🎥｜animation

[fickle sage]

has it stayed the same since?

[twin turtle]

hey can someone tell me where I can get tutorials to generate videos using AI and music ?

[neat jewel]

medjed has a youtube tutorial

google it

[neat jewel]

No, at one point I reinstalled SD and deforum tho

[glacial rock]

I couldn't find good guides online, so I've written my own base on stuff I have figured out in the last few weeks. And yes, there are unique steps to my process I will not share, such as my model which I need to preserve future film work. You can probably see that it's inherited from protogen 😄 (images in order below: Anime, SDv1.5, Protogen, Temporal's). However- you can get similar outputs using different models with the configuration specified in the example prompt. Attached still from a few different models for example which you can get using my guide without my model. Other unique steps are really just applying some filmmaking techniques I've learned through the years.

If you read in the readme, my stated goals are targeting filmmaking, marketing, and more. I am trying to help push the boundaries of coherent animations frame to frame, while also having a degree of control in the outputs. So this is more of part 1 and it's a hobby project! Part 2 will probably talk about some of the tools in the tools folder and how to use them to speed up your workflow or large inputs (thousands of frames). However the tools have readmes and can be used now.

There will be more guides to follow as I push updates to the repo for those that wish to share, so I'd watch it your interested. Taking contributions as well. I am working on a more intricate layered setup right now. Will see how it goes! 😄

Repo Link since the chat has been moving on: https://github.com/TaylorTheDeveloper/Stable-Diffusion-Animation-Processing

[neat jewel]

@glacial rock all good

[bold bough]

https://www.tiktok.com/@themajesticmaidens/video/7189237282392624386

[glacial rock]

Pushed and update to the repo with more details on long video processing as well as fixed some typos in the readme.

[soft bone]

Very useful info, thank you. I have not used flowframes, and I really did not understand this

Output FPS - Recommend 6, 8, or 12. You may need to play with this, but try to match your input frame rate in the end. This will translate to 12, 16, or 24 output FPS at Normal Speed
So suppose you have 24 FPS output from SD and you want to interpolate it to say 30 FPS - wouldn't you just enter 30 as your output FPS? or is there some further calculation involved?

[glacial rock]

I actually just picked up flow frames for the first time about a week ago so I am still learning it. I plan to update this section when I have a better handle on it as well. In general, if you have 24 FPS input and extract all the frames, you will have 24 FPS output. If you flow frame it with a factor of 2X, then your interpolated result with have 48 FPS.

So for 24 FPS footage and if I export each frame, I generally choose 12 x 2 to get 24 FPS output again. However it is slightly slow motion. Then I will speed it up 200% percent in a video editor- like ExGif's speed change utility or Adobe Premiere/Davinci Resolve.

Another way to do it, is to take the 24 FPS output and use the /tools/frameReducer.py to halve the number of frames to make it 12 FPS- but you will actually loose frame information this way- so I tend to prefer the first way otherwise you get stop motion looking output.

Example of stop motion like output- https://github.com/TaylorTheDeveloper/Stable-Diffusion-Animation-Processing/blob/main/documentation/images/paintingAnimation/1024-002.gif

[soft bone]

Ok, I better download it and try this out in practice, but thanks for writing it out, it is a little counter intuitive 😅

[glacial rock]

I'm going to make a note to make sure I make this section clearer in future updates

🙂

I also may be a little off in my explanation- I'm basically leaning on my film knowledge here and there may be another nuance to it I haven't understood yet.

So, here's a grain of salt 😄

[soft bone]

When I think about it, if you want to interpolate and keep the clip at same length, it makes more sense

[glacial rock]

yeah and in my case specifically, if there is a MAJOR issue in a single occasional frame, it makes it easier to delete it because I have additional frames. Less frames give you less to work with in the end. However, my goal is to avoid discarding frames.

[plucky plover]

your zoom is probably 0:(0). change it to 0:(1) that is no zoom

THERMAL VISION https://youtu.be/XuuDMiUN5tk

made with a custom model : )

[soft bone]

One more thing, when you extract the img sequence with ffmpeg -r 24 raw/out%03d.png -i input.mp4, does the -r flag really do anything useful? It's just a bunch of images, right?

[glacial rock]

-r is important because it lets you determine output frame rate

So no matter what the input video frame rate is, ffmpeg normalizes it to -r value befor rendering it out to you

so if you have 120 FPS footage, and want 12 FPS out, -r is a good flag to use

You can also specify -r 24/2, which is the same as -r 12

[soft bone]

Well I have never used that flag, but I always end up with the right amount of images in the folder. I suppose it detects the frame rate automatically. When I ffmpeg it back to mp4 using the reverse operation, then -r comes to play ffmpeg -r 24 -i raw/%04d.png video.mp4

[fickle sage]

Wow yes it is! Rerunning now thank you

[glacial rock]

yeah it will otherwise use the source input frame rate by default

so 30 FPS in 30 FPS out

[fickle sage]

Thankyou thats working for me now!

[plucky plover]

happy to help 🙂 also if you run into any more problems deforum discord has a bunch of support channels!

[fickle sage]

yh found it just now. Thanks : )

[past stone]

https://youtu.be/_ZDMQmJDJRY

[glacial rock]

Latest test refining approach for a larger project.

and for whatever reason my Adobe install is fubar, trying to fix it so I can cut the full music video export I promised (raw and unrefined based on my earlier approach)

[gloomy coral]

Hey everyone! It's great to see so much r&d here, there's so much to learn!

I tried looping a video recently using the seed schedule but it doesn't seem to work, in the 2D and 3D mode, whilst using the same prompt at the beginning and the end along with the respective seed. Is there a workaround for this?

[left raven]

Does anyone know when the DreamStudio Pro release will happen? Emad said it would be this month. Anyone has some information about it? Appreciate

[pallid sluice]

https://giphy.com/gifs/rezdogsfxonhulu-xgfwp70oezGcqJxTjK

[faint rivet]

https://www.youtube.com/shorts/hL079vS9Rvc

[shrewd dirge]

Anyone know how to use your own image in deforum generated video auto1111 and please no old tutorials there outdated now

[glacial igloo]

https://www.tiktok.com/@exosfeer/video/7181867444955254021

[daring warren]

https://www.instagram.com/p/CnnCrMmh6yO/

nofacekolektiv

Testing, testing, testing👨🏻‍💻..
.
.
.
.
.
.
.
.
.
.
.

ai

Instagram

[noble quail]

hey guys! i am working on a AI generated educational history series called AI History that using deform and just posted my first video! It is on the Samurai! https://www.youtube.com/watch?v=Vvukxgxi5LA&ab_channel=AIHistory

[faint harness]

https://youtu.be/lBphYWiaUb8

[faint rivet]

https://www.youtube.com/watch?v=YEWjTSIFm_k

[foggy gyro]

This is very cool. Do you have any info on how you generated the different styles? Doesn't seem like stable diffusion as the frames are very smooth.

[foggy gyro]

This is nice! How many frames per second are you generating?

[noble quail]

its about 30fps

[hallow bramble]

GM!

https://twitter.com/iastitraia/status/1616313595712921601

iastitraia (@iastitraia)

https://t.co/dg269Afclq

▶ Play video

Twitter · Jan 20, 2023, 5:56 AM

[plucky plover]

used my oculus Rift VR headset as a controller for the camera movements https://youtu.be/XJRUCFGD2g0

[rancid bramble]

is this 512 res ?

[plucky plover]

832 x 448 upscaled 5 times to 4k

[rancid bramble]

what gpu use

[plucky plover]

3090

[fickle sage]

https://twitter.com/Art_Love_xx/status/1616495124510347271?s=20&t=f557-gzUH2r5F2BR27-gXA

Arthur (@Art_Love_xx)

Check out my latest #AIanimation for my beat 'Medicine Man'

Made using #deforum, bringing the music to life in a whole new way!

🎹🖌️🎨🎷

#music #AI #stablediffusionart @RTFKTcreators

▶ Play video

Twitter · Jan 20, 2023, 5:57 PM

[balmy stag]

Just put up a free, super easy to use animation website using SD https://twitter.com/genmoai/status/1616482565992636416 - try it out at https://alpha.genmo.ai/ 🙂

Genmo AI (@genmoai)

Announcing Genmo Video, a generative media platform with a new text-to-video model that can generate immersive live artwork from any prompt or any image.
What will you create? 🎨▶️

Free public access: https://t.co/jNdu4QQIcq
Discord: https://t.co/guXCPHBQo7

👇1/n

▶ Play video

Twitter · Jan 20, 2023, 5:07 PM

[plucky plover]

https://www.youtube.com/watch?v=QRl73WPFxqI matrix mode

[mental pecan]

some new tests. and fails..

[ornate pewter]

The colors and everything. I like these animation clips so much.

[mental pecan]

thanks.. needed a break for a bit but back to my animations again.. most go on my IG if you want more..

[ornate pewter]

No thanks

[foggy gyro]

What kind of video did you use for the source? Live footage or a pan of a 3d object?

[mental pecan]

@foggy gyro this was pure SD.. not video init.. took a bit to try to have it rotate around with the proper keyframes

[foggy gyro]

Wow! Yeah rotating around a model is hard. Nice work. What workflow did you use? Very impressive.

[mental pecan]

yea it was a lot of trial and error.. I've beeing using deforum in auto1111, but been doing AI video stuff for almost 1 1/2 years give or take.. also working on a keyframe visualizer i blender but that progress is going slow..

@foggy gyro here are some of the keyframes used if you want to test it out on an object.. im constatly tweaking stuff so if you come up with something neat let me know.. "translation_x": "0:(2.5), 200:(2.5), 201:(-2.5)", "translation_y": "0:(0)", "translation_z": "0:((0.125*(cos(120/153.141*t/45))+0.0)), 380:((0.125*(cos(120/153.141*t/45))+0.0)), 381:(1)", "rotation_3d_x": "0:(0)", "rotation_3d_y": "0:(-0.50), 200:(-0.50), 201:(0.50)", "rotation_3d_z": "0:(0),410:(0), 420:(0.5)",

[waxen lion]

https://youtu.be/Vy9Qd6HVK9M

[glacial rock]

I really like this- great work!

Anyone ever check out 2 minute papers?
https://youtu.be/uzF6CTtjn-g

Another recent one:
https://youtu.be/C9LDMzMRZv8

[glacial rock]

I had some fun with this one, Audio on!
https://youtu.be/_rv1TKFg8WY

[rapid moth]

@waxen lion elite as always

[balmy stag]

whoa

[full trail]

https://vm.tiktok.com/ZMYdjUocd/

[full trail]

Fashion Evolution - animation with SD
https://www.tiktok.com/@vblank_digital/video/7191163439778204933

[dense crater]

https://www.youtube.com/watch?v=3Wo15vcBZ8k

[safe oriole]

my test deforum videos playlist https://www.youtube.com/watch?v=fhrGPEmG49o&list=PLRQLFXfQy4pAu3NEd9FSmjNd3wW8XVFEh

[hallow bramble]

https://twitter.com/iastitraia/status/1617092888378097667

iastitraia (@iastitraia)

How to create audio-reactive animations such as this one? 🤖🎵🌌

Here's a step-by-step guide: 🗒️

1. Choose a song
2. Split it into audio tracks
3. Generate and copy keyframes
4. Paste them into @deforum_art notebook
5. Render the AI-magic

More in the thread below🧵⬇️

▶ Play video

Twitter · Jan 22, 2023, 9:32 AM

[celest delta]

https://youtu.be/KSssj5Bd8Eo

[glacial rock]

I'm curious what folks think of the style here.

[stable sky]

https://www.youtube.com/watch?v=akJIQjO1O1Y Around 5000 minutes of image generation in deforum sd1.5 all prompts at 960x540, upscaled in Topaz Video AI and editing in Premeire Pro - I hope you like it and im always open for feedback and ways to improve!

