.

Tweet media one

@RisingSayak

Sayak Paul

29 days

We're introducing experimental support for `device_map` in Diffusers 🧪 If you have multiple GPUs you want to use to distribute the pipeline models, you can do so. Additionally, this becomes more useful when you have multiple low-VRAM GPUs. Docs ⬇️ 1/4

Tweet media one

4

18

117

0

1

5

@Haofan_Wang

Frank (Haofan) Wang

4 months

@airesearch12 @_akhaliq Possibly, but we don't use any names in the prompt. To make the results more convincing, we also showed the results of ourselves as nobody.

1

0

5

@Haofan_Wang

Frank (Haofan) Wang

30 days

Looking forward to see it on Face-to-All.

@multimodalart

apolinario (multimodal.art)

30 days

InstantStyle demo is out! Upload the picture of an image and whatever you generate will come out in that style You can choose style only blocks or style+layout! 🔥

2

5

50

0

1

5

@Haofan_Wang

Frank (Haofan) Wang

4 months

@ylecun Here is Tony LeCun, enjoy❤️

Tweet media one

0

0

4

@Haofan_Wang

Frank (Haofan) Wang

3 months

Really handsome!

@Xianbao_QIAN

Tiezhen WANG

3 months

The same prompt doesn't work for me. 🤣 This is me "after HF sell or IPO". I don't like oil head / suits though.

Tweet media one

1

1

5

1

0

4

@Haofan_Wang

Frank (Haofan) Wang

3 months

Now, you can use more SDXL base models. Try it online!

@zsakib_

Sakib

3 months

instant-id now lets you choose sdxl base weights 😎

Tweet media one

Tweet media two

Tweet media three

Tweet media four

2

3

16

0

1

4

@Haofan_Wang

Frank (Haofan) Wang

4 months

@0xStacking @cocktailpeanut @_akhaliq There is an opening issue for this feature

Implementing InstantID into Fooocus · Issue #2025 · lllyasviel/Fooocus

InstantID was just released to public and it could be amazing addition to Advanced Fooocus features. What is InstantID: Zero-shot Identity-Preserving Generation in Seconds. InstantID is a new state...

0

0

4

@Haofan_Wang

Frank (Haofan) Wang

9 months

Experimental result of AnimateDiff given with the first and last frames. Credit to our great interns.

3

0

4

@Haofan_Wang

Frank (Haofan) Wang

4 months

🩷

@camenduru

camenduru

4 months

🖼 InstantID: Zero-shot Identity-Preserving Generation in Seconds 🔥 Jupyter Notebook 🥳 Thanks to Qixun Wang ❤ Xu Bai ❤ @Haofan_Wang ❤ Zekui Qin ❤ Anthony Chen ❤ 🌐page: 📄paper: 🧬code: 🍊jupyter by…

Tweet media one

6

32

177

0

0

4

@Haofan_Wang

Frank (Haofan) Wang

1 month

@op7418 因为开始预告说明已经弄得差不多了😃

1

0

4

@Haofan_Wang

Frank (Haofan) Wang

3 months

Update: this website has ceased operation and the domain name will be transferred to us later.

@Haofan_Wang

Frank (Haofan) Wang

3 months

We clarify that is quite misleading but not authorized and has never contacted us for official cooperation. Please pay attention to your personal privacy. We currently only have a project page, Github page, and a huggingface spaces demo. #InstantID

1

4

34

0

0

4

@Haofan_Wang

Frank (Haofan) Wang

4 months

Update about CVPR 2024: 1. One paper with 4(5), 4(3), 3(3). Hope to see you in Seattle if accepted. 2. One paper got desk reject, as our 1st author forgot to withdraw it from AAAI. But, 5 papers submitted to ICML this year, all thanks to my great interns.

0

0

4

@Haofan_Wang

Frank (Haofan) Wang

4 months

Thanks for supporting InstantID in ComfyUI, excellent job!

@ZHOZHO672070

-Zho-

4 months

🤣InstantID in ComfyUI 来啦！仅需一张图就可实现高质量的角色保持！多种风格随心变！ 1⃣模块化更高效，同时支持本地、hub模型 2⃣9种风格随心选，还可与PhotoMaker Styler通用 3⃣3种工作流：特意增加了配合ArtGallery的艺术可视化工作流，助你畅游艺术世界项目地址：

12

41

175

0

0

4

@Haofan_Wang

Frank (Haofan) Wang

4 months

@SimianLuo I love your work, whether it gets accepted into any conference or not, bro. A similar case is IP-Adapter, they are both elegant and well-known works. Time will tell.

0

0

3

@Haofan_Wang

Frank (Haofan) Wang

3 months

Testing.

Tweet media one

@Haofan_Wang

Frank (Haofan) Wang

3 months

MultiControlNet will be supported in InstantID very soon.

1

9

71

0

0

4