[Edit:
Version v5.0 works with latest comfyui (v0.15.0).
If you have any problems, please refer to the FAQ at the bottom of the page or have a look in the comments.
Many thanks to everyone who tested this workflow. Thank you very much for the many inquiries and, of course, for all the knowledge and experience you have contributed. here👍🙂
Special thanks to:
@SeoulSeeker for the "Dead Simple MMAudio" workflow wich are the basis of the audio part here,
@taek75799 for the really well working enhanced models
@Bakazaya pointing to the color issue in version v3.0 and running lots of tests,
@bluntfeather sharing latest experiances with installing Comfyui-Easy-Install,
@nitrovtx for remain persistent in matters of quality and running a lot of tests,
@Icey64 for providing the link to "Comfyui-Easy Install",
@boinobin730 for asking for a First to Last Frame option, running pre tests and responding fast as hell 🙂 and
@SnowShoes311 thank you so much again for all your buzzing 😋]
Features:
Optimized Wan 2.2 workflow, runs perfect on RTX 3060 12 GB VRAM GPU and 32 GB RAM,
"Text to Video", "Image to Video" and "First/Last Frame 2 Video" generation in one workflow, all with easy audio generation,
easy installation/model downloading, all necessary sources are specified,
easy to use workflow, clearly structured, all necessary steps are explained,
easy switches for mode selection,
easy prompt selection for fast prompt creation/testing,
easy switching between "standard" and "enhanced" models,
very fast and smoth high quality outputs up to aprox. 1440 x 960 with 60fps,
2x fast upscaler,
4x fast framerate multiplier,
MMAudio Sampler (generates sound accordingly to the video action),
Triton and Sage Attention option,
A 5 Second long high quality video generation takes about 10 - 15 minutes (see below).
Tested generation times:
As a rough guide value for RTX 3060 GPU: generating a 5 second long high quality 1440 x 960 60 fps video with 6 steps it will take:
t2v: around 10 - 12 minutes,
i2v: around 15 minutes.
Comfyui-Easy-Install with Triton + SageAttention:
This workflow should work with any latest comfyui version >v0.6.0 (Desktop, Embedded, Windows/Linux).
However, comfyui is developing rapidly, and it often happens that some of the custom nodes used are not updated quickly enough or not updated at all. Manual workarounds are sometimes necessary. Furthermore, care must be taken to ensure that there are no conflicts with other nodes.
If you're having difficulties with your existing comfyui system or if you want to run video generation on a separate (parallel) comfyui system, like I do, I would recommend you the following installer: https://github.com/Tavris1/ComfyUI-Easy-Install.
Complete installation of comfyui including manager and some pre configured custom nodes is just one click - really 🙂
Installation of Triton + SageAttention is just a second click - really 🙂 And since it's so easy now, I would definitely recommend it to you for video generation.
Cause it is an embedded version, you can install it parallel to your existing comfyui version without the risk to ruin your working system.
After installation just configure the "extra_model_paths.yaml" file to use your existing models.
After a fresh installation of Comfyui-Easy-Install you might have some issues too, but there are known workarounds - please see the FAQ below.
For testing/understanding/experimenting/changing the workflow:
Click "Toggle Link Visibility" to see the links.
click the Subgraph symbols to open the Subgraphs.
for quick testing you may lower the settings for: steps, clip lenght and video resolution,
be really carefull with modifying Groups or Subgroups (even Titel or Color) cause they are essential for switching,
feel free to try and test other models. Just give me a hint if you find models which deliver better results and fitting the 12 GB VRAM limit.
And as usual: Have Fun 🙂🙂
Short Conclusion:
This workflow is based on elements of a variety of allready published workflows. My "job" was only to put things together, optimize it for a small machine and create a most simple and hopfully user or even "beginner" friendly workflow.
I`m not an "expert" - just a user who wants to get it running on "available" hardware.