[unborn quest]

thank you, this is a VToonify you can chek this here https://github.com/williamyang1991/VToonify

[dull delta]

A man and lion talking each other on mobile phone#

[soft bone]

I've now used FlowFrames. Did 2x to 25fps video, and the result 50fps looks very good.
I wanted to ask you about this

So for 24 FPS footage and if I export each frame, I generally choose 12 x 2 to get 24 FPS output again. However it is slightly slow motion. Then I will speed it up 200% percent in a video editor- like ExGif's speed change utility or Adobe Premiere/Davinci Resolve.

Wouldn't you end up with 12fps, after speeding the 24fps interpolated by 200%? Why, what benefit is there?

[glacial rock]

I was just getting to know Flow Frame at the time and I went from a place with very random outputs, to less random outputs. I wasn't very worries about the timing of things, and more interested in having extra in-between raw frames in case I needed to remove some. This was useful at the start, and in this case, using Flow Frames was awesome. But now that my outputs are more stable frame by frame like in the video I shared above. I have navigated away from Flow Frames and I am using Premiere Pro.

For the clip I shared above, the foreground was 1080p in and 1080p out, with low noise. The backgrounds were 540p and intentionally had higher noise, because it looked nice and took less time to render. They were composited together in Premiere + After Effects.

[glacial rock]

I'm almost done uploading the music video I started working on a few weeks ago. It didn't come out as good as the test film I shared above, but there were many challenging scenes and un-ideal lighting conditions in the original Music Video. I learned a lot just comparing the source film to the output from diffusion, and was apply to apply that to the test film above which came out pretty good.

Just finished uploading! This is based on my (now outdated) process from a few weeks ago.
https://youtu.be/FgOZyv4du4g

And here is the original video:
https://www.youtube.com/watch?v=klGNDDMfOeI

It took about 5 days to render at 1080p, using the same prompt base I shared here: https://github.com/TaylorTheDeveloper/Stable-Diffusion-Animation-Processing/blob/main/prompts/paintingAnimation/.prompt

[pseudo plaza]

I finally finished editing this guide for making really smooth, non-jittery img2img animations.
The tutorial uses After effects to generate in-between frames, but something like Flowframes works just as well.

If you have any interest at all in making img2img animations I highly recommend you give it a watch. Good luck! 😄
https://www.youtube.com/watch?v=Z6pR7IB9in0

[balmy crag]

I wonder if same will work in Divinci?

[pseudo plaza]

Probably! Any program able to generate in-between frames will do. Flowframes is another free alternative 🙂

[balmy crag]

nice

[hushed schooner]

Something I am working on 😄

[silent magnet]

I tried it for a bit tonight. Finally found some time. How do you animate it in automatic1111? I have the UI and can get a protogen model, but didn't get how to do it on a batch. Maybe overlooked smth.

[balmy crag]

@silent magnet there are many ways, deforum, the animator script, jinja2 templates, img2img overlaying, to name a few

[foggy imp]

Oh! I love all the little details, color, and the organic look along with the shift in hue.

[serene veldt]

https://www.youtube.com/watch?v=3hT_i2sSmdU

[balmy crag]

the eyes moving around weirded me out 😄

[foggy imp]

It's watching you!

[balmy crag]

earth, the television channel

[foggy imp]

It certainly has an unnerving feeling, but it kind of works--makes me think of a wood carving, really.

[balmy crag]

that's trippy, subbed love the longer videos

[serene veldt]

cool 😎 thanks brother.

[serene veldt]

yeah, davinci resolves comes with 4 different "optical flow" algorithms. (btw the cpu heavier/slower ones don't always look better, it really depends on your source material, so it often pays to experiment)

[faint rivet]

Deforum Hybrid Video - Freakazoid Alien Wave
Rendered 1024x1024, upsampled to 4K 60fps
https://www.youtube.com/watch?v=bQR2kF4FiTI

[faint rivet]

My Bleeble-Blorp Don't Jiggle Jiggle

[glacial rock]

Img2Img batch mode is here

But to complete an animation sequence, you need to put images into an editor or use flow frames

[grim crown]

Messing around with my Stable Diffusion home-made app. Here is a random walk though latent space. The direction of travel is a bit too random. And it is only 7 steps per frame. I will make some better ones soon. https://www.youtube.com/watch?v=N6ZNBvNN8_8

[plucky plover]

another video using my oculus rift as camera controls !
https://youtu.be/8oJDYQkTo90

[grim crown]

That's cool. Did you record the oculus position and then render it back slower?

[subbed]

[glacial rock]

Tried a take on Family Guy with disastrous results.
https://youtu.be/AKF5x6qsXLw

[faint rivet]

Alien Musk - Let That Sink In - Deforum Hybrid Video https://youtu.be/qEDkdi-xwzw

[glossy plover]

Wow, that looks awesome. Have you used Google colab to generate this? My pc is not strong enough for this.

[hushed schooner]

Hey there, there is a deforum notebook you can use to do it. https://deforum.github.io/

@glossy plover

[glossy plover]

Oh, thank you very much. Is there a good tutorial on how to use it?

[hushed schooner]

https://youtu.be/FehfaLbTUjk

thats the one I used, but if you need any more in depth search youtube for deforum notebook

[glossy plover]

Thank you

[glossy plover]

This is gold= tnx

[hushed schooner]

its a lot of fun

[glossy plover]

Do you know if it's possible to use as start frame your own image, or is it only based on prompt?

[hushed schooner]

there should be init options somewhere

control f init

[glossy plover]

Ok, I will check

[cobalt hinge]

hey all, I made this https://youtu.be/l6GFTit7Wn0

[faint rivet]

https://www.youtube.com/watch?v=5nlVTpM8q2k

[daring warren]

https://www.instagram.com/reel/CnwhjwUhy6E/?igshid=YmMyMTA2M2Y=

[fallen frigate]

"AQUATIC TIME JUMP" follow me on twitter for more of these kinds of videos: https://twitter.com/J0HN9R1M3/status/1617548521464565761?s=20&t=PkvoaYwHIrwKPAG2kwv7Gw

30XIS PRIMUS DIFFUSIUX (@J0HN9R1M3)

*USE HEADPHONES!

"I am an Old Time traveller & this is a sojourn gone wetly wrong. "
#stablediffusion #AIart #Deforum #aiartcommunity #aianimation #animation #trendingvideo #timetravel #weirdanimation #AIArtistCommunity #AIArtworks #aianimation #art

▶ Play video

Twitter · Jan 23, 2023, 3:43 PM

[gilded vigil]

https://www.youtube.com/watch?v=qtGT9aOjRnE
Green Glitch .. some industrial techno music I made with an Stable Diffusion animation as music video

[fiery fox]

wow, I love this! is there a particular model or prompt that you're using to get images in this style?

[rose bough]

@hushed schooner

[hushed schooner]

?

[languid vale]

I recreated the ENTIRE Bee Movie with Stable Diffusion, but I can't post it on YT due to a false copyright claim! It's transformative. Twitter only lets me post 120 seconds: https://twitter.com/TheFoxipso/status/1617583357650702337

Foxipso (@TheFoxipso)

Apparently 130,447 AI-generated frames of Stable Bee Movie violates @Dreamworks' copyright? Pff. If that's not transformative I don't know what is. I have to wait for YouTube to review and surely deny my appeal. Stay tuned!

▶ Play video

Twitter · Jan 23, 2023, 6:01 PM

[serene veldt]

thx🙏 if I remember to do so i will look up the workflow next time im at my comp, almost certain though its a mixture of multiple models.

[ornate pewter]

YouTube's copyright is absolutely horrible.

[glacial rock]

@languid vale you may be able to bypass by speeding or slowing down the film by 1.05 or .95. Not sure if they account for that now or not. 🙂

[blazing star]

📝🎥
https://www.instagram.com/reel/CnwtQrdO7Oj/?igshid=NjcyZGVjMzk=

[young rock]

Hi guys, can someone please show me how I can start making cool AI videos? Is there a link to a tutorial somewhere? Thank you

[balmy crag]

here are some tutorials https://www.youtube.com/playlist?list=PLizyrpTqy7WVdTp-MPfxjXLOSlTtIDONT

[glacial rock]

@young rock I have a guide here
https://github.com/TaylorTheDeveloper/Stable-Diffusion-Animation-Processing

[mental pecan]

Trippy Travels - Dreaming Computers

[karmic atlas]

hi, we make music and than with stable diffusion videos

https://www.youtube.com/watch?v=76z1O4zZdLY
https://www.youtube.com/watch?v=YPftrJfnYco&t=11s

[queen tulip]

/create animation with a car

[agile eagle]

👀🫠
.
Deforum Txt2img, no int, no affects applied

[plucky plover]

playing with different models! here is a sketch model https://youtu.be/TLtmVivHDb8

[amber fiber]

https://tenor.com/bFGWW.gif

[amber fiber]

/tenos

[amber fiber]

https://tenor.com/NhQ5.gif

[glacial rock]

Do you have a link to where one can download the sketch model?

[plucky plover]

yes i used this one! https://civitai.com/models/1221/anime-pencil-diffusion

[wooden talon]

Sweet!

[glacial rock]

So, I am not actually very happy with the final result and don't plan to make this public.

The smoke and lighting effects, plus lower quality 70's video made this very hard to process. But I wanted to share it with the community. It took ~30 hours to render at the cost of fidelity. Full fidelity would have taken 8 days. I may revisit at another time, but for now will shelve this specific effort.

Viewer Discretion advised! Queen's Bohemian Rhapsody.

https://youtu.be/9NX7DWocWVc

[neon shard]

does anyone have good examples of animations that dont look like ur tripping out on shrooms?

[glacial rock]

Yes, I do. This is my focus right now with diffusion.
https://github.com/TaylorTheDeveloper/Stable-Diffusion-Animation-Processing

[glacial rock]

example outputs here: https://github.com/TaylorTheDeveloper/Stable-Diffusion-Animation-Processing/blob/main/prompts/paintingAnimation/prompt.EXAMPLES.md

[neon shard]

sorry if this is a weird question but... would it work with porn?

also why is bohemian rhapsody flagged as 18+

Oh... nvm I see why now lmfao

What would be dope would be if it divided the video into segments, and you can then redo segments that were botched, like when subjects disappear

[glacial rock]

Yep! So I did do that here for a music video I made for a friend years back. 75 or so scenes. Took almost a week to render at 1080p
https://youtu.be/FgOZyv4du4g

However the Queen song was a low priority side project, and I am focusing on stable outputs or semi stable outputs like this one here:
https://youtu.be/_rv1TKFg8WY

[neon shard]

it still looks like im tripping tf out... it doesn't look bad it just gives me serious LSD flashbacks

Now I wonder if old-school deepfaking is still alive and kicking

or if that technology has improved 🤔

[glacial rock]

So the issue with Diffusion is it is inherently random. I don't do deepfakes because of the ethical issues surrounding it, not contributing to that dumpster fire.

[glacial rock]

The wrong turn video was an earlier project based on my original approach, and the Taken for a ride video combines that approach, some refinements, and studio techniques. Foreground is high fidelity and background is low fidelity- which is why the background is so random in the 2nd video. But even with improvements there are still major limitations.

[neon shard]

oo u know what I'm thinking

if u dont already use it, you can use depth-aware masks like midas

That way the backgrounds stay untouched and only the subject is changed

that would look dope af

[glacial rock]

Feel free to share more info on the depth aware masks. I use another model, in my linked guide above, to remove backgrounds.

[neon shard]

Yea it would totally work in theory, I use it all the time, Im just not sure exactly how your program works

Ah let me drop the vid from where I got the idea

https://www.youtube.com/watch?v=TmOtHWNnPZM This is the one. In a nutshell its just an extension script that senses depth in the image, and you can tune how deeply/shallowly you want the img2img be affected. So if u set the contrast high, it will essentially only edit stuff that pops out the most, which are usually people, and leaves backgrounds untouched

Example of it in action, high contrast mask = background untouched

[glacial rock]

Are the frames out of order? cool colors.

