Parseq tutorial 1: Fine grained control of Stable Diffusion and Deforum - Keyframing & Interpolation

Parseq is a parameter sequencer for the Deforum extension to the Stable Diffusion Automatic1111 UI. You can use it to generate animations with tight control and flexible interpolation over many Stable Diffusion parameters (such as seed, scale, prompt weights, noise, image strength...), as well as input processing parameter (such as zoom, pan, 3D rotation...).
- Access Parseq here: sd-parseq.web.app/
- Source and docs (including setup instructions): github.com/rewbs/sd-parseq#re...
- Import these keyframes to start from a blank doc: gist.github.com/rewbs/cacdc99...
00:00 Introduction
01:46 A tour of the Parseq UI
06:00 Starting from a blank slate
09:32 3 important settings to check in Deforum!
10:45 Our first animation
11:05 Predictable seeds and auto-uploading
13:16 Adding a 3D zoom
14:50 Tweaking interpolation functions
17:08 Adding rotation and translation
18:09 Next tutorials...

[@latentessence]

there's no AI replacing you soon, man. Amazing work and documentation

[@tradehut2782]

Not AI but more intuitive apps are bound to come for deforum. Its always lik😊e that

[@corysimpson5825]

In 2 years this video's gonna have 10M views. Enjoy the coffee ;)

[@benasido7621]

100% agree

[@Soundream]

Really great stuff Robin! Looking forward for the next tutorials, really great tool! Parseq - audio analyser is also really cool! Hope you can dive in a little bit on it! Keep the great work mate 👍

[@sL1x999]

you are nuts! amazing work thank you!

[@alekseyzhuravlov]

Fantastic! Tweaking interpolation functions seem to be extremely powerful, looking forward to reading the docs

[@rolyars]

Thanks for all this great work!

[@LibeShorts]

cannot wait to have the next tutorial!! thanks you very much!!!

[@Pauluz_The_Web_Gnome]

Great stuff….gonna try it tonight!

[@PrototypicalFrequencies]

Great tutorial! Thanks so much for writing this program and sharing it with us!!! And a walkthrough too 🥰

[@kcelewicz]

Robin - this is amazing! Thank you!

[@iLEZ]

Cracking application, and an even crackinger tutorial. I can see that some UI elements have been added and changed, but the gist is VERY clear and concise. You can tell that you are rightfully proud of your great work here. Thanks a million!

[@mahoognus]

Wish I knew about this sooner. I've been using math functions in Desmos + trial and error to create smooth motion in Deforum. Amazing work! Thank you!

[@delayedchaos]

Thank you so much! For however my brain works this has been the cleanest most intuitive iteration I've seen. I really struggled going in and adjusting minor settings in A1111 and I'm kind of stubborn so I don't want to ask stupid questions of the deforum devs. This makes it so I don't have to ask those stupid questions and I can directly interface with the frames in a visual way and I appreciate that so much. Thank you again for all your hard work on this!

[@ronnykhalil]

yas! really terrific primer. Cannot wait for the next ones!

[@vellov2]

Absolutely love this! Thank you!

[@GrandHorseMusic]

This is amazingly helpful, thank you so much for your work and explanation. I'm often left baffled by tutorials but I got this working first time. Very much appreciated!

[@matsuda206]

Fantastic tool and tutorial, thank you so much!

[@friendofai]

This is exactly what I was looking for, about to dive in now. Thanks so much!

[@user-fn8wl8de9f]

Wow, man, just amazing! I am so happy to see a tool like this one created. Feels like magic

[@throughiris8676]

Wow, this is amazing, as a professional animator / CD. I wish you designed all the other programs I use! Great work!

[@ATLJB86]

Your the man Robin!

[@Energy_Sheep]

感谢大神！赞！Thankyou for this lesson🥳

[@Braunfeltd]

cheers, this is great. learned basic deforum than ran into this on my feed and blew my mind on the possibilities now. this helps alot for people like myself that are visual rather than trying to see it in our head haha. thanks

[@maltas22]

Great tool thx for your effort.

[@deeceehawk]

Thank yo uso much! I'm studying the options, awesome job Robin!

[@arothmanmusic]

WHOA. I just started toying with this tonight. It's incredible!

