I've been trying to learn the best ways to generate videos from images on my PC, and many of my searches led to Pinokio, most of which recommend the Wan2GP package. There were a couple other packages recommended as well, but I wanted to ask if anyone has used this for their generation.

I've dabbled with it, but most generations take so long that playing around for best results takes more time than I have. I was hoping someone might have info/advise on which package they use (Wan2gp or otherwise) and what models/settings they tend to use.

bobfox said:
I've been trying to learn the best ways to generate videos from images on my PC, and many of my searches led to Pinokio, most of which recommend the Wan2GP package. There were a couple other packages recommended as well, but I wanted to ask if anyone has used this for their generation.

I've dabbled with it, but most generations take so long that playing around for best results takes more time than I have. I was hoping someone might have info/advise on which package they use (Wan2gp or otherwise) and what models/settings they tend to use.

I haven't used Pinokio, but I have written a guide on local video generation with ComfyUI and WAN 2.2: https://rentry.co/bagelmonster-wan-wip. This contains all of my settings, models used, and methodology for video generation from a base image. There will be a bit of a learning curve, and if you decide to follow the guide and run into any issues please DM me so I can add instructions or clarification to it. Once you have ComfyUI running everything can be done from your browser.

Here are the DaSiWa WAN 2.2 Lightning models I've used: https://civitai.com/models/1981116 (I'd recommend the newer BoundBite model for best prompt adherence) and his ComfyUI workflow: https://civitai.com/models/1823089. You will need a pretty beefy GPU to run these locally at a reasonable speed, I would recommend nothing lower than 16GB VRAM. If you have fast RAM you may be able to get away with less VRAM, I just don't know if it'll be slow to the point of being unusable.

If you have a flaky connection speed like I do and have problems downloading from CivitAI, DM me and I'll try to help. CivitAI broke download resumption recently so if it keeps failing you may have to try a mirror on HuggingFace, which does support resumption.

If you'd like to stick with Pinokio, it does look like Pinokio has a ComfyUI utility available (https://beta.pinokio.co/apps/github-com-pinokiofactory-comfy) which might help you set it up, or you may be able to use these models with WAN2GP by just telling it to do 4 steps.

bagelmonster said:
I haven't used Pinokio, but I have written a guide on local video generation with ComfyUI and WAN 2.2: https://rentry.co/bagelmonster-wan-wip. This contains all of my settings, models used, and methodology for video generation from a base image. There will be a bit of a learning curve, and if you decide to follow the guide and run into any issues please DM me so I can add instructions or clarification to it. Once you have ComfyUI running everything can be done from your browser.

So I found out ComfyUI is available in Pinokio and I was able to set it up as your guide described, and it works :D I'm legit stoked! There were a couple hang-ups but I was able to troubleshoot the missing files. Thanks very much for the guide :)

By chance do you know if there's a limit to how many frames can be generated per prompt?

bobfox said:
So I found out ComfyUI is available in Pinokio and I was able to set it up as your guide described, and it works :D I'm legit stoked! There were a couple hang-ups but I was able to troubleshoot the missing files. Thanks very much for the guide :)

By chance do you know if there's a limit to how many frames can be generated per prompt?

I'm glad you were able to get it working! If you don't mind, would you be willing to say what GPU you have and your first/subsequent generation times? I feel like most people won't have a 3090 or workstation GPUs like I do and would like to add more data points to the guide.

I would recommend no less than 2 seconds and no more than 7 seconds of video per prompt in general. By default it's 16 frames per second with the lightning models I linked, so 32-112 frames total. You can go to 1 second for small character movements if needed, but I wouldn't recommend going larger than 7 seconds, and I would very much recommend keeping your frame counts to multiples of 16. Note that WAN tends to try and loop from the input frame at 5 seconds from my experience, so if you're trying to create something that loops I would do 90 frame segments to make the final segment looping easier.

Also remember that while you're trying to get your prompt down, you can generate at super low resolutions with a fixed seed to make iteration faster, then increase the resolution when you're happy with how a segment looks.

bagelmonster said:
I'm glad you were able to get it working! If you don't mind, would you be willing to say what GPU you have and your first/subsequent generation times? I feel like most people won't have a 3090 or workstation GPUs like I do and would like to add more data points to the guide.

I would recommend no less than 2 seconds and no more than 7 seconds of video per prompt in general. By default it's 16 frames per second with the lightning models I linked, so 32-112 frames total. You can go to 1 second for small character movements if needed, but I wouldn't recommend going larger than 7 seconds, and I would very much recommend keeping your frame counts to multiples of 16. Note that WAN tends to try and loop from the input frame at 5 seconds from my experience, so if you're trying to create something that loops I would do 90 frame segments to make the final segment looping easier.

Also remember that while you're trying to get your prompt down, you can generate at super low resolutions with a fixed seed to make iteration faster, then increase the resolution when you're happy with how a segment looks.

I have a GIGABYTE RTX4070S EAGLE OC 12G
only 32gb of ram atm.

I actually fiddled with some settings and rendered a 5 sec video at 24fps, ending dimensions of 992x800 (which is odd cause it was not what I had in the comfy settings), but it took 15 mimutes according to the job details if that helps. I also used a lora from civitai. I'd be happy to try a benchmark if you want to provide details, I can run it and reply.

That said, I'm just sort of doing this as a hobby with no intention to publish anything for now. Maybe someday I -might- try a project, but it's not likely.

bobfox said:
I have a GIGABYTE RTX4070S EAGLE OC 12G
only 32gb of ram atm.

I actually fiddled with some settings and rendered a 5 sec video at 24fps, ending dimensions of 992x800 (which is odd cause it was not what I had in the comfy settings), but it took 15 mimutes according to the job details if that helps. I also used a lora from civitai. I'd be happy to try a benchmark if you want to provide details, I can run it and reply.

That said, I'm just sort of doing this as a hobby with no intention to publish anything for now. Maybe someday I -might- try a project, but it's not likely.

Sure, I've DMed you with the requested benchmark parameters so as not to spam up this thread. For the ending dimensions not matching, did you enable any upscaling options in your workflow?

My account on this site is too new for me to send DM's. I got your message, but for right now, I can't: I am having an issue with ComfyUI where it says "Failed to fetch server logs" and won't run any jobs, and I have no idea how to fix it x.x

bobfox said:
My account on this site is too new for me to send DM's. I got your message, but for right now, I can't: I am having an issue with ComfyUI where it says "Failed to fetch server logs" and won't run any jobs, and I have no idea how to fix it x.x

Judging by your profile creation date and time, that limit for DMs will actually be lifted in like 10 minutes. The limits are imposed for a straight 7 days, and that's when the timing lines up for your profile.

bobfox said:
My account on this site is too new for me to send DM's. I got your message, but for right now, I can't: I am having an issue with ComfyUI where it says "Failed to fetch server logs" and won't run any jobs, and I have no idea how to fix it x.x

It sounds like the ComfyUI server portion isn't running, it may need you to manually run it or some other step in Pinokio to get it running. Your web browser just connects to the local ComfyUI server. You may be able to test this by trying to go to http://127.0.0.1:8188 (assuming it set up the default ComfyUI settings) - if you can get there it means ComfyUI is running but Pinokio can't reach it for some reason.