There are many things I don't really understand. If you find mistakes or better solutions please give me a hint.
And I really hope that even "beginners" have a chance to go the first steps...
Frequently Asked Questions (FAQ):
For quick and better overview I will try to merge all known issues here - step by step (please be patiant). If your issue is not listed here, please have a look in the comments first. Most issues have been allready discussed.
Comfyui Nodes 2.0:
Turn off Nodes 2.0 in comfyui (use comfyui menue). Actually not all custom nodes are supported.
Comfyui crashes after generation while vae decode, upscaling or frame rate multiplying (Rife VFI) without any error report:
This is a RAM problem (not VRAM). Increase your swap file (min. 64 to 128 GB) or set it to automatic management on a fast drive with at least 100 GB free space.
JW Nodes (JWFloatToInteger, JWIntergerDiv, JWImageResizeByLongerSide), soundfile missing:
For the workaround look here and here:
python -m pip install soundfileFresh Comfyui-Easy_Install Installation (missing soundfile and Pytorch v2.9.0 issue with SageAttention on Windows:
For full conversation look here.
Open cmd in python_embedded folder:
python -m pip install soundfile python -m pip uninstall -y torch torchvision torchaudiopython -m pip install torch==2.8.0 torchvision==0.23.0 torchaudio==2.8.0 --index-url https://download.pytorch.org/whl/cu126Slider Nodes - how can I modify the "default" values:
Right click the slider node, choose Properties and set the values you like 🙂🙃
Description
Completely redesigned workflow for easier use.
Same functionality as version 4.0 + easy on/off switch for audio generation.
Works with latest comfyui versions (v0.15.0).
FAQ
Comments (71)
Why is there always a kind of bright flash/flare at the end of the video when I use frame to frame? Can it be fixed?
Can it be fixed? Not really. Try other images and prompts or go the easy way and use a cutting tool to cut off the parts you don`t want.
love the work your put into this workflow its amazing. but i ran into a problem and i have no idea how to fix
i keep getting this error:
BigVGAN._from_pretrained() missing 2 required keyword-only arguments: 'proxies' and 'resume_download'
Hi- thank you. Did you googled your error message? You have an issue with MMAudio. Comfyui and all nodes up to date? Any node conflicts?
@arkinson Yes I was able to figure it out Was able to use Chatgbt to figure out the exact problem I am going to post the solution
Thanks for this workflow it's amazing. I personally don't understand where the upscaler is in the workflow or how I still need to learn how to prompt the audio better but you did an amazing job.
I'm using smoothmix wan 2.2 gguf on a rtx 3090 with 32gb ram and using the goated ComfyUI easy install after my last portable install broke. If anyone has slowdowns constantly and using a setup like mine try using --cache-none. Consistent generations <6mins for a high quality 5s video.
Hi - thank you so much for your feedback 🙂 The upscaler you find in the "Upscale + Rife + Audio" subgraph in the output group. Don`t worry too much about audio. MMAudio pruduces only very simple noises. If you want much better audio with synchronised speach, have a look at my other workflows. LTX-2 is brilliant, but there are actually not many loras availlable.
Honestly, didnt expect it but it worked out of the Box , nice job! Sure I needed to configure my own GGUF and so on but it works as planned. Thank you!
Thank you and happy generating 🙂 Let me know if you find any bugs.
I am not sure what happened with my original comment. I can't access it - the window pops up and disappears.
This has been a really good workflow for me. I am using an older version and the block swap node has been removed. I am also using an older version of Comfyui as it has a tendency to screw up my workflows.
Now, I have a 5090, so I made a couple adjustments for larger models. Then I added a wan2.1 Lora to add details. It's fast and follows commands pretty well. I am particular to the FFLF version, which helps with consistency. The details LoRa really helps with realism. Otherwise skin textures get a bit plastic looking.
I keep meaning to check the latest versions of the workflow but with my limited time, I am sticking with the older version. It's very, very good, though.