[@jungresonant9772]

You bloody genius! I've been wanting something like this since I first started with Deforum last October. The trial and error aspect can be really frustrating and Parseq will greatly reduce the unwanted randomness. Also saves me a hell of a lot of time typing out hundreds of keyframe coordinates for my beat based work. Thanks, will def donate a coffee.

[@hi4620]

Very useful extension! Much more convenient than the standard deforum

[@AstralProjects]

Thanks for this!! I mean the whole thing, explanation and all. I've been trying to run animations for weeks with unsatisfactory results.

[@edoard3]

you are a god

[@Archilives_UE5_Ai]

thanks for this great thing, i have been using deforum for a few months now, but parseq is really hard for artists, hope with your efforts will bring simpler

[@InkDropFalls]

Im Back .....Just wanted to give you some KUDOS ...the changes you made to new UI since this video ...have been fantastic ...everything so far looks like improvement.......Nice one MATE ... would be handy to have another audio analyser video with voice to explain how to use it ....of course i understand your busy with updating and changing and improving app so video content is going to be slower ....You doing good stuff mate .....hoping i get good enough to help bulk out some tutorial videos to help promote it

[@shlomitgueta]

thank you very much i will try now :)

[@optimoos]

I have a comment :) God bless you for this! the amount of work needed to create such a tool... thank you so much for this! I've been busting my ass to create better Deforum animations, and your tool will help bring it all to the next level. you rock!

[@Leyv_Ai]

Hi

[@opticonor]

Incredible!

[@Cazametroides]

This is awesome!!!!

[@digital_magic]

Awesome tool and tutorial,..i am very impressed by it 🙂

[@nikarabodywhisperers]

awsome stuff bro

[@Julz_ru]

You rock mate

[@JuanDArango]

Bro you are a good person to humanity. Thanks

[@Nalestech]

Wow! Thank you! for your effort!

[@rewbs]

My pleasure!

[@faddlewaddle2615]

Most excellent! Thank you. Oddly, I had trouble switching to 3D and carrying this out (whilst using the default setup). Would get an error. Had to bring in one of my own 3d motion projects and take it from there. From there? Awesome!

[@PASSENGEROFSHITOFFICIAL]

great thankyou so much !!!

[@Tedrer]

Thanks for that!

[@VisualsByMasaya]

This is a fricken game changer.

[@InkDropFalls]

Hope you do the audio synch video.....Really excited to learn more ....

[@elifmiami]

This is the most amazing tutorial I have ever seen! It also seems like a really useful tool for stable diffusion. I can't wait to watch more! (Btw I'm really curios how to do combining 3D y-axis rotation with x-axis pan to rotate around a subject)

[@JuanRamirez-di9bl]

This!

[@rewbs]

Thanks! for the rotation around a subject you can find an example here: github.com/rewbs/sd-parseq/discussions/106

[@davewoodhead2452]

Thank you Robin for amazing work. Your web app makes controlling Deforum so much better. You've instantly improved the quality of my animations. This tutorial is excellent, fun and easy to follow along, I've learnt a lot this morning. The moment you changed a curve from linear to cubic spline was like wooooah! Can I ask if there's a modifier that can create an exponential curve? I use the handles in After Effects curve editor on every job to make those dramatic but smooth changes to interpolation. If there's a way to do that in Parseq that'd be amazing.

[@rewbs]

Hi! Thanks for the kind words, I'm very pleased you're finding it useful. For curves, check out the bez() function for Bezier curves. Tutorial 3 shows them in action.

[@davewoodhead2452]

@@rewbs exactly what I need, thank you so much, this is a game-changing tool you have created.

[@jordyandrews]

hope part 2 is out soon!!

[@rewbs]

Just published! :) kzbin.info/www/bejne/hqe6lJ6KZbWcoZo

[@Leyv_Ai]

Like si hablas español y usas Parseq, gracias Robin, eres el Steve Jobs del SD

[@jorn-jorenjorenson5028]

Wow, started with deforum animations a couple of days ago and now I stumble over this tutorial (thanks, YT algorithm!) and that is excatly what I need! Thanks so much for sharing this tool and for the tutorials! Btw, parseq immediately reminded me of parsec on my old Ti99/4a. Is it coincidence, smart AI or you putting that green space ship sprite in the intro?