[glacial rock]

ahh cool! I will need to check it out in more depth later on. I'm using resnet50 to remove backgrounds right now, and I supplement them with a green or blue for keying so I can then composite the video elsewhere. However if I ever need to keep the backgrounds, I'd use this for sure.

[fallen frigate]

"WATER KNOWS ME" Follow me on Twitter for more videos like this: https://twitter.com/J0HN9R1M3

[mossy ridge]

https://youtu.be/DkXCLOIDUUg does anyone know how to get anime results like this ?

[glacial rock]

@mossy ridge you can try this process linked above, but use anime models from the community. I don't really use anime models and likely your prompt will be different, but the general process is similar to create good stable animations end to end.

[mossy ridge]

i am trying to use anything 4.0 model now

[glacial rock]

Cool!

[mossy ridge]

the link you mentioned . should i use stable diffusion localy ?

[glacial rock]

I recommend it if you have the hardware

[mossy ridge]

i only have a 1070 nvidia gpu and 32 ram is it good ?

i know its not good

but is it possible?

[glacial rock]

Yeah you have plenty of processing power.

I have 2070 8GB ram

https://gpu.userbenchmark.com/Compare/Nvidia-RTX-2070-vs-Nvidia-GTX-1070/4029vs3609

[quasi hamlet]

How can I find promts to this? And do I need to pay for this promt?

[royal zenith]

https://youtu.be/S3d8pQ7nNwM

[tawdry shuttle]

Hey guys, been experimenting with deforum a little bit and I've got some pretty interesting results:
https://www.instagram.com/reel/CnvQBuuprM4/?utm_source=ig_web_copy_link

think I`ll be doing more of these kind of videos just to see how much better I can do

[plucky plover]

been playing around with my oculus rift and converting them to keyframes. dialing it in to make it smoother 🥳⭐️
https://youtu.be/FnbHLpVpbBM

[ionic zenith]

"Meanwhile the technology was evolving so fast that I worried my video would feel outdated by the time it would be ready."
It felt so nuts taking two months to create a music video with Disco Diffusion, at the rate at which the technology progressed... The next video used Deforum, it's also in the article 😊
https://medium.com/@antoine_guenet/about-4-years-ago-i-completed-the-writing-of-my-first-solo-album-6cf0ab399741

[glacial rock]

to what specifically?

[wooden talon]

Awesome!

[ionic zenith]

The one made with deforum: https://youtu.be/eOg4X92OMaA 😊

[barren mulch]

https://youtu.be/pPNGE_0rV7I

[rocky matrix]

LIke and subscribe ! https://youtu.be/wnMvDjWvC9o

[lethal steppe]

Anything better than Deforum out yet? Anyone try the tune-a-video copy?

[faint rivet]

better than deforum? hah, nonsense.

tune-a-video? what is that? I'm just using tuna-video

[dense surge]

img2img with static sd images cropped and scaled can generate interesting movements if arranged as a spritesheet.

[ashen stag]

comic hiphop Colombia ¯_(ツ)_/¯

[pseudo plaza]

You should probably start by having a look at this:
#🎥｜animation message
It's a tutorial specifically for avoiding trippy results. Wish more people did similar stuff, 97% of animations are deforum morphs or trippy/flickering videos. 🙂

[velvet cobalt]

girl steampunk realistic

[glacial kayak]

https://www.youtube.com/watch?v=8g5IAg5odUQ

[plucky plover]

when you wake up delirious at the casino 🎰🎰🎰
https://youtu.be/ZwmJuWTRbdA

[dark rune]

Parallax Background Tutorial using Stable Diffusion, Photoshop & After Effects

https://www.youtube.com/watch?v=xLi_BVaSBfU&ab_channel=JAMESCUNLIFFE

[sharp depot]

yesss I like it, good vibes! I love that scene, but if Miyazaky sees it he will be very sad. You made Hayao cry 🤣

[vast cipher]

Made this locally on my MacMini using the unmerged image2image PR in the Apple stable diffusion repo. https://youtube.com/shorts/ShXKxwuS9ZI?feature=share

[hoary ether]

New animation ❤️ https://www.tiktok.com/@ukiyoanimation/video/7193056392574553350

[glacial rock]

Unstable moustache... Audio on!

Not as crazy about this one... but looked kind of neat and enjoyed trying this model out. Audio on!

[glacial rock]

good work! Subject (car) came out really stable.

[dense crater]

https://www.youtube.com/watch?v=fF68ROoPzVQ

[faint rivet]

I have achieved balloon supremacy
https://www.youtube.com/watch?v=BJqUu1km8lk

[quasi hamlet]

How do I make a animation please

[mossy ridge]

hello , is it possible to use inpainting in batch mode
Message #💬｜general-chat

[faint rivet]

https://www.youtube.com/watch?v=ktYHEf0_ZKk

[faint harness]

https://youtu.be/uq_t1cYHSYg

[fallen girder]

"Sector 7" | Deforum Animation (Stable Diffusion v2.1) [4K] - Workflow in the video description
https://youtu.be/oGC9t0X_658

[weak stratus]

Ötme Bülbül Ötme - Doğukan Polat feat. Kerkayas (Official AI Music Video) [4K]
https://youtu.be/YMi-dy3P0lE

[plucky plover]

super computer brain 🧠💿💻⌨️
https://youtu.be/3n5OPvHgLXQ

[mossy ridge]

https://www.youtube.com/watch?v=sIoQSw71_GI

[livid fossil]

https://youtu.be/1T6S40MNEwo

[hoary ether]

New animation 🙂 https://www.tiktok.com/@ukiyoanimation/video/7193422148302589189

[soft bone]

https://www.youtube.com/watch?v=fKJdLd8XxnY

[daring warren]

https://twitter.com/NoFaceKolektiv/status/1619107689812070401

No Face Collective 🥷🏼 (@NoFaceKolektiv)

Testing custom upscalers 👁️

#digitalart #animation #stablediffusion #psychedelic #cyberpunk #AIart #AIArtwork #AIArtCommuity #AIAnimation #DigitalArtist #Diffuser #ETH #nftart #deforum #scifiart #discodiffusion #AUTOMATIC1111 #AIartists #musicvideo

▶ Play video

Twitter · Jan 27, 2023, 10:58 PM

[spring moon]

Really rough idea

[fading turtle]

YES

[faint rivet]

The close-up shots are so good.

[red sapphire]

Warp + Segmentation + Upscale

[noble quail]

https://youtu.be/XXANMhv4Bag

[hoary ether]

New animation 🙂
https://www.tiktok.com/@ukiyoanimation/video/7193714101535657222

[plucky plover]

new mario game just dropped https://youtu.be/hjPMdtjBeic

[near badger]

PC98 cg style

[serene veldt]

https://www.youtube.com/watch?v=l2K4FlpqAdU

been doing more slow long stuff 🙂

[mossy ridge]

https://www.youtube.com/shorts/Dnx3zleowYA

[full grove]

https://youtu.be/H-RU0MwcAGU this is me and @sharp relic first fully diffused music video! Enjoy 😊

[fast dome]

Awesome!

[astral bane]

https://www.youtube.com/watch?v=3t0X3q5a9dQ

[spring moon]

Did some more work on this. I am messing with some footage I recorded for a tutorial on a chord progression from the 80s I love.

[hot basalt]

Wow this is amazing

[weak stratus]

Thank you

[faint rivet]

https://www.youtube.com/watch?v=miSizi5Z4KA

[magic frost]

улыбка

[pseudo plaza]

Really nice style!

[daring warren]

https://twitter.com/NoFaceKolektiv/status/1620046921644802048?t=ESvbTcz-YbV4D-y7SK8png&s=19

No Face Collective 🥷🏼 (@NoFaceKolektiv)

Pigeons over the Nile 🕊️

🎨Stable Diffusion, Deforum.
👨🏻‍💻Flowframes, Topaz.
🎶@toundra

#digitalart #animation #stablediffusion #psychedelic #AIart #AIArtwork #AIArtCommuity #AIAnimation #DigitalArtist #Diffuser #ETH #nftart #deforum #AUTOMATIC1111 #AIartists #musicvideo #AI

▶ Play video

Twitter · Jan 30, 2023, 1:10 PM

[plucky plover]

for all my painters out there!!!!! : ) https://youtu.be/10dt0_LPid8

[fallen girder]

"Insanity" | Audioreactive Deforum Animation (Stable Diffusion v2.1) [4K] - Workflow in the video description
https://youtu.be/WBCQ8PwJaIE

[regal aurora]

Run Rabbit! Run!!https://youtu.be/oPyaJNXQq50

[fading turtle]

https://imgur.com/a/VmWIP0p

fun little looping logo anim I made a week or two ago

[open crypt]

do someone have a working link for the deforum discord? it seems it's expired

[sudden pelican]

Hi!!! Love this

[glossy ravine]

Thanks!!

[fringe flare]

Strange, there is no invite button at this moment

[vast onyx]

I couldn't find deforum discord either

Here are a couple of music videos I made with deforum
https://youtu.be/rSD4xZTDo9Q

https://youtu.be/FsVskNtNazk

[vast cipher]

Processed this overnight on my MacMini using the SD v1.5 model converted to CoreML, with the image2image pull request (hopefully merged soon). This was the first time I added some movement, by slightly zooming each generative frame using CoreImage.

also have been toying with randomly changing the prompt used throughout the video from a list of prompts.

[bright moss]

that is amazing

[vast cipher]

thanks...always tweaking my code. I want to try changing the prompt less frequently to give SD more time to generate imagery, since the zooming creates differences frame to frame, it is less likely to get 'stuck' like it did when I had a more static frame.

[viral wagon]

how to create animation

[red mirage]

Super good congrats!

[mellow kettle]

https://www.youtube.com/watch?v=WTQIZG0ngZo&t=133s

[dense crater]

https://www.youtube.com/watch?v=qDtLNaG5_Uo Lyric video with background generated in Stable Diffusion Deforum. Settings in the description!

[worthy bear]

I'm happy to show you the music video I've been working on for the past months. I will be grateful for your likes and comments. thanks! https://www.youtube.com/watch?v=GrUfoKK3-6E

[agile eagle]

Whoa

[agile eagle]

Inspired

[faint rivet]

https://www.youtube.com/watch?v=LaXI9vV7uFM

[inner glade]

#🎥｜animation

[true belfry]

With these ai generated videos. How do you make one that isn't just a imgtoimg of every frame?

[vast cipher]

Made this overnight from a beat I made yesterday in Ableton. Changed the zoom speed to be a bit faster. Thinking I might try adding some rotation next. I also tried changing the prompt less, so SD has more time to do its thing. Still can't get the blend from video to generative to be like I want. I don't mind the jumping around, I like that...but I wish that it was smoother as I raise the strength before becoming fully generative. Need to mess with it more.

[faint rivet]

https://deforum.github.io/

[vast cipher]

Most people use that, I go a more DIY route, using the Apple repo. https://github.com/apple/ml-stable-diffusion

[golden hemlock]

hope you enjoy, this episode visits the place where Ai's go to chill using a variety of Ai techniques https://youtu.be/kvNCR8n9L5g

[hot basalt]

I make music and am trying to master making animations that flow well with my music but the learning curve is steep. I’m willing to provide music for your animations if in return I can use the animations on my YouTube channel as well. Msg me if you’re interested. Here’s my YouTube…I have hundreds of instrumentals I haven’t posted and we can go through some of you’re interested. https://youtube.com/@SixMixes

[hot basalt]

I can make some very nice images and can provide prompts, seeds and images as well

[bright moss]

with a different aspect ratio, these could be pretty interesting youtube shorts that I would watch

[vast cipher]

yeah on my list is figuring out how to do 9:16, since TikTok is my primary platform for making these. So far the swift code crashes if I give it a non-512x512 image, so I dunno. I guess cropping is my only option? I feel like I've seen size settings on other SD tools...