Hi - thank you for your feedback. In the later versions I only added MMAudio (check version descriptions). It is not a "gamechanger" (<-- I really hate this word 🤣) but it adds some fun to every clip with simple noises. Latest version v5.0 is a complete design and logic rebuilt, but with same functionallity as previous versions (caused by latest comfyui updates v4.x did not work anymore). And if you like to generate videos with speach - have alook at my ltx-2 workflow. There are actually just a few Loras availible, but ltx is really a lot of fun too 🙂
Thanks for the suggestion. I've been thinking of checking out LTX.
And, I will tell you - so far, this is about the best wan2.2 workflow I have messed with. Like I said in the original comment, I changed it to run larger models since I have a 5090. But, this thing has been smooth as silk.
@hdean You definitely should have a look at LTX, especially with a fast gpu. It is much faster then Wan, you can generate much longer videos and speach/audio is brilliant and absolut funny. There is just one con actually: there are only a few loras. But you can generate the first part in Wan and after it you can use LTX v2v to tell a story or some nonsens at least in one run 🙂
I`m glad you like my work. The idea to publish a workflow was born some month ago by the same problems you desribed: endless testing of odd non documented workflows wich did not work or not suite your hardware finally or terrible model mismatch, etc.
Yes, my workflows are strictly optimised for speed and quality with low hardware. And I get a lot of feedback from users with more powerfull hardware, they are doing some simple model modifications like you and are happy with "full speed" and higher quality outputs 🙂
@arkinson @arkinson Well, ya do good stuff - check this LoRa to improve the videos - 🔗 [Download Link](https://huggingface.co/vrgamedevgirl84/Wan14BT2VFusioniX/resolve/main/OtherLoRa's/DetailEnhancerV1.safetensors?download=true)
@hdean Thank you for the Lora link, will test this soon.
@arkinson Let me know how it works for ya.
It works without errors. Congratulations and thank you. With a few minor fixes suitable for my system (such as bf16 instead of fp8 clip and VAE), it does its job very smoothly and without tiring the system. Even though I find the Random First Frame Image situation unnecessary, I believe there is a logical explanation. My only question is whether the upscale algorithm can be improved... The upscale seems to add some 'pastel tones' to the frames. Maybe it can be tried with SeedVR, but I can't try it because I don't have a technical background and I don't know. Other than that, everything is wonderful.
@mracar Hi - and thank you so much for your feedback 😋
Random first frame image: For myself I used it often for automated overnight runs 😉All you need is a good prompt, ideally with some wildcards and a bunch of suitable images. The next morning, you're sure to find some good clips. 🙂
SeedVR: Interesting, but I have no experiances with it. Did you used it for yourself allready? Main questions: Is it fast? Will it run with 12 gb vram? What is the advantage against the used RealESRGAN 2x upsaler??? Do you have a special SeedVR node in mind?
Testing: Open the "Upscale + Rife + Audio" subgraph and simply replace the upscaler node. Or create a simple test workflow: "Load video -> SeedVR -> save video.
@arkinson I have no words for SeedVR advantages, sorry I can't help if it works with 12gb vram or if it is fast. As I said, I'm not on the technical side of the business. I probably need to tinker with some settings. However, your work is fast and does exactly what it is supposed to do, for which I am grateful.
@mracar You don`t need any technical background nor any experiance to create the mentioned workflow with just 3 simple nodes - just a lot of time and computing power for some serious tests 😉
I ran into a startup crash when using this workflow with MMAudio. Posting the fix here in case it helps someone else.
Issue:
ComfyUI failed to start with errors related to BigVGAN._from_pretrained() and later is_offline_mode import errors.
Cause:
Recent versions of huggingface_hub (1.x) and transformers (5.x) are not compatible with the current MMAudio / BigVGAN implementation. Stability Matrix had auto-installed newer versions, which caused the mismatch.