[@rewbs]

Thanks! Haha not a coincidence at all, playing that game is one of my earliest memories. :)

[@jorn-jorenjorenson5028]

@@rewbs Awesome, surely the best game for the Ti, I loved it! Until those really mean and very quick alien ships came... : D

[@karakadir8860]

wow I have no words, this is so amazingly well implemented, I so much appreciate that! did you create this alone or did you have some help?

[@rewbs]

It's a solo project for now but contributions would be very welcome! Note that I'm using a lot of preexisting libraries so kudos to all the authors.

[@DonaldDrennan]

This looks amazing and I can't wait to try it out. One thing I immediately wondered is, would it be possible to take a keyframe curve from Blender and use it the way you're using equations? and conversely, export these equation keyframes from Parseq to Blender? It might be interesting to combine things created in Blender with Parseq backgrounds for example.

[@Navasundar]

I’m trying to figure out ai animation out for a media class I’m teaching high school. One way or another we are going to be exploring ai, ai image generation, prompt engineering and animation. It was a stop motion unit but give the explosion of ai and ai tools I threw it out and we are just free styling our exploration. See this has me at least hopeful I can get students excited about what we are doing by seeing actual results. Thanks. Any tips on setting up 1111 and SD on our workstations would help too. I’ve had a lot of hurdles organising dept. giving access to websites and stuff, but I am determined to make what they are learning relevant to world they are inheriting!

[@colorinspaceambientmusic]

Excuse the thunder lol. Only a Brit would apologies for the weather.

[@karakadir8860]

definitely gonna get coffee from me

[@emankhalid3948]

love u

[@maxinebishop]

Люблю математику

[@Jerichozis]

Fun fact, a guy Xpuct who made Deliberate model hates Deforum with all of his heart and saying that deforum videos are epileptic bullshit. Such irony that deliberate videos are best, lol=)

[@--signald]

Great tool, Robin. Thanks so much. Quick question: you added some 3D parameters at the end, but I didn't see you switch the Auto1111 Animation Mode to 3D. Did you do that and forget to mention, or did that final animation happen in 2D mode?

[@rewbs]

Yes you're absolutely right, in this tutorial I forgot to show the switch to 3d mode in the Deforum UI. Well spotted!

[@jcaballeroramos]

Thanks! Amazing but is working for google colab version of deforum? Or is it only for A1111?

[@bsharamistreh8348]

Hey Thanks for the amazing content, i have a question regarding the values of different parameters, for example translation x,y,z or rotation x,y,z sometimes it is by pixels change and sometimes it is by degrees and for zoom in 2d mode it is a value between -1&1. what happens in parseq when we exceed the value that is allowed and does "show as % of max" still works when we enter higher value than needed? is there a way to see all information about values, because this confuses me alot. Thanks :)

[@JosephEudave]

Great work you have done here. Can it work with Comfyui?

[@studio85amsterdam]

Thanks man really interesting. Two questions, you mention strength as how one frame efects the next, where do you have this on de forum normally do you know. I never used this. Also you mention I can do this when running colab remote, I thought Lab1111 is when you have it running locally, how do I set this up? Any tutorial in mind?

[@studio85amsterdam]

ok I found out the strength parameter is strength schedule. this is really good for me to know but i do get wild animations typically. i wrote in your other video too, the one talking about music sync is muted. only hear drum sounds, is that how its supposed to be. i dont completely get your point about the seed. i thought keeping it constant was supposed to make an animation consistent. if the seeds are sequential number does that mean that they are similar to each other. never seen a guide that says seed number 1-1000 is like this 1000-2000 is like this, is there such a guide. sorry i asked a lot of questions but im average knowledged on the topic and want to get better.

[@rewbs]

Hey thanks for the questions! Strength is a really important parameter and finding the right value is always going to involve some trial and error as you balance between consistency, artifacts and responsiveness to prompt changes. As for the seeds, no, whole numbers close to each other will not result in more similar images. Any number 0 or above is a valid seed. I like using sequential seeds but there's no real benefit over using any repeatable sequence of numbers. Yes the the other vid is supposed to be that way, I didn't have time to do a voice over. I will do a proper tutorial on that topic soon. As for a1111 and Colab: a1111 is "just" a python program you can run locally if you have the hardware, or in a hosted environment like Colab with a bit of tweaking. There are a few Colabs set up to run it, e.g. github.com/TheLastBen/fast-stable-diffusion

[@user-bv3jh8jw9n]

If I use my own image as init, what strength value for the init should I use in UI?