[latent laurel]

deforum ?

[faint rivet]

https://www.youtube.com/watch?v=jq4Pf1RbILQ

[glacial kayak]

https://www.youtube.com/watch?v=77X-afnQ_us

[copper wind]

Hello all!

I meed some help regarding Stable Diffusion video generation.

As there are so many models and UIs available, can someone guide me which is the best and convenient to create a series of images (coherent or linear) and then convert it into a video.

[copper wind]

This is what exactly i am looking for. Can you please give basic steps for what model and UI should be used@

[noble quail]

All the visuals are SD with Deform: https://youtu.be/oX7ZGeFwWyo

[serene veldt]

https://youtu.be/cwBi5MCwHD0

[warped ridge]

all i wanted was for img2img to do gifs. do I call it gif2gif? giffusion? haha

[fading turtle]

This is made in touchdesigner 🙂

[daring warren]

https://twitter.com/NoFaceKolektiv/status/1621177551262744578

No Face Collective 🥷🏼 (@NoFaceKolektiv)

History kaleidoscope 🎬

🎨Stable Diffusion, Deforum.
👨🏻‍💻Flowframes, Topaz.

https://t.co/suH69TRJyx

#digitalart #animation #stablediffusion #psychedelic #AIart #AIArtwork #AIArtCommuity #AIAnimation #DigitalArtist #ETH #nftart #deforum #AUTOMATIC1111 #AIartists #musicvideo #AI

▶ Play video

Twitter · Feb 2, 2023, 4:03 PM

[blazing star]

This entire video (including character performance) was generated using AI.

https://www.instagram.com/reel/CoLuzZGLAba/?igshid=OGQ2MjdiOTE=

[faint harness]

https://www.tiktok.com/t/ZTRsKHBc9/

[ebon sparrow]

https://twitter.com/8co28/status/1621391623795113984?s=20&t=efSshCh96O5yxHWntG9_cg

852話 (@8co28)

AI
初音ミク以外のキャラを考える。
miHoYo　原神　胡桃
motion https://t.co/zWSz2hQYAM

Likes

9469

Retweets

1497

▶ Play video

Twitter · Feb 3, 2023, 6:14 AM

[vital yacht]

New 100% Deforum Animation created with "Clarity" custom model for SD 1.5 - 🔴 soft NSFW (maybe)
https://www.instagram.com/reel/CoHyNturTXh/?utm_source=ig_web_copy_link

[shrewd knot]

@vast cipher omg i love what you did! I have done install local 1111, but want to use coreml version, but i don't get the idea of anything (like convert or diffusers..) can i get a hint to run SD as coreml plz? / I got m2 macbookpro, installed anything at all.

[ebon sparrow]

https://twitter.com/8co28/status/1621645116976013312?s=46&t=RR8bUTjSyh6Z72ZMLMNqbA

852話 (@8co28)

AI
実験
miHoYo　原神　胡桃
motion https://t.co/zWSz2hQYAM

Likes

860

Retweets

116

▶ Play video

Twitter · Feb 3, 2023, 11:01 PM

[dense surge]

Is this first order model?

[vast cipher]

Went down to the city overnight so I left my MacMini processing...apparently it crashed after 18hours of images...but i got this out of it. https://youtu.be/M-5c1OtjWEY

[slender spruce]

I’ve taken the thin-plate-spline-motion-model and been building my own

[lethal steppe]

any Tune A Video guis?

[rancid raptor]

Hi team. I recently came across this video: the guy said he uses Stable Diffusion to do the warp transition. What function does that in SD?
https://www.tiktok.com/@filianorenoire/video/7191148587911318790?q=filianorenoir&t=1675539176441

[daring warren]

https://www.instagram.com/reel/CoPiVr4oAi_/?igshid=YmMyMTA2M2Y=

[glad viper]

https://youtu.be/R5oBiaNyicI

[near badger]

persona5 style -> masamune shirow style -> photorealistic style
original
https://youtu.be/H0xLBDCPfiE
https://github.com/s9roll7/ebsynth_utility

[mossy ridge]

Teen girl having sex with teen boy

[mossy ridge]

Breh 🤣

[vast onyx]

AI music video I made tonight SD Deforum with audio sync.
https://youtu.be/fwTYRvNOnSk

[rugged prism]

Loop back wave in automatic

[stable sky]

My best animation so far, everything I've learned is in this one - I hope you enjoy it and your feedback is most welcome. https://youtu.be/DVZCimjx3kA

[hoary ether]

New animation 🙂 https://www.tiktok.com/@ukiyoanimation/video/7196677628428897542

[rancid raptor]

"seed travel" addon. It warps between seeds. It's on github

[mossy ridge]

Anybody here who knows how to use Animator v6? I don't know how the syntax is for Keyframes and the prompt templates

[gentle nova]

https://www.youtube.com/shorts/eA0fBm0etlk

[gentle nova]

how was this done, coherence is pretty good

[near badger]

Using the loopback function included in the extension,
Looping back 5 times with low denoising strength(Euler a 0.15).
I also use the function to cut faces in the same size each time using face detection, enlarge them, and then img2img individually.
Finally I am generating intermediate frames with ebsynth.

[gentle nova]

Thanks for the workflow

[simple maple]

ꜱ'ᴀɪ'ʟᴏʀ ᴍᴏᴏɴ 🌙✨ https://twitter.com/MewnCat/status/1622324578977562626

MewnCat (💙,🧡) (@MewnCat)

ꜱ'ᴀɪ'ʟᴏʀ ᴍᴏᴏɴ 🌙✨

#stablediffusionart #AIArt #sailormoon (1/3)

▶ Play video

Twitter · Feb 5, 2023, 8:01 PM

[vast cipher]

Processed this overnight using a beat I made this past weekend. Uses CoreML, with my own swift code handling prompt changes, zooming and rotation. I think next I'll investigate how to make it audio reactive. https://youtu.be/s2K0aDxijxk

[visual oar]

Hi, I haven't created videos since Deforum SD v0.5, I see now it's on v0.7, I'm guessing for this version and for SD 2 a new .ckpt is needed. Does anyone know the one that needs to be downloaded?

[blazing star]

This entire animated character performance was generated by AI.😱

https://www.instagram.com/reel/CoUxETKL48-/?igshid=OGQ2MjdiOTE=

[daring warren]

https://www.instagram.com/reel/CoVD00NoKmk/?igshid=YmMyMTA2M2Y=

[faint rivet]

You don't NEED a new checkpoint, but you CAN get one.

WebUI version of Deforum uses auto1111 interface to select models. And, you can even use the depth model with deforum.

[mystic mountain]

i made dis

and I made another one 👀

[elder badge]

incredible

[elder badge]

wow. Michael Jackson as an Egyptian pharaoh is the best too

[visual stratus]

Thank you for tutorials by the way. Y'all are gods.

[elder badge]

wow!!!

[pale island]

https://www.youtube.com/watch?v=DQOw5w0A5KE

[elder badge]

how did you do it?? is there a tutorial?

[mental pecan]

A journey though a cosmic portal of time

Audio track was converted to a curve with a set of key frames that where adjusted and smoothed out.. this was probably the 20th iteration..

[smoky atlas]

How do you make animations using SD?

[vast cipher]

Didnt have time to dig into audio reactivity yet, but i processed this overnight. https://youtu.be/KEgmKpoNL4s

Converted the v2 base model to CoreML using the image2image pull request, jumped from ~1.4 step/s to about 1.9 on my MacMini M1.

[livid wasp]

KEEP IT PSYCHEDELIC!

https://youtube.com/shorts/CNaV8tyJGxg

[rocky matrix]

Made with deforum v.05 https://youtu.be/37uLDMncK80

[balmy crag]

@sweet edge 10 seconds, 10 fps, DPM++2Sa @ 25 it

0 | prompt | a tropical garden |, 3 | prompt | roses and daises in a tropical garden |, 5 | prompt | yellow hummingbirds in a tropical garden |, 7 | prompt | a tropical banana tree |

[sweet edge]

am gonna try the same and see if I get something

[balmy crag]

ya just replace the commas with newlines

[sweet edge]

I'm sort of confused though, you rendered 100 frames right? Is it supposed to be 30, 50 and 70?

[balmy crag]

not quite, yea 100 frames but per second it's just 10 seconds long

so keyframe 7 is frame 70, and keyframe apparently means second duration ;0

[sweet edge]

ah okay thats what I thought, is that always how it works? the key frame is by second?

omg

Im so stupid

It just clicked in my head...

[balmy crag]

that is slightly confusing

[sweet edge]

waaaaaaaah

It literally says time_s

and I fr was thinking in frames

God I feel like an idiot, I appreciate all of the help though! Am going to follow your suggestion above

[balmy crag]

yea i'm trying a 25 second one now, I think mine mess up when I try and rotate/zoom, it doesn't reset or maybe I have to manually reset the effect

[sweet edge]

are you using any functions?

I think you can make it zoom in and out with a function

[balmy crag]

only transform, but mine get blurry after. idk, waiting for it to render

[daring warren]

https://www.youtube.com/watch?v=NNAB1kqWlbk

[polar matrix]

How did you do that? Please describe.

[bright moss]

really cool, I love how things stretch and get distorted and then suddenly become a whole new thing that's beautiful

[faint rivet]

Playing with the instruct pix2pix model in hybrid video with deforum.

[mossy ridge]

Star Wars

[mental flare]

https://youtube.com/shorts/N6PTh-CphJc?feature=share
https://youtube.com/shorts/LxhvJUqhjHc?feature=share

[worthy bear]

https://www.youtube.com/watch?v=GrUfoKK3-6E

[daring warren]

https://www.instagram.com/reel/CoaB6ozoPj_/?igshid=YmMyMTA2M2Y=

[faint rivet]

instruct pix2pix
https://youtu.be/dhJBL7dRinE

[noble quail]

https://youtu.be/jFwGfeFKHeg

Just posted episode 2 of my fully AI generated talk show - ConanDiffusion

[serene veldt]

https://www.youtube.com/watch?v=pVEww6EGr8M

[sullen kiln]

is there an API for deforum? does automatic 1111 have an API with Deforum end point?

[daring warren]

https://www.instagram.com/reel/CocU2SEIKUN/?igshid=YmMyMTA2M2Y=

[coral gazelle]

https://www.youtube.com/watch?v=ZpeDNdhAOEw - made with love, SD, chaiNNer, DAIN, and first order motion model

[acoustic ice]

https://youtu.be/hZ2IQZ2oz78

[opal pollen]

I directed this years manifest for Oneplus using AI and Stable Diffusion. In it's form it's a tiny love letter to their customers of course, but also to AI-animation, combining customs models and tons of new techniques I developed for this production to bring some of the previously impossible shots:
https://www.youtube.com/watch?v=iUEtLgPOea4&ab_channel=OnePlusIndia

[faint rivet]

https://streamable.com/kt799q

[faint harness]

https://youtu.be/Hv0xpOjZlHY

[mental pecan]

thank you...

Another test of some audio synced to the visuals..

[feral lintel]

Cool work! Is this generated by a input video with camera movement?

[lost latch]

https://www.youtube.com/watch?v=ytCeYDOThjE

[fallen girder]

"Meltdown" | Audioreactive Deforum Animation (Stable Diffusion v2.1) [4K] - Workflow in the video description
https://youtu.be/sqq69mW1hH8

[daring warren]

https://www.instagram.com/reel/CoetxbKox84/?igshid=YmMyMTA2M2Y=

[hoary ether]

New animation 🙂 https://www.tiktok.com/@ukiyoanimation/video/7198547089264020741

[stable crest]

https://giphy.com/gifs/fuckthatsdelicious-viceland-fuck-thats-delicious-l2JJO0D0JpgoU5OTe

[little glen]

Thanks! This was pure txt2img

[mental pecan]

Trippy Trap Beat

[haughty isle]

I've uploaded a Parseq video tutorial that covers everything you need to know to get started: https://www.reddit.com/r/StableDiffusion/comments/10z6n0y/parseq_tutorial_1_fine_grained_control_of_stable/?utm_source=share&utm_medium=android_app&utm_name=androidcss&utm_term=1&utm_content=share_button

[vast onyx]

Made a new song & created a video for it using Deforum
https://youtu.be/xWaMRCm0GFE

[spiral knot]

https://www.youtube.com/watch?v=xhU3nZazrZ4 Small clip of my next short. Needs more work, will be longer, and will come out by next week. I will do a video break down of my process too.