Fix (worked for me):
Go to:
StabilityMatrix\Data\Packages\ComfyUI\venv\Scripts
Shift + Right Click → Open PowerShell
Run:
.\python.exe -m pip install huggingface_hub==0.34.4 .\python.exe -m pip uninstall transformers -y .\python.exe -m pip install transformers==4.44.2Restart ComfyUI.
After aligning:
huggingface_hub → 0.34.4
transformers → 4.44.2
Everything booted normally and MMAudio worked.
Just posting in case others hit the same dependency mismatch 👍
@yamypro Hi - thank you for posting your fix for Stable Matrix 👍
don't have to downgrade, you simply remove hkchengrex 's mmaudio node and put kijai/ComfyUI-MMAudio instead. it has bugfix (google bug msg & see git logs)
I had a fresh comfy installation and tried this because of similar issues but ended up breaking the whole comfy and it won't boot up at all now XD
this workflow is actually so good thank you
i’m running an rtx 3080 12gb with 32gb ram and it’s literally working so well.
i set the res to 800px and the high/low steps to 2/3 and 5s and it takes like 300~350s to finish an entire video with the wan 2.2 14b q8 gguf and lightx2 lora.
@fluffywool Hi - thank you so much for your feedback 😋 I am glad it is usefull - and that`s the point it is made for 😉🙂 You might have a look at my newer LTX-2 workflow too. It is a lot of fun to play with sound and speach....
Thank you for the update.
But im getting such bad upscale results with the Audio nodes now compared to what I got in 3.1
Is there any way you could make one without the Audio nodes?
im not sure why but it just really doesn't give the results it used to.
@Megasherru Very strange, cause not much changed in the video generation part, since adding the audio part - except the Rife nodes. You can check this by opening the upscale subgraphs. Did you run some serious side-by-side tests (same prompt, seed, settings, etc.)?
@Megasherru I just did a quick comparison test. There is definately no difference in video upscale output between workflow versions 3.1 and 5.0 (even with the mentioned different Rife nodes). So there must be some issue on your side.
@arkinson Ill test it out a bit more then, maybe something I did wrong, ill check some more and see :) thank you for your quick reply
Really nice, works on 6gb vram, rtx 4050 laptop
Cool! Thank you for your feedback 🙂
Anyone know why the style of faces changes a lot? specifically, a semi-realistic rendering style input image gets turned into realistic video style. I didn't have this issue with the normal WAN2.2 base model. I'm using the Q6 versions instead of Q8, but otherwise same one as in your workflow
Which gguf models are you using? Base models + lightning loras or the NSFW Fast Motion models suggested in the 4.x workflow? The fast motion models in my experience immediately change faces. I've had the best luck with base models that have lightning lora built in.
@ainewb14 Interesting. Could you link the models you use? Do they generate good movements? Cause the standard models + lightning Lora mostly generated slow motion.
@arkinson https://huggingface.co/jayn7/WAN2.2-I2V_A14B-DISTILL-LIGHTX2V-4STEP-GGUF/tree/main
Thanks for the great workflow!
@arkinson they generate good movement, not as fast as the fastmotion models but they don't change facial features. I use them with Painter I2V node set to around 1.50 or so.
@ainewb14 Thank you for the link and your information.
how can i change the upscale model?
@fzdxgfchgvjhkjl Open the subgraph and change the model to what you want.
Hello, I struggled for 3 hours to install JW, but it's all good now. I have a model in safesensors mode, so I disconnected the 4 UNETs and connected 2 KJ ITV broadcast model loaders. The main problem is that when I click "run," it creates an image (the end image) and not the video. I can't figure out where the problem might be coming from ?