[@mannychop]

hi im sure this is some pilot error over here on my desk, but I don't see Parseq in my Deforum > Init area. Is there something else I need to enable somewhere?

[@mannychop]

yep pilot error, I had to update my auto1111 and now its there

[@Tigas4ever]

I'm so much the guy in the animation... lmao

[@Tigas4ever]

I feel much better now! :D

[@dagovegas]

Hello! I have a problem when adding the zoom translation_z at min 13:20 . When adding the values 1-128, nothing changes in my video, (it's not zooming-in). I've notice Robin changed the animation mode to 3D and i've already did that but still my video doesn't zoom in. Can somebody help me or been to the same issue?

[@cashyappurushottam427]

Great tutorial! Though I'm having a problem when using 3d . there's an error showing in the UI, and the cmd window says "ValueError: Number of output components does not match number of values returned from from function f" ?? What's happening Sir?

[@rewbs]

Hey! That's most likely not related to Parseq. I recommend jumping on the Deforum discord and asking for help there, or raising an issue on the Deforum GitHub.

[@UnashamedlyHentai]

"Copy From Txt2Img/Img2Img" buttons under the prompts would be great

[@jasonsutis3387]

Could I somehow use this to do an image to image at specific frames? I have 3 images That i want to generate in betweens of each one morphing into the next one. So i have two images and i want to start with one and end up with the other over 30 frames.

[@rewbs]

There is a feature called guided images in Deforum which can help with this. Or, you can create a slideshow video of your images and use it as an input video on hybrid video mode.

[@p0ckettv]

Where is that comment you mentioned with the starting code?

[@nikarabodywhisperers]

Im running through this tutorial the 3rd time , Im doing everything the same , but results are any thing but consistent ... Any suggestions ?

[@rewbs]

Hi! Sorry to hear that. Could you join Discord (follow the chat link on the Parseq UI) and go to the Parseq support channel, and then share your full rendered Parseq data and Deforum settings file? Or if you don't want to join Discord just paste those files somewhere and I'll take a look.

[@j3ffn4v4rr0]

Is there a way to use Parseq with the Colab version of Deforum?

[@rewbs]

No, parseq currently only works with the a1111 extension. However, there are ways to run a1111 on Collab or on other services like rundiffusion and runpod.

[@chrispennie5830]

Thank you for this program!! It is incredible!!

[@Grumpy-Fallboy]

all this AI advance reminds me "Blade Runner movie 1982" Enhance Scene, where he uses computer to change angles and detail on a picture.

[@ohaderprofi]

hello, thank you very much for your tutorial, unfortunately I get after inserting the Copy output from Parseq to deforum, this error message: "UnboundLocalError: local variable 'parseqName' referenced before assignment" this is displayed after I pressed the Generate button, do you know what I can do? thank you very much. unfortunately it does not start to generate in deforum.

[@rewbs]

Hi, this usually means you need to update your Deforum extension. Could you please make sure you are on the latest version and let me know how it goes?

[@ohaderprofi]

@@rewbs Thank you very much for the tip and your work!!!, it worked(partially).... (more about this below). My steps were as follows: 1. I edited the file "webui-user.bat", opened it in the editor and added "git pull" to it (to enable automatic update). 2.Then i checked the extensions tab for update for Deforum, it was up to date. 3.Then i downgraded my pyton version(from 3.10.10 to 3.10.6). ***Then I found out*** that I should have all options turned on in the Parseq option: "Managed fields" (so that all settings are done by Parseq). Then the error ("UnboundLocalError: local variable 'parseqName' referenced before assignment") doesn't happen anymore. ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- I have more questions: 1.Do you know if it comes to problems if one would use quite many prompts? For example over 50 prompts? 2.would it be possible to add an option to the prompts to enter also times(seconds)? so instead of: From; Frame: "0" ... To; Frame: "120" to From; Sec: "0.00" ... To; Sec: "5.00" . it's totally late, must go to sleep urgently ^^

[@Richard-hx2fv]

can i use it in comfyui？

[@mertakkoca7309]

"UnboundLocalError: local variable 'parseqName' referenced before assignment" I am making a mistake. Can you help?

