[Old Workflow] Create art with AI using InvokeAI Unified Canvas

In this video I show how to make art with AI by creating two new images from sketch (digital painting) to final render using InvokeAI's new Unified Canvas and Stable Diffusion.
Model Used:
huggingface.co...
Prompt:
red panda knight holding pole axe, Banner Saga and Artgerm, 8k, highly detailed, exquisite [grainy out bad+ poor low basic amateur+ blurry haze worst juvenile unprofessional+ failure crayon beginner] [oil thousand hands dance car bird plane city house sports text watermark]
Invoke AI Tutorials:
Unified Canvas - • InvokeAI - Unified Can...
Setup Guide - • Best NEW AI tool? Invo...

[@3oxisprimus848]

THIS is a BEAUTIFUL example of an Artist using AI as.a. TOOL! 💡

[@coreym162]

That's all it ever was and all it can be used as. A tool.

[@TechCarnivore1]

This is amazing. Shows the sheer potential of these open source tools. So glad everyone has access to these tools and the ability to train their own models.

[@albertbozesan]

Incredible. I love how close you were able to keep it to your original sketch. You had a very clear plan from the start, enough artistic skill to put it on the page, and then used the AI as the tool it was meant to be. Can't wait to try this out!

[@pjgalbraith]

Thanks Albert keen to see how you get on! The new unified canvas is a nice glimpse at a tailored UI for img2img.

[@albertbozesan]

@@pjgalbraith let me polish up my sketching skills for a few years, don’t wait up!

[@RamiJames]

I can't describe how excited I am by this tech

[@OtherWorldExplorers]

If there was ever any argument for AI being art this would be it. You started out with a panda and you ended up with, what most of us would agree looks like a very distinguished fox. If that doesn't underscore the creativity that it takes, let's not forget that you having to go through and manually manipulate the image time and again time, adding the gauntlet, the background fire and smoke elements, and going into Photoshop and making mask to make certain elements come forward. If that doesn't underscore creativity I don't know what does.

[@sociallyresponsiblexenomor7608]

Fantastic tool for artists who have the technical and artistic fundamentals- because they can paint over, refine, fix, adjust. I think AI assisted artists will be a great thing for the first while.

[@milestrombley1466]

Proof artists and ai can work together.

[@alessiosanto1296]

I test the image prompt on an empty canvas first. It saved me a lot of time with img2img, when the prompt resembles the input image well (small details, perspective, …).

[@pjgalbraith]

That's a good tip! At the very least to get the right style you want.

[@aiartbx]

Real artists needs to watch this video. As amazing as this tech is I'm pretty sure a good artist can paint something like this manually in the same amount or less time. This is the type of production workflow ai would be adopted and ai isn't perfect as this video shows. It's a very helpful tool indeed but its not going to kill all the artist jobs.

[@pjgalbraith]

I'm sure some artists could but the good thing is that this workflow scales to a broad range of rendering styles and subjects, and will continue to get faster. It never was going to replace artists enmasse for a whole host of reasons, some due to limitations in the tech, others more fundamental. Unfortunately the internet tends towards polarisation. I know a lot of artists using it but the stigma around it means they don't publicly share their workflow which is a shame.

[@aiartbx]

@Patrick Galbraith I agree. Also there are still alot more artists that don't even know or have used it. You are definitely the only person who shares production worthy workflows on KZbin so thanks for that.

[@dissonancE..]

This has been my workflow from the beginning. This and using it to generate stocks (say a cave or textures) for photobashing without needing to spend hours collecting the right stocks. I also love passing my old art through as well to give it a fresh ai update and see what I can get. Another great video Patrick 👍 Have you checked out the noise painting technique by Lynn Cole?

[@pjgalbraith]

Yeah it's amazing! That's why I started making videos to show how great these tools are. I haven't heard of Lynn Cole I'll check it out thanks for the heads up.

[@infographie]

Excellent

[@PriestessOfDada]

Hey patrick, killin it as usual in this one

[@_skiel]

3min into the video and already sub'ed ! great tutorial - thank you !

[@pjgalbraith]

Awesome, thank you!