[languid badge]

How did you make this with 0 jitter?

[spiral knot]

Everything is animated with premiere pro and after effects.

[mental pecan]

tOTal WIP.. just messing around..

[faint harness]

https://youtu.be/iqsFNvmRM8A

[dry pier]

one of my last animations
https://youtube.com/shorts/2GoQ_-HAWtI?feature=share

[left raven]

Is there a room or an option to create a room for animations? Appreciate

I mean voice channels

[heady karma]

Hey there! I try to do some loopback wave animation for 20 created images, anyone know a solution to that?
Seems like, i am to dumb. Tried now the last 8hours😅

I only found scripts, where i could set specific prompts at frames or time but no images.

[daring warren]

Thoughts?

https://www.youtube.com/watch?v=WQI-gH0ZitU

[vast onyx]

@daring warren cool effect, I assume soft light or overlay over AI stuff

Also Golden Light is a banger track, rewatching the original song video now

[daring warren]

Just frame post processing, it came out very smooth wiith the right tools 🔥

[daring warren]

Agreed 🔥

[faint rivet]

https://www.youtube.com/watch?v=ZXsuwbJhpTQ

[faint rivet]

https://www.youtube.com/watch?v=riOTTpP_ttk

[vital yacht]

New Animation created with Deforum Stable Diffusion (ProtoGen X 5.3 Model)... 🔴 Maybe soft nsfw
https://www.instagram.com/reel/CodG10kJF_8/?utm_source=ig_web_copy_link

[lilac tide]

@faint rivetThat is good.

[stable cedar]

https://youtu.be/eaYY4sbupY0

[feral lintel]

Cool, thank you!

[warped ridge]

Had a potentially bad idea -- folding a giphy/tenor GIF searcher into the gif2gif script. Like

Going to see what apis are like. If it's low effort I think it'd be fun.

[jagged token]

Hello, a question that you use for that result

[little glen]

I'll try to answer, what's your question?

[normal helm]

#1072220168534642768 frog

[plucky plover]

https://www.youtube.com/watch?v=gqxO1kz0cw8

[bright moss]

that's sooo consistent, wow!

[gleaming depot]

#aboywitharrow

[noble quail]

hey everyone! My project inspired by Nothing, Forever - ConanDiffusion is now live on twitch! https://www.twitch.tv/conandiffusion

[dark sage]

One of my first deforum animation. https://youtu.be/dlMoZqH_aGY

[brave ingot]

https://www.youtube.com/watch?v=LBvSXxMqD1s&ab_channel=agathacrusty

[mental pecan]

https://youtu.be/b4QOaQGsw4s

[faint rivet]

https://www.youtube.com/watch?v=sJxxlkRrziM

[proven sleet]

https://www.youtube.com/shorts/EwIq6n94-OQ
Stable Diffusion + ChatGPT + D-ID

[ripe plaza]

I'm trying to make animation using deforum, it generates the first few frames properly but later it gets distorted and the image gets filled with weird colors. Am I missing something? How do I fix it?

i just added translation x and z keyframes

[vast onyx]

Some models just don't work too well, but your issue could be something else. I noticed weird stuff with SD 2.1

[ripe plaza]

i'm using 1.5

[ripe plaza]

i just restarted SD and it's working now

[vital yacht]

AI Animation 100% Deforum Stable Diffusion - "Zombies Cold War" + Original Audio
https://www.instagram.com/reel/Cop_S5AvYD3/?utm_source=ig_web_copy_link

[gilded vigil]

Chibiiii!

[vast cipher]

Experimenting with different ways to mess with image2image over video. Started with this live performance i did last weekend. Then wrote code to slowly oscillate the strength parameter like a LFO, and each time the value gets to its lowest (closest to the real video) I change the prompt and seed. This is using a v2 base model converted to CoreML, my code is all swift. https://youtu.be/f5NrOu6M9sA

[daring warren]

Appreciate some love on the reel, IG hates me.

https://www.youtube.com/watch?v=bovJS2mDJtg

Reel

https://www.instagram.com/p/CorvEXbIS-Q/

nofacekolektiv

The Canteen🌹🇲🇽

🎨Stable Diffusion, Deforum.
👨🏻‍💻Flowframes NMKD, Topaz, RunwayML.

× Rendered and post processed with AI, 2k 60fps. Each step is as important as the main output.
× A Share is always appreciated 📲, IG doesn't have much love for my account.
× I will always try to help and assist, but I'm not making any tutorial or guide, if you are...

Instagram

[fallen girder]

This is a music video I've been working on for a client
https://youtu.be/G9SUkE929dU

[glossy osprey]

Is there a way to use random seed for each generation in batch img2img? I’m using automatic1111

[safe oriole]

Trains through the ages with deforum https://www.youtube.com/watch?v=HMffOrcEjEU

[stable cedar]

https://youtu.be/3imiSusAr1s

[gilded vigil]

Aliens!

[near badger]

Sometimes I get asked by DM
but I'm simply using this extension to output.
There's no particular process that requires technique.
https://github.com/s9roll7/ebsynth_utility

[faint rivet]

https://www.youtube.com/watch?v=_kKrssV0ojA

[noble quail]

hey guys just posted the latest episode of my ChatGPT written talkshow ConanDiffusion - featuring Paul Rudd and a "clip" from his latest movie. Also a new segment where Conan reviews AI generated images, and Chris Rock at the end! All voices are 11labs! https://youtu.be/Wd5yRlYM798

[vernal topaz]

I've been playing with animated transitions from one state to another recently. It's quite fun. Here's me on a bad coffee day. https://www.youtube.com/watch?v=u9NO0IeuZ1g (Details are in the YouTube description.)

[lost latch]

https://www.youtube.com/watch?v=5Jxcl_Ixcmo

i'm on chapter 3 of my retelling of the hobbit

[polar echo]

Web3#: animations

[royal zenith]

https://youtube.com/shorts/CPIR5cM12vc?feature=share

[gilded vigil]

Roboooots!

[arctic latch]

Thatsvawesome!

[faint rivet]

https://youtu.be/IL-aIzTh_lg

[sudden phoenix]

@twin lion sd2iclone

[twin lion]

Terrific

[elder island]

ControlNet + 200 frames

[gilded vigil]

Chiiibi sings!

[nocturne zephyr]

/dreamperfum

[daring warren]

I'm sorry that quality conversion sucks, will try with different encode next time!

https://www.instagram.com/reel/CoxQMRco4Iu/?utm_source=ig_web_copy_link

[fallen girder]

"Fixer" | Deforum Animation / Video Input (Stable Diffusion v2.1) [4K] - Workflow in the video description
https://youtu.be/lrwr0R_lj4M

[livid wasp]

pix2pix mode is crazy 😂

https://youtu.be/p2vdJIvMJ1k

[gilded vigil]

Sing Sing Sing!

[mossy ridge]

https://cdn.discordapp.com/attachments/413297512309325825/1076234397117849650/Untitled.mp4

@winged coyote

[warped ridge]

So I'm exploring a novel method of SD animation. I had a thought, "what if all frames were generated simultaneously?"

So i stuck 28 frames of a gif on a 2000x2000 sheet:

Seems to generate all frames much more consistently, if very small

results are interesting at least

[red mirage]

This is actually a fascinating idea. Generally with multiple people img2img can freak out a bit, but you're right this is pretty consistent. Did you use ControlNet at all?

[warped ridge]

yea controlnet was what kept all the cells in-line (mostly)

there is some tracking issue where the borders warped, but something maybe i can rein in

it has no idea what to do with rick astley's seductive moves

I'm noodling a workflow that involves shrinking the frames to get as many to fit on a 2048x sheet, generating against it, breaking the frames out, upscaling the frames, then sequencing them.

[warped ridge]

tempering my expectations of this approach, but we'll see

[faint rivet]

https://www.youtube.com/watch?v=DO55NBBzndQ
Here's some more found footage from the set of my imaginary hit Scandinavian children's television show...

An eccentric model maker and French animator named Otto Mattique' was hired to produce sets for the Scandinavian children's TV show, 𝑲ₙₑᵣₖ​𝓥ᵃᶠᶠˡᵉ?!¿¡. Every month, he would create elaborate sets on his own, and then drive them down to a locked warehouse for the TV studio. But, isolation didn't treat Otto very well, and a previously undiagnosed mental condition began to eat away at Otto's sanity. The TV studio decided to go with another set designer, and Otto was fired. There was just one problem... Due to multiple oversights by the studio, he never received the paperwork, he still had the key to the warehouse locker, and he was still getting paid every month by the studio! So, Otto just kept going, while his art descended into madness. The imagery becomes more detailed as his mental condition worsens, sometimes evoking disturbing imagery from his childhood, his time as a political prisoner, or even from his time in the war. This is his story. The studio started running out of storage space eventually, so they donated them to different organizations. Eventually, they were destroyed, as organizations closed or ran out of space for the enormous sets. However, there were many still images and some animated footage made by Otto himself. His techniques were so strange that modern cinematographers still can't figure out how he did some of the shots, like the ones depicted here. But, most of footage is said to be lost. If any found footage comes in, I'll be sure to share it with you on this channel.

[daring warren]

https://www.instagram.com/reel/Co0C8y9ouJB/?igshid=YmMyMTA2M2Y=

[rugged prism]

Great work!

[warped ridge]

😄

[pallid lance]

https://www.youtube.com/shorts/-u-J6OHWj5E

coming soon

[cinder hamlet]

hola

[lusty briar]

Here is my latest song and diffusion animation:

https://youtu.be/3j95vtwO05E

[lusty briar]

Sweet. Video input right?

[round jay]

ControlNet provides control over Stable Diffusion model via methods such as Human Pose. XR Animator app can be used as a pose helper, accepting FBX/MMD motion/image/video as input, animate the 3D figure and save as image/video https://github.com/ButzYung/SystemAnimatorOnline/releases https://hub.vroid.com/en/characters/6524757472248303508/models/1735312209306041515

[mental pecan]

small test using stable warp.. https://youtu.be/OWEMwUkp8Kw

[round jay]

A tool for editing pose images that can be used with ControlNet. https://huggingface.co/spaces/jonigata/PoseMaker

Keyframe generated with ControlNet img2img using the canny edge detection model & animated using Ebsynth

[mental pecan]

@round jay those are dope.. as a 3d animator im looking forward to taking these free animations from https://www.mixamo.com/#/ and pushing it though stable/controlnet , image2image, maybe deforum some day..

[faint rivet]

https://www.youtube.com/watch?v=fYUg0MbeLoE

[slender spruce]

Control net let me fix one of the issues I had with dog faces, I was super happy how this came out (thin plate spline to animate)

[round jay]

lol this is pure gold

[round jay]

• Top: Output generated with ControlNet + EbSynth
• Bottom: Input video sequence from photogrammetry 3d scan

[faint rivet]

https://www.youtube.com/watch?v=OB-Uwdvy0L4

[faint rivet]

Some Instruct Pix2Pix but with some good old fashion Perspective RANSAC motion. See description for settings...
https://www.youtube.com/watch?v=1PcO9G38NWE

[ebon sparrow]

ControlNet Video Tool

https://twitter.com/pirrer/status/1627193071577468930?s=20

fox hsiao (@pirrer)

ControlNet Video Tool ，說好不哭。
方法 : 上傳影片 -->文字 : 人物變老人 --> 完成
有空有閒，就是多混幾段加音樂剪在一起，就會是一份作品 (?) @fffiloni

▶ Play video

Twitter · Feb 19, 2023, 6:27 AM

[faint rivet]

Alright, I went ahead and upscaled to 4K for this one... It deserved it.
This shows a new feature I'll be adding to deforum: hybrid_flow_factor_schedule. This is flow factor -1, in other words, reversed flow.
https://www.youtube.com/watch?v=gfZVCh4yVI4

[faint harness]

https://youtu.be/e5Ao-nqf_Rg

[daring warren]

https://www.instagram.com/p/Co5ABNiopSI/

nofacekolektiv

Rocket Man 🚀🌌
🎨Stable Diffusion, Deforum.
👨🏻‍💻Topaz, RunwayML, RIFE.