As usual, there might be lot of issues if you try to use other models. You have to check for model + loader compatibility, vram usage, etc.. Just try to run systematic tests.
anyone found "wan french tongue kissing v1 (tungue kiss).safetensors" ? :-P
otherwise, great, got it working in ~15 mins, latest git comfy on linux. 5s i2v gen time on 4080s is less than 1min
Hello! I like the idea of adding noises to the WAN video but, I've been trying to run the full of the workflow, but I don't seem to be able to get past this:
This workflow uses custom nodes you haven't installed yet.Installation RequiredInstall RequiredMMAudioFeatureUtilsLoaderInstall RequiredMMAudioModelLoaderInstall RequiredMMAudioSamplerin subgraph 'New Subgraph'I've reinstalled whole comfy, set permissions to weak, downloaded manually the node, set comfy version to earliest available which is 0.18.0, installed the version from Kijai, on comfy manager keeps saying it will update but keep popping the same message, I even have all the models downloaded already... Is left anything to do to run the MMAudio ones?
@ad_Marzzel Mmh, seems you still have node conflicts, cause the error message says the sampler is not installed. You might have a look at the github pages for workarounds, use the comfyui trouble shooting guide, disable all not necessary nodes, etc.... This all needs patiance, knowlage and lot of time.
The easiest way for you might be a separate installation with Comfyui-Easy-Install. Tooks 2 mouse-clicks and around 30 minutes of time (see my short guide here at the model description). After installation open the worflow and install only the necessary custom nodes for this workflow first.
The "first frame to video" part is missing the Lora Loader for the lightx2v lora, without that, you will only get hazy outputs.
First of all: Thank you so much for this great workflow (I'm using v2.2). I'm using ComfyUI Windows portable nvidia 7z because I can't get the manager to work with the desktop version.
Everything is working fine, but I'm having an issue with fflf. I'm getting this message:
"JWImageResizeByLongerSide
_.execute() missing 1 required positional argument: ‘image’"
I should actually have all the nodes. A friend who introduced me to ComfyUI and recommended this workflow unfortunately couldn’t help me either and suggested I check the comments section here.
I would be very grateful for any help in resolving this issue. One more thing: I’m a complete beginner with ComfyUI.
@neverwinter437 Hi - thank you. It`s a known bug with the JW nodes. See my short FAQ at the end of the model description here for a workaround. Btw. As a comfyui beginner I would recommend you a separete comfyui installation just for video generation with Comfyui-Easy-Install (see my description here too).
Hey, Arkinson - your FAQ links for the JW nodes just point back to your page. I am trying to help Neverwinter out - we are buddies. I had to install the JW nodes using git and they work fine for me. So, if you could point to the FAQ regarding those nodes it would be helpful.
@arkinson Thank you very much for the quick response. HDean is the friend who suggested I post a comment here. He’s also been a big help with ComfyUi.
I tried using an older version of the workflow—version 2.0—and this problem doesn’t occur there.
@hdean @neverwinter437 Thank you for the hint. Civitai drives me crazy 😣 The link adresses in the FAQ are right (pointing to comments where we discussed this issue). But Civitai opens the model page instead of the comments. I see this behavior for about 1 or 2 weeks now. I even can not open comments I actually get with a right click in a new browser tab anymore. I tried different browsers - no luck. Opening comments is often try-and-error actually. Sorry - I know, that did not help you 🙄 The problem is, I can not find these discussions for myself in the severel hundreds of comments....
So let`s try from the beginning. The reson of your issue is, that the JW nodes not work properly on your system. The workaround is to install soundfile manually with the given command. You will find some more help on the github page of the JW nodes here: GitHub - jamesWalker55/comfyui-various · GitHub.
But once agein, I would really recommend you to use Comfui-Easy-Install. Every time I get too much trouble with my existing installation, I just install a fresh version, copy my settings and I am on the road again in less then 40 minutes. My last install is around 3 weeks ago - just two mouse clicks + installing missing notes and every thing runs including Manager and without any hussle about node conflicts, soundfile patches, etc. Uhhh, long talking, but maybe it will help you - or at least some others 🙂
@arkinson Thank you very much for your detailed response; it explains why I can't access the posts.