[@orionsbelt927]

Yes this is exactly what i was looking for, thankyou for putting the video out for it

[@pjgalbraith]

No problem 👍

[@googleyoutubechannel8554]

Good to see what the state of the art workflow for cutting edge AI is like for a professional artist. As an amateur this is kind of what I expected.... SD is basically a fast, enthusiastic, but shoddy, colorist with poor sense of depth (amazing it gets any at all from only raster understanding of world)

[@pjgalbraith]

It's improved a lot since this video now that ControlNet is available.

[@gaetanoisgro6710]

Having this working with AMD GPUs would be amazing, such a funny tool

[@ddiva1973]

Fantastic!!!!!!!

[@AscendantStoic]

This is the way ;)

[@wizards-themagicalconcert5048]

Great content mate,well done,subbed !

[@gohan2091]

I've been trying this with photo realistic portraits. I'm no where near as comfortable as you are here but I'm learning slowly. Your video will certainly help. Please consider doing a video on non cartoony images such as photo realistic.

[@pjgalbraith]

Realism is definitely less forgiving. I'd recommend using a photographic model. Check out civitai for models.

[@gohan2091]

@@pjgalbraith civitai? OK I'll check it out. Have you heard of realistic vision model? I've been using that. It's pretty nice for photo realistic images. Do you know how it compares to civitai? Unfortunately I have an AMD GPU so can't really experiment with different models as every generation takes so long. I have found an Invoke AI Web UI though which I can use but it's out of date

[@pjgalbraith]

@@gohan2091 civitai is a place to download models. But if you aren't running locally you may be a bit limited. See if you have any way to use the noiseoffset Lora which helps with contrast and low lighting.

[@loremipsumexe]

a wide angle shot with foreshortening may be easier to produce by first overshooting the margins so as not to mess with the silhouette of the subject during the iteration process, and then cropping for a final result.

[@pjgalbraith]

That's a great idea! I'll try that next time.

[@gogidolim]

This is exactly what I want. But I want to create something like busy wet market scene without any central characters.

[@nuvotion-live]

Great tutorial. Now that photoshop has some of these image synthesis features do you find yourself using Invoke less?

[@coreym162]

This stuff will never be on autopilot like many A.I naysayers suggest. There is no one button answer and never will be specific to the user's preferences without user input.

[@mitrooper]

Death of true art.

[@coreym162]

Gatekeeper...

[@Thesamurai1999]

Would you consider making a tutorial on controlnet?

[@pjgalbraith]

It's coming. I have a lot of footage but no time to edit right now unfortunately.

[@KaiBuskirk]

Wow!!! danke ŌPŪ/Kai

[@monstercolorfunco4391]

that's weird, the new version doesn't even have image to image strength, just updated today, unusable no proper settings. nice vid i wish it was the same here.

[@madjunir]

Really cool. Am wondering if these final images would be acceptable for copyrighting since you did the initial sketches and just tweaked them with the AI. I read a few places that if you let the AI generate the full image in SD, MIDJOURNEY, etc. then the copyright office will not accept them.

[@pjgalbraith]

No idea but you would think so given the design, composition, pose, etc... were preconceived. I think the main issue is with txt2img.

[@DFMoray]

How long do these take verses just painting it yourself?

[@Amelia_PC]

Is there a tensor version to download instead a ckpt file? Tensor is safer.

[@pjgalbraith]

Yeah it's in the same link I put in the description

[@GS195]

Awesome. I wish I had a 24gb gpu to be able to use it though.😢

[@pjgalbraith]

It only needs 4GB according to the docs

[@KevinTurner-aka-keturn]

4 GB might be cutting it too close for this kind of inpainting workflow, but 8 GB is comfortable enough.

[@funicon3689]

im running it on a 2070S

[@tonycheezal5650]

Hi, i just joined invoke ai recently, and i see your links to your model has quite a lot to download. Normally the tutorial told me to download from civitai, and the file is just safetensors, 1 file only. May i know how do i install your model on invoke ai?

[@karter.design]

Amazing work, the final result was faithful to the original draft. Also, this model is very inpaint aware. Can you share the model link and the invoke ai yaml files ?

[@pjgalbraith]

Thanks! Check the description for the model link. Yaml is the same as standard 1.5 model just update the path to the ckpt and use the standard vae.

[@iz5808]

Can't to get invokeai to run (cuda out of memory). It might be you know any alternatives with the same functionality? It seems A1111 is a bit off with generating images from sketches :(

[@pjgalbraith]

You can do a similar workflow using Automatic1111 the latest version even has a sketch mode, but you may need to do more manual merging in a photo editing program.

