Colorize and Restore Old Images

Подписаться 1,2 тыс.

Просмотров 4 тыс.

50% 1

This video provides a guide for colorizing and restoring old images using Unsampling and ControlNets in ComfyUI with Stable Diffusion. This is a follow-up of my video on "Reimagining" images and uses the same workflow with a few tweaks. For a more in-depth look at how that workflow works, you can watch that video here: • Reimagine Any Image in...
Links
Custom Workflow: comfyworkflows.com/workflows/...
ContolNet Aux Preprocessors: github.com/Fannovel16/comfyui...
ComfyUI Noise Nodes: github.com/BlenderNeko/ComfyU...
Stable Diffusion 1.5: huggingface.co/runwayml/stabl...
Improved VAE: huggingface.co/stabilityai/sd...
ControlNet Models: huggingface.co/comfyanonymous...
Chapters
0:00 Overview
0:12 "Reimagining" Recap
0:25 Viewer Suggestion
0:50 Workflow Walkthrough
2:09 Things to Try
2:38 Wrap Up
*Narration created using Elevenlabs' SpeechToSpeech Synthesis

Хобби

Опубликовано:

11 июл 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 27

@Metalman750BC 5 месяцев назад

Wow .. very impressive!

@uk3dcom 7 месяцев назад

Thanks for the mention. 😁 I'm now playing with image to image but keeping the face recognizable when applying various art styles. Surprisingly tricky.

@HowDoTutorials 7 месяцев назад

I have noticed that the image resolution can have a big effect on final result. It seems like 768-1024 is the sweet spot.

@randomscandinavian6094 7 месяцев назад

This is very interesting! I’ve failed to colorize well with Stable diffusion myself but this tempts me to try it again. I’m just beginning to figure out ComfyUI though so I need to get further into that. It’s a bit overwhelming. What I’ve done before before is to use an online free colorization page and then enlarge and overlay the result as a color layer over the full size original in Photoshop to use as a base. You can’t see the pixels in the color transparency blending mode. And then basically hand coloring over that with brushes in the soft light blending mode. And adding some color noise over that to make it look like early color film. Anyway… thank you! So little info out there on this type of work.

@HowDoTutorials 7 месяцев назад

This will save you so much time! ComfyUI can definitely be intimidating at first, but once you get the hang of it, it’s a breeze. I’m working on a “How to Get Started” video that should be out tonight or tomorrow. That should help get you going. 😁

@randomscandinavian6094 7 месяцев назад

@@HowDoTutorials that sounds great!

@HowDoTutorials 7 месяцев назад

A little later than anticipated, but here it is! ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-guqwYQNWTuk.html

@randomscandinavian6094 7 месяцев назад

Great!

@aliyilmaz852 2 месяца назад

Amazing share! Thanks again, I am old and have lots of b/w photos. Will give it a try. And If I can I will try to swap the faces with current ones :) Maybe you can teach us how to swap faces, will definitely appreciate it!

@nawafalhinai1643 Месяц назад

were should i put all that files in the links?

@pn4960 7 месяцев назад

Wow amazing ! Do you think it works for converting b&w drawing into full color ? EDIT : I've made it work for a drawing, it's not as performant but very cool non the less

@HowDoTutorials 7 месяцев назад

Ooh! I’ll have to mess around with that a bit.

@davewills6121 6 месяцев назад

Question!! My son is doing a project on ''The Mesolithic period'', he wants to use some examples of AI art for his talk on the subject. The problem is, all my attempts using Comfyui are a mutated group of cavemen, he's nervous as it is, he said my AI art will make him the laughing stock. So, are there any simple PROMPTS that i could use to produce good results?. cheers

@HowDoTutorials 6 месяцев назад

I would try SDXL models, as earlier models can struggle with multiple subjects. You might also try using ControlNets with OpenPose. If time is of the essence and you don’t have an SDXL model setup, you can try using SDXL here: clipdrop.co/stable-diffusion?. Best of luck!

@1ststepmedia105 7 месяцев назад

Each time I use your work flow I get the following error message: Error occurred when executing BNK_Unsampler: 'ModelPatcher' object has no attribute 'memory_required' any help?

@HowDoTutorials 7 месяцев назад

I haven't run into that issue personally and I'm struggling to find solutions online. If all else fails, you can try installing ComfyUI from scratch: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-guqwYQNWTuk.html

@HowDoTutorials 7 месяцев назад

If I had to take a guess, it might be caused by not having enough RAM/VRAM. You could try using a lower resolution version of the image and see if that helps.

@krystiankrysti1396 7 месяцев назад

thres hould be one final node that merges luminosity from original with colors from restore, i do that manyually

@HowDoTutorials 7 месяцев назад

That’s a great idea if you’re wanting to restrict changes to color only! My idea with this was to also help with graininess and other issues with image quality to “restore” as well as colorize, which replacing the original luminosity would undo.

@squirrelhallowino29 4 месяца назад

I have some cool sketches i've made i wonder if this will work to colorize that and well just give the AI some freedom to bring them to life~~

@HowDoTutorials 4 месяца назад

This should work well for that!

@Marty 4 месяца назад

The voice sounds also ai generated :D is it?

@HowDoTutorials 4 месяца назад

It’s generated using elevenlabs speech to speech, so I record narration with my own voice and then run it through the AI to refine it.

@bandinopla 6 месяцев назад

this is pure luck... i've tried the workflow with another pic and it does create a colored photo but with many things diferent from the original. Especially clothing. But is it interesting tho...

@HowDoTutorials 6 месяцев назад

It definitely can feel that way, but once you mess around with the workflow a bit, you'll start to get a feel for how to adjust settings for each image. The most important things to adjust are: 1. ControlNet strengths/activation: Try adjusting the strength of the controlnets. You might even try bypassing them entirely to start and add back as needed. Different images are affected differently and certain controlnets are better for certain purposes. (i.e. Lineart is best for creating illustrated styles) 2. Resolution: This workflow is sensitive to resolution and can give poor results outside of a certain range. Generally speaking, you want both dimensions of the source image to be between 512-1024. Going a little above that range tends to be okay, but going below creates consistently poor results. Best of luck!

@Rixynator 5 месяцев назад

for some reason it takes stupidly long to process. doesn't make sense.

@HowDoTutorials 5 месяцев назад

Is it only happening with this workflow? If not, and you have an NVIDIA card, make sure you’re clicking on the correct .bat file. Otherwise, it will use the CPU which will be quite slow unfortunately.