I'll try the suggestions you provided here. As I mentioned, I've now switched to Workflow v2.0. fflf works without any issues there.
@arkinson Hey, thanks for the quick response. I was telling Neverwinter that you are pretty on point with getting back to people. I really need to share some of what I have done using your workflow. I think I might have mentioned that I made a few adjustments to account for my 5090. I think I kind of push the boundaries by generating at 1024 X 1024 most of the time. So, my generation times are far slower than what they might otherwise be. But it's worth it for the sharper videos.
I keep promising myself I will check the newer workflows, but I find myself not wanting to make the adjustments or changing what is already working quite well. And you did tell me about the sound generating, but I tend to use VibeVoice and then throw it into a lipsync so I can have longer videos with consistent voicing. And, after much experimentation, VibeVoice works amazingly well with inflection.
My turn to get a bit long wided, eh? Thanks again.
@hdean @neverwinter437 Thank you both again.
Definitely try Comfyui-Easy-Install - I know, I sound like a salesman 😅 but I tested nearly all comfyui versions over the time and there is no one wich is so easy to handle on a Windows machine - including the SageAttention option (wich is the second mouse click) and speeds up the generation a lot.
And you really should test (my 😉) LTX workflows. It´s a completely new experiance in comparison to Wan. I use these "old" Wan workflows sometimes only for some special Loras to generate a start video, wich I can "tune up" via Video to Video in LTX.
@hdean With your gpu RTX 5090 you will really profit from LTX-2.3, cause it is made for native resolutions of 2k and 4k. Any higher generation resolutions deliver significantly better outpus here - all without frame multiplying and final upscaling....
If you like to try it, just let me know, to give you some hints to edit my workflow for better hardware.
@arkinson I have tried ltx but so far it refuses to do anything on my system. I am planning on a new version of Comfyui and then the update. I haven't tried that installer and I am a little hesitant. Trouble is, I need a bigger HD and I have 2 2tb drives that are getting to the almost full point. I do have another external. But that calls for certain organization I am too lazy to get to at the moment.
@hdean Yeah, full HDD`s - we all know this wretched state of affairs 😂
I bought a 1 TB SSD for comfyui only recently - and what happened? It is full 🤣 No worries about LTX - I just published workflow version v4.0 🙂
@arkinson I tried it and I need a new version of Comfyui. Thinking about downloading it tonight. Probably I will have shit break, which is why I rarely update comfyui.
@hdean Yes, sometimes you are right: Never touch a working system!
I`m forced to update permanently, cause I get a lot of user questions/issues regarding my workflows and it makes no sense to discuss if my own sytem would be completely out of date....
@arkinson Not a surprise. But you are working on things I do not have to worry about.
Still haven't updated. Managed to get sick as hell and have not had the energy to start messing around. But, I do intend on getting the lates version and loading the necessities into it. I have to versions, so I do one and then see what crap I have to deal with. Thankfully, the most I usually have to do is transfer custom nodes and models. So, not a big deal, just tedious. And I am usually in the middle of some feature video for me DA site.
@hdean Hi – at first, I really hope you're doing well 🙏👌
Just some hints to organize comfyui according to my experiances:
1. Use a central model folder and just edit the extra_model_paths.yaml file. This way you can link all your models to every new system just by copying the extra_model_paths.yaml file to the new system.
2. From ...comfyui/user/default path copy your workflow list and comfy.settings.json to the new system.
3. Never copy existing custom nodes to a new system - just open the workflows you are really use and simply load all missing nodes via the manager. This reduces the risk of node conflicts significantly.
As mentioned, the whole setup of a new system only costs me a few mouse clicks and round about 40 minutes of downloading/installation time.
@arkinson Believe it or not, I am actually working on a new version of Comfyui right now. I basically do the same thing you're talking about, except I have s hit ton of models on different hard drives.