× 4k 120fps, downgraded for IG.
× Share if you like it 📲
× Much love for the ones who support 🙌🏻🖤

×××××××××××××××××××××××××××××××××××××××××××××
.
.
.
.
.
.
.
.
.
.
.
.
.

Instagram

[round jay]

https://avatarposemaker.deezein.com/

[round jay]

STARGATE

[inner crest]

Yes I have the Jaffa. Now I try to get the cast !

[civic lily]

https://vimeo.com/797802975 using FILM interpolation with diffusion, and some point advection as a guide, short writeup at http://the.strange.agency/projects/advection-diffusion/

[faint rivet]

https://www.youtube.com/watch?v=NsPAnZm5N-k

[faint rivet]

https://www.youtube.com/watch?v=wJ97SXgT2g4

[faint rivet]

https://www.youtube.com/watch?v=V_nnePlmjC4

[vast cipher]

Feeling left out cause I don't have access to all the fun new depth and controlNET toys using CoreML, but I am working more on my swift code. Got RealESRGAN uprezzing implemented which gave me the 9:16 aspect ratio I've been wanting to use for my posts to social. https://youtube.com/shorts/1z1m0r_BGm4?feature=share

[warped ridge]

a bit more stable. gettin' there.

[quaint creek]

test

#🎥｜animation morning in USA

[round jay]

not there yet, but its getting closer, you know what i mean 😛

[round jay]

2. Use ffmpeg to convert your video to pngs of each frame
3. On the img2img tab, run a batch pointing at the folder where the pngs from step 2 are, prompting your LORA or whatever from step 1. Enable ControlNet and use both the OpenPose preprocessor and model. In some cases the other options will work better, like if you aren't doing full body shots.
4. Use FlowFrames on the resulting images to generate the final video
(Bonus thing to try: prompt for a single-color background, then replace it with the same image in every frame. Otherwise, set the seed on step 3 to a specific number to reduce the change in the background between frames)```

[bright moss]

absolutely amazing

[noble quail]

https://youtu.be/_pEhMRkaam4

Just finished a special new episode of ConanDiffusion (my AI-generated talkshow) featuring Wonder Woman and Cleopatra!

[plush vigil]

Sir i keep getting this error im using webui, and put in the folder not load

[pallid lance]

https://www.youtube.com/shorts/H56TC3p6L-E

[round jay]

In this test, I can confirm that the thickness of the Openpose bone affects the image. Too thin or too thick can cause unwanted problems. If a problem arises when directly inputting the openpose bone, it is necessary to adjust the thickness.

Multi ControlNet makes it possible to use multiple ControlNet models, but there is a high probability that strange mixed images will be created if two overlapping images of unequal shape are used at the same time. This problem can be prevented to some extent by lowering the Guidance strength or weight of ControlNet located in the back. In this case, since the depth is the background and the openpose is the character in front, lower the Depth strength to 0.3.

[magic relic]

Hmm good point

[warped ridge]

https://tenor.com/view/bill-nye-bill-nye-the-science-guy-scienceguy-consider-gif-7391226

"Guidance strength" specifically refers to what % of generation steps will be guided. I tend to keep it as low as possible, particularly with open pose.

For example, 0.25 guidance strength will stop guiding at 8/32 steps.

"I got the pose I wanted; stop fucking with my generations."

What'd be cool is if they could make the slider -1 -> 1, so indicate guidance strength coming in at the end of the steps rather than when it stops

[haughty isle]

A new version of Parseq is live! https://sd-parseq.web.app/.

Biggest new features are a brand new prompt management experience better suited for handling multiple prompts, and a new "Preview" section, allowing you to quickly see the value of all fields and the fully rendered prompt on each frame. There's also a range of other UI improvements and a few new functions & variables. See https://github.com/rewbs/sd-parseq#version-0140 for details.

[round jay]

Background with segmentation and the person with openpose with two ControlNets!
① segmentation, ② openpose, ③ output result!

Background with depth and the person with openpose with two ControlNets!
① depth, ② openpose, ③ output result!

[inner crest]

Made with ControlNet

[inner crest]

Guess what Iam going to do:
https://youtu.be/aKSrFWShKKQ

Oh yeah I will make it come true !

https://tenor.com/view/austin-powers-yeah-baby-gif-22521962

[signal gulch]

experimenting with control net
https://www.youtube.com/watch?v=FSbmgMqBkl0

[wooden forge]

BRDR

[daring warren]

https://www.instagram.com/p/CpANgTkozpA/

nofacekolektiv

Dissolve🫠
🎨Stable Diffusion, Deforum.
👨🏻‍💻Topaz, ControlNet, RIFE, RunwayML.

× Share if you like it 📲
× Test with ControlNet, 2k 30fps, this one was a bit tricky, couldn't get better results with more frames without glitching the interpolation.
× Much love to the ones who really support 🖤

Instagram

[west pond]

un logo reebot, UNDER ARMOUR, nike

[raw jackal]

A reminder for you lovely folks frequenting the animations channel~

I'd really, really love to hear your feedback on our server survey if you have 1-2 minutes to fill it out.
The survey is being closed <t:1677517200:D> so make sure you fill it out if you haven't already!
(If you would prefer to remain anonymous, simply hit "next" on the first question without putting your discord username in!)
https://docs.google.com/forms/d/e/1FAIpQLSdpaL4lcXvdexbLEnAnP-G3w5e56nXuq_0-CfWDFKpmpAlSww/viewform

[grim glacier]

how was this made plz?

thats one of the few animations that doesnt look like a shitty acid trip

[plucky spear]

The camera hovers above a vast, dimly lit chamber, filled with piles of glittering treasure. The treasure is a dizzying array of gold coins, bejeweled goblets, silver candelabras, and precious gems, all arranged in haphazard stacks that reach up to the ceiling. The air is thick with the musty scent of old wealth.

[faint harness]

https://youtu.be/7JtM0MAIJYk

[limber bobcat]

I follow instructions from https://youtu.be/3FZuJdJGFfE

[daring warren]

https://www.instagram.com/p/CpCw2KwIqwr/

nofacekolektiv

Flexin' on this robots every bone and muscle 🔥

🎨Stable Diffusion, Deforum.
👨🏻‍💻Topaz, ControlNet, RIFE, RunwayML.

× Share if you like the quality so more people see it📲. It hurts to see effortless stuff going viral😅
× 4k 120fps (30 for IG), fully uploaded on YouTube.
× Much love to the ones who really support 🖤🙌🏻

Instagram

[grim glacier]

bring on the fembot wives hell yeah

[inner crest]

Multi-ControlNet: OpenPose + normal map. Not perfect. Poor quality of hands and I had to inpant a lot of frames. But I think ControlNet allows stuffs that was impossible before. Also, I think with a strong embedding or model where face, haircut, clothes are well defined. It may possible to be really close to perfection. If they could sync their model with SD2.1, animation would be improved also. SD2.1 is annoying for a lot of stuffs. But hands, eyes, faces are almost perfect with the correct prompt.

[grim glacier]

too trippy for my taste

you can use depth map to make the clouds consistent so that only the girl changes

[inner crest]

nope because look at the original video:

depth map didnt allow me to create new stuffs and modify the hat. Or I had to lower the weight so much it was useless

But probably, if I wanted just to lightly modify the video. Depth map would have been a better choice

[grim glacier]

There has to be a way to re-process it, so first process using canny controlnet and second pass using depth map somehow

[inner crest]

Canny go deep into the details

I tried 😄

I swear

OpenPose is great, because it doesnt focus a lot on details

therefore, I could change the woman as I wanted

[round jay]

#🎥｜animation message

[inner crest]

shit

it was the same video you used ?

holyshit

yeah it is the same 😄

loool

what method did you use ?

[bright moss]

are you combining these in an external photo editor like photoshop or GIMP?

[grim glacier]

no offense but that looks even worse lol

ima upload one to show yall wassup

working on it although its nsfw

[inner crest]

https://tenor.com/view/areg-tarakh-tarakh-areg-tarakh-demq-judging-you-gif-17135953

[grim glacier]

thats one uqly mf

[round jay]

https://cdn.discordapp.com/attachments/1025266140445933648/1078133614342062191/Fpmg9y2aMAATNHr.jpg open full size and ZOOM IN!

[inner crest]

damn the legend was there

[bright moss]

since when can you layer controlnet? That's awesome!

[mossy ridge]

those are impressive videos

[round jay]

👍 😂

[bright moss]

what exactly is this part doing to the image?

[round jay]

segmentation of the background for generation

[inner crest]

background yeah

Me I think I will try to create a custom embedding

with clothes, face, haircut

[bright moss]

using a normal map?

[inner crest]

then use Control with it

I feel like Im just scratching the power of this script...

[inner crest]

https://tenor.com/view/next-level-thats-what-we-need-next-stage-top-level-level-up-gif-14109554

Same video with less fps. So like that @grim glacier wont puke on his keyboard

[grim glacier]

roflz

yo. can I post nsfw vid here

?

no nips just uncovered boobies

[inner crest]

https://tenor.com/view/cage-priest-dance-gif-14793366

I would say no 😄

damn this gif would be awesome with COntrolNet

but define what is uncovered boobies ?

because when you post the video, there is a symbol like an eye. You click on it and you may post stuffs unholy

[grim glacier]

I can post the original but im afraid it might get removed cuz theres a bit of nipple showing trolol

[inner crest]

just do a spoiler attachment

[grim glacier]

nah fudge it ima wait till I finish my own and then I'll post it and if I get banned then fuck it lol

oic

kk gimme a mo

[inner crest]

there is a bot that will scan your attachment

it wont allow you to post it if it is too sensitive

https://tenor.com/view/anchorman-amateur-hour-gif-20673985

[inner crest]

lol

https://tenor.com/view/get-out-leave-gtfo-get-out-of-here-gif-6207127

[warped ridge]

the hunt for consistency across frames can't always produce winners

[fringe carbon]

So I'm starting this new major record label in Oslo/Amsterdam/Los Angeles/Houston and we're lookin for someone who can help us make some awesome animations. Money is no issue, the music will be decent, and you'll get to travel with us to a lot of festivals.

Anybody interested?

[karmic atlas]

yeah, check out blnk studio
https://www.youtube.com/@blnk__studio
https://www.instagram.com/blnk__studio/
we produce already music animation to our videos, also animation series is in development

[round jay]

you need to step up your game boy! 😉 @inner crest

[grim glacier]

niiiiiice. At least the clouds are consistent

5 years from now we're gonna look at this shit and say damn... those were the dark ages

who's ready to see my masterpiece. I can only post it for a few seconds cuz it'll probably get taken down.

[grim glacier]

there has to be some sort of algorithm that can stabilize that face motion

like it shouldnt be that hard for a computer to do

kind of like stabilize video but not for movement

[warped ridge]

the entire frame is distorted. it's a work in progress. haha

GPU go brrrr

[round jay]

yeah needs a lot of work XD

[inner crest]

https://tenor.com/view/liz-what-gif-22720185

you apply a mask ?