[@rewbs]

Hi! Could you confirm you're using the latest version of the Deforum extension? That issue should be resolved in some updates that were released about a week ago.

[@mertakkoca7309]

@@rewbs The issue was fixed after upgrading, thank you. :))

[@Tedrer]

Hi! I didnt found some trubleshoot. so Im asking here. In the last run I getting this problem: RuntimeError: The size of tensor a (336) must match the size of tensor b (320) at non-singleton dimension 1, trying to understand why... Thanks!

[@rewbs]

Hi! This is most likely not related to Parseq but rather the underlying invocation of Stable Diffusion. I've seen this error when using unexpected dimensions for the input image or the desired output.

[@Tedrer]

@@rewbs Yes, it is the input dimensions! Thanks!

[@Tedrer]

@@rewbs Hi again! Another issue, I got this line: ValueError: eta must be 0 for reverse sampling. Any idea? Thanks for all the effort!

[@rewbs]

@@Tedrer Hi, not sure about that one, but I suspect this one is also unrelated to Parseq. Perhaps you could try using another sampling algorithm, or check the a1111 settings to confirm "eta (noise multiplier) for DDIM" is 0?

[@Tedrer]

@@rewbs Hi! Yes it is 0. its happen again right now when I changed some of the Parseq dimensions schedule values...cant point my finger on the problem source

[@randomprocess7876]

Hi guys. Any colab link which has paseq?

[@rewbs]

Any Collab that runs a1111 and the Deforum extension can be used with Parseq (Parseq is a standalone web app which produces output that you supply to the Deforum extension).

[@user-vp5hi9wu2k]

I am getting this error"UnboundLocalError: local variable 'parseqName' referenced before assignment". Any solution? Thank you

[@rewbs]

This usually means you are on an old version of the Deforum extension. Can you try updating it?

[@user-vp5hi9wu2k]

@@rewbs Thank you for the reply. I already updated it and its working. Parseq is really wonderful and thank you for the great effort. Really looking forward to your music sync tutorial. 🙏

[@InkDropFalls]

forgot to mention have to turn on 3d in deforum latter on towards the end .....in doing so you get an error TypeError: argument of type 'NoneType' is not iterable ..any idea why how to fix that ...sort just made last hour of following steps a waste o time

[@rewbs]

Hey, sorry to hear that. Could you possibly join the deforum discord and paste the full error message in the Parseq support channel? Or alternatively raise an issue on GitHub? discord.com/channels/1011314943380959313/1050887765799207043

[@InkDropFalls]

@@rewbs SOLVED so in video when edit 3d ....you have to turn on 3d in deforum ...it does say on perseq .... However as said i get error .... Solved it by ..TICK OFF .... 3D Depth Warping: use_depth_warping OFF ... can be found just above place you paste perseq text in

[@tomweller6352]

@@InkDropFalls Thank you so much for that information!!! I was spending days(!) in finding the solution to this error!

[@InkDropFalls]

@@tomweller6352 happy you sorted it ...that made my comment worth while ...thanks for reply i do often think if any of my comments have a positive impact and worth making .....I guess they do ....Thanks too

[@maxinebishop]

How do you make it to 500 frames?

[@rewbs]

You can just change the frame number of the last frame.

[@AlbtzGaming]

suggestion 1: background sound is bit loud

[@rewbs]

I agree: the sound isn't great in this tutorial. I got a better mic for tutorial 2 onwards. There's still background music but hopefully my voice is clearer.

[@technoprincess95]

well it is very hard to understand, i've watched days, still confused.

[@deulin]

How to make the number of frames more than 119 ????!!!! What is the point of making such a powerful tool with such a stupid limitation???!!!! Really, all the people in the world make videos 120 frames long????!!!!

[@deulin]

For those who suffer with this issue, there is a life hack. You need to go down to the "Keyframes grid for field value flow" and change the number of frames in the last line of the table.

[@rewbs]

Yep, there is no such limitation. It's just a matter of changing the frame number of the last keyframe.