I am about to try to load one of your latest models and get the nodes needed. I can't make the older version work on my new Comfyui - the JW nodes are DOA. Kind of sucks. But it's life. Hopefully, I can make use of the new LTX.
I am going slow. Right now I am recovering from a rather nasty cold that kicked my ass good. Probably, I will get to testing the new models after I make certain my old necessary workflows are working. I still like the SDXL models tremendously, and I need to ensure those workflows are stable. Then I get to play!
@arkinson I just tried loading your workflow with the MMaudio - turns out there is no way (unless I am wrong) to load specific audio. And I definitely need that. I need to figure what I need to do simple video with your workflow with the non gguf models. I have no clue about ltx though.
By the way, I sent you a private message some time back. I figure thats a better way to chat than this.
I install all the models and loras but when i run it there was an error called "No module named 'sageattention'" happened in KJNodes. Does anyone have a guide to fix this? Thanks.
@IGotBanned67 Did you installed SageAttention on your system?? If not, turn the SageAttention option off.
Thanks for the workflow, but I am having some issues.
First, the video quality is horrendous. Faces and body shapes change instantly, and the video itself is so blurry that I can't look at it without getting nauseous.
Second, the workflow prefers to use RAM more than the GPU. Ram doesn't go below 90% (32GB), but the GPU fluctuates constantly. It's so bad that Comfyui sometimes crashes because of memory violation (probably a problem on my end). VRAM fluctuates too, but not that much; it tends to be at a high usage rate.
Third, the workflow keeps using RAM and VRAM long after creating the video. I don't know how to fix it other than restarting Comfyui.
Also, I don't know what to do with "01.2 Random First Frame Image". I use the same path as the picture I use for "01.1 Single First Frame Image" but I don't know if I'm doing it right.
@toygurt1 Uhh - you guys stay in these pretty "old "Wan workflows, instead using (my) new LTX-2.3 workflow 🙂 Anyway - some short hints that may help you with this "old school" stuff:
First of all: update all components and make sure you have no node conflicts.
Blurry outputs: Check twice you have downloaded AND selected the right models. Have a look in the comments too - we had this "issue" discussion very often here.
RAM usage: That`s normal with Comfui core. Make sure your swap file is in the range of my min "specifications".
Random first frame image: This is just a pretty usefull option for completely automated overnight runs to generate several videos with different start images. Instead of a single image you set an image path containing a bunch of images you want to process randomly/sequentially "overnight" by setting RUN = "instant" for endless generation execution. So. next morning you might have generated some pretty cool videos.
Short answer: For single generations use only "Single First Frame Image".
@arkinson Thank you for the answer! I didn't know it was an "old" workflow. I've heard WAN was the best for video creation, so when I searched it and saw "3060 12GB," I didn't think too much and tried it since the comments were recent.
I think I had the right models, so I don't know what went wrong. I had about 22GB swap file, maybe that.
Random first frame image is a great idea, wow. But I wonder, does it have a chance to pick up the same image over and over, or does it choose each picture once?
In the end, I decided to ditch this one and try your LTX2.3 generator; that seems to be working great, thank you.
@toygurt1 A quick comparison between Wan and LTX pros:
Wan: Lot`s of Loras available. I still use it sometimes for specific generations- just because of the Loras. For example: A short Wan clip with simple noises for the start->LTX V2V to extend the clip for some more action and brilliant speech.
And yes, a lot of users still use it.
LTX-2.3: Complex audio/speech/singing/lip-syncing out of the box and much better video generation at all, but not too many Loras available actually.
But keep in mind: My LTX workflows are made for advanced comfyui users. Even if the usage is very simple finally, you might need some comfyui knowlage to get it running und to understand the pretty complex logic in the background.
Blurry outputs: you really SELECTDED the right models?
Random first frame: If you select "random" it will do what it is called: It will randomly pick one image from your folder each run. If you set it to "incremental" it will take the next image each run.