[warped ridge]

i've been getting a more "consistent" result in terms of the subject, but at the expense of this caustic type distortion

[inner crest]

In this video, I use a strong positive prompt like (((pink_afro))), (((unicolor))), (((pink_unicolor_shirt)))

and I think the result is more consistent

[warped ridge]

(((((((((((((((((((TOM CRUISE)))))))))))))))))))

[grim glacier]

The thing is that they keep the same facial expression during the whole video, but I gots an idea

[inner crest]

lol

[grim glacier]

first lemme see if I can upload my piece without nitro

[inner crest]

but yeah it can be a good idea

a positive prompt like ((((Zendaya)))) or ((((Rihanna)))) may increase the consistency

I mean a star face

[grim glacier]

What I'm thinking of doing is using the same prompt throughout the whole video, then redo it with the same seed but a different prompt for the facial expression during certain points

[inner crest]

in the second video I didnt do that

[grim glacier]

then splice the videos together for a smooth transition

[inner crest]

and the consistency is not the same

[inner crest]

lol good luck

[grim glacier]

now if somehow we could batch resize>inpaint the face, that would fix the distorted faces in full body shots too

At this point doing old school deepfakes seems more plausible lolz

ooooh actually

that gives me an even better idea.

[inner crest]

yeah exactly you use deepfakes for your animation 😄

but I think the prompt allows more control

until I find out a good embedding

since I can only use SD1.5 models...

https://tenor.com/view/no-rolling-eyes-gif-14437483

[round jay]

most embeds and loras are for 1.5, fidning anything good for 2.1 is very hard lol

[inner crest]

I create them

nothing is better than what I make...

https://tenor.com/view/fresh-prince-hair-cut-will-smith-hair-did-new-hair-cut-gif-12045822

[grim glacier]

u dont want 2.1

wait till 3.0 which will be 1024 for those of us with monster gpus

Take a peek before mods see it

My idea is to use a full body shot or something similar to this, just as a base

then do a deepfake [of the original] and so that all the real facial expressions are kept

[inner crest]

a custom model I created in order to get images for my next embedding

...

Go get something similar with SD1.5

[grim glacier]

and then just substitute the face on the SD video and you have essentially a true-to-life celeb body and a true-to-life face

whereas classic deepfakes only deepfake the face

[inner crest]

wtf 😄

lmao

damn bro

[grim glacier]

genius or what

[inner crest]

lets see the original video ?

[grim glacier]

One sec

[inner crest]

because you probably lower the weight

[grim glacier]

yep. the original is an asian

but I currently dont have any hot asian girls dreamboothed... yet!

nitro is not letting me

Ok there ya go

its uploading

[inner crest]

but you see in your prompt you should increase the weight for the lingerie

and explain the color,

[grim glacier]

Oh I did

[inner crest]

describe it, detail it

[grim glacier]

Im surprised it worked this well

but my idea is to do multi-prompted deepfakes

as a #2 option, my #1 would be hybrid classic deepfake mixed with SD deepfake

[inner crest]

maybe it may work

[grim glacier]

Anyone here have any idea on where to start on old-school deepfakes? Like the classic donald trump ones

afaik those have no real interface its just cropping a bunch of faces

[inner crest]

me I like to push the limits of the script. So I try new stuffs

[grim glacier]

maybe if I really get a lot of time in my hands I can come up with a script that will let you change the prompt after X frame.

[inner crest]

like the original video and my video may be really different

[grim glacier]

That sounds like a fantastic implementation

[inner crest]

lol good luck

[grim glacier]

with chatgpt anything is possible these days

Oh, btw the original got blocked.

[inner crest]

lol

[grim glacier]

probably cuz theres more booby in there lolz

[inner crest]

https://tenor.com/view/you-sick-old-pervert-you-sicko-youre-disgusting-thats-wrong-paul-reiser-gif-15608448

I swear so many nude generated with ai 😄

[grim glacier]

why else would anyone use it?

[inner crest]

probably why my models and embedding dont work on civitai 😄

lool I swear I create a stromb embedding on ancient Egypt: 300 downloads

and this girl with TI Embedding - Style-Hamunaptra (Ancient Egypt) - 1.5 Base

makes 8k downloads

[grim glacier]

no shit lmfao

[bright moss]

pretty sure because that person had several very successful embeddings before hand and already had quite a following so their followers downloaded it boosting it to the top of the embeddings page. I think they are in the top 5 on civitai if I am thinking of the right person.

[inner crest]

so obviously. Me and my egyptian cat...

[bright moss]

ok nvm

they are 22

[inner crest]

lol

it is also SD1.5

me I only create SD2.1 embedding

[raw jackal]

We're getting a little close to Rule #4 with this one chief 🙂

[bright moss]

and it only has 7k downloads

[inner crest]

it is in the #1047197565365538826

I didnt generate it

[raw jackal]

👀 Is it?

[inner crest]

you may close her post 😄

and say to use mine

it is censored 😄

and for kids 😄

[raw jackal]

Just to clarify the model is fine - it's that specific picture that's iffy

[inner crest]

#1063200420178165892 message

no no no

[bright moss]

I have 5 embeds that are over 1k downloads, 2 are over 2k, and all of mine are over 350 at least.

[inner crest]

it is unsafe for children 😄

[raw jackal]

Thanks for the details! Appreciate it!

[inner crest]

look at all these nude women

https://tenor.com/view/nun-fanning-gif-11713370

Im shocked...

[bright moss]

It wasn't even trained with images, it was just tokens which works a lot differently. Just putting a bunch of words that sd already knows into a file and using that instead of writing out the words yourself. Probably used an extension like Embedding Inspector for it.

[inner crest]

yep and 8k...

when hard workers like us get nothing...

https://tenor.com/view/life-is-unfair-unfair-sobbing-crying-emotional-gif-15178308

[grim glacier]

mfw the most popular models are actually just blends of other models

[inner crest]

lol I swear

[grim glacier]

shit ive been thinking of uploading my own custom blend because it works mwah

hey lettem do it. the AI community is like communism. one for all and all for one

[inner crest]

cough Protogen cough

[grim glacier]

protogen + ultrarealistic porn model + vision AI

with a slight touch of 1.5 for extra flexibility

[inner crest]

SD2.1 😍

[warped ridge]

not sure if this is irony or just ultra-meta

[grim glacier]

not sure if ironic or u called that blurry face impressive

inpaint that joint at the very least if u gonna defend 2.💩 models

[bright moss]

looks like a low res image, just use hires fix

[inner crest]

yep it is a raw image

[inner crest]

Why making one One Punch Man video when you can make three ?

[bright moss]

Three Punch Man

[mossy ridge]

that's for the off topic channel at most, not that discord most likely. strange account tbh

[left raven]

Does anyone know what the last tweet of Emad can mean? He says "Some fun AI-generated media dropping Sunday... Future is coming fast, time to nap..." Text to video maybe?

[faint rivet]

https://www.youtube.com/watch?v=ocYOl0KB0uk

[fading turtle]

https://www.youtube.com/watch?v=-RHeR6HeU_Y
this one is special. AI stuff made all with stable diffusion inside touchdesigner

[frozen swift]

I spent today breaking down a video2video workflow using controlnet + ebsynth + nerf (for funsies)

https://twitter.com/bilawalsidhu/status/1629632110456602631

Bilawal Sidhu (@bilawalsidhu)

Multi ControlNet is a game changer for making an open source video2video pipeline. I spent some time hacking this NeRF2Depth2Image workflow using a combination of ControlNet methods + SD 1.5 + EbSynth.
🧵 Full breakdown of my workflow & detailed tips shared in the thread below ⬇

▶ Play video

Twitter · Feb 25, 2023, 11:59 PM

Do ya'll like this level of detail in a thread? Because it's the longest tweet thread i've made so far lol

[cursive orchid]

super helpful. I'm keen to see more of your experiments. Looks really good

[civic lily]

Using LiDAR as seed image to guide diffusion, animating start_schedule param over time https://vimeo.com/802403310

[hot umbra]

First time playing around with Deforum
https://www.youtube.com/watch?v=-P5Rq2PYFT0

[mossy ridge]

I think you guys are going to appreciate this

[mossy ridge]

How many frames was this dude ?

[inner crest]

24fps

[mossy ridge]

damn

How much time it took ?

[limber dagger]

hello everyone, I need some resources for mac. I couldn't use stable diffusion animation with my mac, is there any good resources? thank you

[warped ridge]

i am but a humble gif merchant

[boreal obsidian]

how can i use batch inpainting

[daring warren]

https://www.instagram.com/p/CpKorwxIdNN/

nofacekolektiv

Expanding my Deforum knowledge to some things I've never tried before like parameters based on guided images. I had to do something about it and test it with the newly film interpolation, and what better than a banger fromfor it 🔥

Just a little test that I liked with the new logo concepts 🫶🏻

🎨 Stable Diffusion, Deforum.
👨🏻‍💻, FILM, edited with...

Instagram

[mossy ridge]

Original for the sake of comparison

[waxen basin]

Finally able to remove jitter on AI videos! Used DaVinci’s deflicker plug in.

[rigid cliff]

@waxen basin just wanted to post the video i saw this in an realised it is your video.

when you have content like that, you are allowed to post the video.

https://www.youtube.com/watch?v=PeXRFH2I9Is

i don't think anybody minds

the "automatic dirt removal" plugin in the chain is pretty genius actually. nice that corridor put so much time and effort into these things.

[faint rivet]

https://www.youtube.com/watch?v=McltGTRf2rM

[faint rivet]

https://www.youtube.com/watch?v=KUcMIqESsLo

[faint harness]

https://youtu.be/kffQBPfQ8g4

[noble quail]

Introducing ColinDiffusion! We are switching Conan's name to steer clear of any copyright issues. The same show, with a handful of improvements! You can watch tonight's monologue premiering at 9:15 PST. With the full episode to come later. https://www.youtube.com/watch?v=h9lr2HN8iiM&ab_channel=NoReruns

[sharp depot]

https://youtu.be/McyooBBYKJQ

[crisp pumice]

https://www.reddit.com/r/StableDiffusion/comments/11e12j3/colouring_bad_apple_using_sd/

[daring warren]

https://www.instagram.com/p/CpNwvcbIfMK/

nofacekolektiv

Feeling the heat, playing with fire, dancing with danger - some things are just too tempting to resist.. 🔥

🎨 Stable Diffusion, Deforum.
🪄 Topaz labs, RunwayML, edited with Capcut.

× Share if you like it 📲
× Much love to the ones who support 🙌🏻🖤

Instagram

[plucky plover]

recreated a music video shot on iphone with controlnet img2img https://www.youtube.com/watch?v=UjE-H24n2M8

[faint rivet]

https://www.youtube.com/watch?v=ciDnQ3N_PQw

[keen kelp]

WIP https://streamable.com/t0afr3

[faint rivet]

https://www.youtube.com/watch?v=in4lqTVw4I4

[jade prawn]

Amazing! is this with deforum?

[keen kelp]

Its a custom system based on Touchdesigner and 1111 https://streamable.com/6m6l4o

[vital yacht]

🔴 Soft NSFW
New AI Animation 100% Deforum: "The Elohim's Gan-Eden"
https://www.instagram.com/reel/CpFhpr5tsId/?utm_source=ig_web_copy_link
And this is another one, 100% Deforum AI Animation: "The Lucifer's Transmutation"
https://www.instagram.com/reel/CpOBAAds3PQ/?utm_source=ig_web_copy_link

[faint rivet]

I almost deleted this test... but, it was funny.
https://www.youtube.com/watch?v=bbHy_A52FW0

[inner crest]

@round jay look at this video:
https://www.youtube.com/watch?v=ptEZQrKgHAg

it would be awesome to have the same thing with XR-Animator

But now, I understand for hand we need depth + canny + openpose

[radiant field]

Need help in deforum

After new update of deforum in automatic 1111

Its not like an animation anymore

It is giving me result like every img is different

[warped ridge]

So I had a thought. Batch blending can produce some more consistent but blurred/outlier frames. Rather than blending them together, you perform SSIM or histogram comparison across the batch images, comparing them to the previous frame. Rather than smashing them together, have them fight it out to see which is "most alike" the previous frame.