[@pjgalbraith]

You could also try this Photoshop plugin github.com/AbdullahAlfaraj/Auto-Photoshop-StableDiffusion-Plugin

[@TheRogueJedii]

I tried doing this with a sketch I made of a castle, but the end result looks nothing like my original sketch. Help please!

[@pjgalbraith]

I would recommend using the ControlNet extension with automatic 1111 webui. It has a sketch to image model. I'll be doing a new video soon on my current workflow.

[@HalkerVeil]

And I'm supposed to believe that AI does art for me. This is about as much work as stock image manipulation. OOH but all these so called artists were fine with that. I never thought in our age we would find another group of Luddites by definition with their sophism.

[@pjgalbraith]

Unfortunately photomanipulation also took time to find acceptance and for a long time was considered cheating. There is definitely a culture clash happening, in software engineering automation of work is highly valued by the community. But artists embrace and highly value what engineers would call "toil". The journey to being an artist is difficult and often unclear and it takes a long time to be able to create work of a professional standard. I think the reaction by many artists while disappointing in many ways is natural and expected. But I hope by showing new ways these amazing tools can be used the word will slowly get out and we can bridge the divide.

[@HalkerVeil]

@@pjgalbraith Well nobody is stopping them from toiling away in the dirt really. Doesn't mean they get to decide for everyone that we can't use automation in every other aspect in the world. If governments and laws back them up over progress, then there is going to be another kind of culture clash on the way.

[@chariots8x230]

This AI looks really amazing! Thank you for sharing this video 💖 Would it also be possible to train the AI using images of your character, and then have the AI reproduce your character in different poses and scenes? Can you copy a pose from another image into your custom character? Can you pose two or more of your custom characters together? I haven’t seen any videos so far that show AI being used to copy complex poses from images, or group poses that involve at least two characters. I’m hoping it won’t be long before this is possible. 😊

[@pjgalbraith]

It is possible to train the AI on your characters using Dreambooth I have a video on how to do it on my channel, you may want to base the training on Anything v3 if it is anime style. Regarding poses once you've got a model with your character you can pose it, but you can't copy a pose except by doing Img2Img on the image you want to copy the pose from. So it's not as simple as saying copy this pose, but it is possible with a bit of work. Over time this kind of thing will become a lot easier with zero shot training (probably this year sometime). Hopefully that helps!

[@pjgalbraith]

I have actually done this before using a character my brother designed. The cool thing is you can also see your character in different ways like as a real person, or as a Pixar character. It's pretty fun to do, I might make a video about this process soon.

[@chariots8x230]

@@pjgalbraith Oh I see. I’m looking to get exact poses, so I’ll be waiting for the feature that will let me copy poses from images, and apply those poses to my characters.

[@chariots8x230]

@@pjgalbraith It’s definitely really cool to transform an original character into different styles. I’d love to see a realistic version of my characters, so I can imagine what they might look like as real people. It would be very interesting to see a video on this process :)

[@ay.l4998]

Interesante

[@mischamim744]

A lot of people are PRing Midjourney. I tried it, but I didn't like the fact that you can't edit it like InvokeAI - I haven't installed it yet, but I want to do it later.

[@pjgalbraith]

Yeah Midjourney is easier to prompt but InvokeAI and the Stable Diffusion ecosystem is a lot better for creative workflows and gives more control.

[@user-nc2hs4rp7l]

5:35쯤 부터 ==> 핵심!

[@bambinoesu]

it took me 1hr35mins to generate 1 image with invoke ai, with my intel had graphics card. I'm crying 😭. it's not efficient enough

[@pjgalbraith]

Are you sure it's using the GPU? Sounds like it's using CPU for inference.

[@gogi931]

Can we conclude that ai speeds up artists work?

[@MisterBroad]

How about upgrading your drawings by talent and hard work?

[@pjgalbraith]

🤔

[@MisakaMikotoLuv]

if thats how you feel, then i hope you only use cave walls and rocks when drawing. technology is off limits for you apparently

[@MisterBroad]

@@MisakaMikotoLuv Oh, you think honing your skills without taking lazy shortcuts makes you a troglodyte? Where did you study again?