Doing 6-batch generations, you'd have 5 additional frames with a chance to be more consistent with the previous.

So far, success is "meh." haha

A bit less flickering, but overall probably not worth the sextuple generation. Maybe if variation strength could be tuned or a better image comparison method.

[fading turtle]

https://imgur.com/a/FrkmiDc

[fading turtle]

I love this gif. this looks like a wild stress test for frame 2 frame

[warped ridge]

Seriously. The motion, varying colors of subject and background.

[fading turtle]

I might give it a try with my current test setup

im curious how much it will break haha

[rugged prism]

Read the faq

Answer is there

[odd spindle]

flowers in space

https://giphy.com/gifs/rain-high-cannabis-B8lrRCcd4gpcFgdV0W

[slender spruce]

Sad Keanu singing to himself

[slender spruce]

Could this work across time instead? Like blend 6 frame steps together with weights by histogram similarity or something. So still one generation pass

[fallen frigate]

Motion Comic created with Deforum (text in Premiere pro)

[noble quail]

https://youtu.be/QWEiCPMi4kA

[warped ridge]

I think this is how traditional deflickering works (exponential moving average of groups of pixels across images). I'm exploring options to reduce the occurrence of flickering, not reduce the appearance of it.

[tired forge]

Dunno what you've done done to have this good result with hands. I tried everything in Blender, hands are still bullshited

[round jay]

canny edge, controlnet

[warped ridge]

@round jay are you high

[round jay]

Let's do something about guitar strings that tend to break down with Controlnet!

[tired forge]

Where do you find this 3D models?

[warped ridge]

what are you doing with these?

Also, FYI, you can stack them pretty darn high and exceed the 2048x limit with some very minor code changes

[round jay]

https://www.reddit.com/r/StableDiffusion/comments/119o71b/a1111_controlnet_extension_explained_like_youre_5/

[round jay]

2nd image is photoshop/ illustrator, anything that is used to edit images, you just edit the edge XD, then you apply the fixed canny map in controlnet

[tired forge]

Yeah, i know how yo use ControlNet with multiple sources, this is why i asked you.
I don't speak japanese x), i'm french

Hmmm used lens, i will try it tonight, thanks 🙂

[tired forge]

@round jay
What did you use for the blue haired girl's hands with the gun and the knife? The hand with the forearm does not look like the hands made with the skeleton of "Hand & Foot Depth"

銃とナイフを持った青い髪の少女の手は何を使いましたか?前腕のある手は、「Hand & Foot Depth」のスケルトンで作られた手のようには見えません。

[faint rivet]

https://www.youtube.com/watch?v=rhQFJRt75xc

[warped ridge]

Lost

Added an inpainting feature to gif2gif. Makes it easier for me to paint as little area as possible.

[brave ingot]

https://www.youtube.com/watch?v=rZpMomuBzTY&ab_channel=agathacrusty

[faint rivet]

https://www.youtube.com/watch?v=UglNAHKo7ZA

[fallen girder]

"The Path" | Deforum Animation (Stable Diffusion v2.1) [8K] - Workflow in the video description
https://youtu.be/VWcPEC44VJk

[slate swallow]

of a singer (╯°□°)╯︵ ┻━┻

[round jay]

Blender and Hand Mapping, openpose, depth, canny, Controlnet https://github.com/jexom/sd-webui-depth-lib

[pallid lance]

batflash haha

[faint harness]

ControlNet HED model, In my custom Deforum notebook!

https://www.tiktok.com/t/ZTRWjYHYS/

Damn stuck frame lol
https://www.tiktok.com/t/ZTRWjqKja/

[daring warren]

https://www.instagram.com/p/CpVLLskI-bo/

nofacekolektiv

Just a Stormtrooper vibin' to a banger🔥

🎨Stable Diffusion, Deforum.
🪄Topaz labs, RunwayML, FILM, edited with.

× Since IG considers that my content sucks, a share is always appreciated, it's becoming really frustrating how I'm struggling with the account 😅🫶🏻📲
× Creative process was a little different this time, including the amazing FILM interp...

Instagram

[inner crest]

I could do that even before COntrolNet...

When it will be possible to use it for an animation ????

https://tenor.com/view/when-today-tomorrow-in50years-confused-gif-16929382

one image is for pussy

7500frames is for real men

having great hands in SD2.1 is easy

soooo

all these tricks to get that with stupid SD1.5 models are not really useful...

and yeah it is a kara kesh

https://tenor.com/view/vala-mal-doran-claudia-black-stargate-sg1-tumblr-smile-gif-21442308

[misty dagger]

2.1 animation

[faint rivet]

1979 Antarctic Expedition Tribute Animation - Deforum Stable Diffusion https://youtu.be/Znme8_FC8b0 via @YouTube

[fading turtle]

https://imgur.com/a/5rywD0L

[zenith roost]

Hello there- what size/smapler/steps do you recommend for quick prototyping? something that give decent results but is fast generating to give an idea of the scene w/ zooms/translations?

[daring warren]

https://www.instagram.com/p/CpXwpPwoLPD/

nofacekolektiv

California dreaming 🌅

Important update:

Hey there people! As you may know this account and instagram always had a difficult relation, and since some things changed in my life lately (for good), I will have less time to dedicate to social media. So I decided to prioritize other channels and means for growth where I don't find any kind of "limit...

Instagram

[vernal topaz]

I tested StableDiffusion with ControlNet (depth preprocessor and model), and I think it's very capable. I didn't pay much attention when I started the batch, but the result works really well. It's a random seed for each frame, which was intentional. https://www.youtube.com/watch?v=JAmDANA21Yc

[mossy ridge]

Creare un booktrailer

#booktrailer smile

#1072220168534642768 smile

[jade prawn]

That's incredible, I've just got SD setup and generating images in TD, still a total newbie at it but the power and control is amazing, thanks for sharing the clip.

[round jay]

Mask Controlnethttps://www.reddit.com/r/StableDiffusion/comments/11hhnqs/multidiffusion_region_control_a_prompt_on_each/

[round jay]

1. Make two line drawings, one for the front and one for the side
2. Create a middle split with deforum
3. Paint the output line drawing using ControlNet
4. Specify the lip and eye patches at the prompt when coloring All have the same seed value

Note:
Background of the final color picture is removed with i2i's batch function + ABGRemover

https://github.com/Kryptortio/SDAtom-WebUi-client-queue-ext
https://github.com/KutsuyaYuki/ABG_extension

[fossil dome]

Make a Apple

[upper geyser]

https://youtu.be/_HvYXAF4Ea8

[ocean rose]

Hi all, I am trying to generate a Controlnet animation in a locally hosted web-ui instance. I can generate single images no problem, but whenever I feed an image sequence to it - it crunches through all the images, but the controlnet reference image does not follow the sequence (so the model is locked in a single pose). Any idea how to fix it? Thx!

[spiral knot]

I've been hand puppeting my stuff in after effects. https://www.youtube.com/watch?v=nTkZJ_XhGeI

[plucky plover]

controlnet img2img side by side! https://www.youtube.com/watch?v=5TWbSpxmyRA

[little otter]

.makeimg"man"

Status

[slender spruce]

Time for that Game of Thrones/Muppets crossover

[inner crest]

SD2.1 and a powerful embedding...

[dense ocean]

If you go to scripts at bottom of img2img, do you have controlnet m2m script? I think that is for this use case though I haven’t used it

[dense ocean]

Nice, is that using the Corridor Crew method of training yourself in particular style and using video deflicker at the end?

[plucky plover]

nope!!! this is just a regular model with multicontrol net. no deflicker.

[dense ocean]

Oh interesting, you don’t have much flicker considering. So just a consistent prompt and feeding canny/depth/pose of the original frame in?

[plucky plover]

correct! i use normal map and canny models.

i have a 30fps input video and delete half the frames . render at 15 fps and interpolate it back to 30

[dense ocean]

I feel like there has to be more ways to improve things even more. Like some process that generates 10 renders for each frame and takes the one most like the previous frame.

[plucky plover]

that would be really cool 🙌 another thing that is needed is ability to put in multiple prompts like how you can in deforum. with batch img2img it's one and done

[dense ocean]

I need to try deforum

[dreamy current]

Added a feature to the Seed Travel script so you can use Structual Similarity Index Metric to fill in frames instead of having to guess how many you need. Just set a value of how similar you want the frames to be and it starts adding them until it is good enough. (Or until it has to give up, since sometimes you don't get a better candidate frame no matter how small step you take. This is why it "stutters" a couple of times.) And sorry for the potato quality. 🙂 Rendering video only using SD takes too long on my poor gtx1080ti to make it hi-res.
https://www.youtube.com/watch?v=4kraYJVsx8Y

[warped ridge]

So cool. I've tried SSIM and histogram comparison similarly, but only on images created in batch. "Generate until you meet a SSIM threshold" is awesome idea.

BTW SSIM became much more reliable when only comparing the middle of the image.

[dreamy current]

Hm. That would be nice to try. Might get rid of (some) of the stutter. 🙂

[dreamy current]

Seems to work really good, new test-video should be done in 2-3 hours.

[hot umbra]

10 minutes of slowly zooming 4K deforum animation with signalis music
https://www.youtube.com/watch?v=uGHnxC9QuIY

[warped ridge]

the framerates being out of sync makes me smile more than anything

[dreamy current]

Pushed to github now. 🙂

[warped ridge]

Woot. I just did the middle 50%. Split it into 4 columns and SSIM'd the middle two. Probably a more elegant or efficient way. Totally going to dig around and see what you did.

[dreamy current]

I did it the simplest way possible (for me). You can select a square in the middle that is X% of the height or width (whichever is smaller). Or use the entire image.

[inner crest]

just a test. Now the next level is to master Parseq...

ControlNet + Deforum + Parseq and my power will be absolute

https://tenor.com/view/im-going-to-be-god-lucifer-morningstar-tom-ellis-lucifer-im-gonna-rule-the-world-gif-21415599

[inner crest]

ControlNet works in Deforum...

https://tenor.com/view/jiren-full-power-gif-21883776

Nothing will protect you from my outstanding animations, poor mortals...

[sudden phoenix]

See that's the problem with being neuraldivrigant. The first time I read that it went into my head as "Nothing will protect you from my outstanding animations and poor morals"

had to read it twice 😛

[hearty tundra]

I noticed today, but is there a way we can start an animation from init image? I couldn't figure it out

[faint rivet]

https://www.youtube.com/watch?v=rxDh_WCujbI

https://www.youtube.com/watch?v=bvFDyCfIKXg

[ebon sparrow]

COOL!!
What music is used in this video?

[inner crest]

Of course, there are also possibilities to insert images during animation. It is what I try to figure out.

And since there is a scribble mode in ControlNet. You just stop your animation. Draw what you want. Resume. And boum you are the god of Deforum.

[fading turtle]

just realized i never shared the results
https://imgur.com/a/l3vpIAe

the colors / gif quality does WILD things with the frame matching

https://imgur.com/a/CfuIr7N

[warped ridge]

https://twitter.com/TheBoogley is the original artist FYI

He has some fun/interesting animations that really challenge my processing ideas

Speaking of which, https://github.com/LonicaMewinsky/gif2gif now (finally) outputs completed GIFs into the WebUI results gallery.

[karmic atlas]

Hi everyone, we're blnk studio where we use generative AI to produce animations. So far we've been producing only music videos, but in the future we would like to produce an animated series as well! And hopefully using gen-1
see our newest music video and leave a comment 😄
https://www.youtube.com/watch?v=bspwTALOBws

[faint rivet]

Luckily, I still had the file. It was Organic Guitar House - Dyalla.mp3 from YouTube Audio Library... no attribution required for that one.