[ARCHIVED THREAD] - FramePack - Generate AI videos from images using your computer's GPU (Now ComfyUI Megathread!) (Page 4 of 17)
|
Quoted: I didn't pay that much attention but I think it was around a couple hours. Thing is, I was watching youtube videos most of the time it was working. The process ran in the background and didn't seem to make my machine sluggish. Here's the clip converted to gif. The quality before conversion was MUCH better. I supplied the pic and gave the instruction 'man walking towards camera swinging helmet'. https://media0.giphy.com/media/v1.Y2lkPTc5MGI3NjExZm1ucTFpb2sxZTFlODZ4eHpjemxlaTlyM29zYTdsOXQ0b3ptMXcxdiZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/4b9e07b45BobIZPoSX/giphy.gif Quoted: Quoted: Quoted: Thanks to this thread I installed framepack and made my first 5 second vid. It was all pretty easy but it took a while because I am using an i7-12700h with 64GB RAM and RTX 3060 with 6GB VRAM. I was surprised at the quality of the output. @CapedCrusader Approximately how long was “a while”? I’ve got a similar setup. I didn't pay that much attention but I think it was around a couple hours. Thing is, I was watching youtube videos most of the time it was working. The process ran in the background and didn't seem to make my machine sluggish. Here's the clip converted to gif. The quality before conversion was MUCH better. I supplied the pic and gave the instruction 'man walking towards camera swinging helmet'. https://media0.giphy.com/media/v1.Y2lkPTc5MGI3NjExZm1ucTFpb2sxZTFlODZ4eHpjemxlaTlyM29zYTdsOXQ0b3ptMXcxdiZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/4b9e07b45BobIZPoSX/giphy.gif Thanks. At least I have a ballpark idea of how long this takes with similar specs. |
|
I deleted everything but Python and 7zip again this morning, and then downloaded Framepack again. I went through all the stuff, update/run, and my result was the same. After less than a minute, the boxes on the right disappeared, the *Start Generation* or whatever was clickable again, and my jpg was saved as a png in the folder. |
|
Quoted: I deleted everything but Python and 7zip again this morning, and then downloaded Framepack again. I went through all the stuff, update/run, and my result was the same. After less than a minute, the boxes on the right disappeared, the *Start Generation* or whatever was clickable again, and my jpg was saved as a png in the folder. You should view this as a wonderful excuse to buy a new gaming laptop. And no, I don't really play games but I like having a fast, snappy computer to use (and sometimes do demanding tasks like this). |
|
Quoted: downloaded it. Now extracting the file and Windows says I have 1 day left to extract. Not sure your Gateway 2000 will run it anyway.
|
|
Quoted: Thanks. At least I have a ballpark idea of how long this takes with similar specs. Quoted: Quoted: Quoted: Quoted: Thanks to this thread I installed framepack and made my first 5 second vid. It was all pretty easy but it took a while because I am using an i7-12700h with 64GB RAM and RTX 3060 with 6GB VRAM. I was surprised at the quality of the output. @CapedCrusader Approximately how long was “a while”? I’ve got a similar setup. I didn't pay that much attention but I think it was around a couple hours. Thing is, I was watching youtube videos most of the time it was working. The process ran in the background and didn't seem to make my machine sluggish. Here's the clip converted to gif. The quality before conversion was MUCH better. I supplied the pic and gave the instruction 'man walking towards camera swinging helmet'. https://media0.giphy.com/media/v1.Y2lkPTc5MGI3NjExZm1ucTFpb2sxZTFlODZ4eHpjemxlaTlyM29zYTdsOXQ0b3ptMXcxdiZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/4b9e07b45BobIZPoSX/giphy.gif Thanks. At least I have a ballpark idea of how long this takes with similar specs. @bluemax_1 I’m currently running the exact same setup processor/ram/card, although 8gb vram on the 3060 (did they make a 6gb version?) You will be able to see the first (1 second) vid portion in around 5 min. I never finished downloading a five second clip, but carrying the same pace, 25-30 min. If you want to dive into the deep end, I can let you know what to expect from that setup on Wan 2.1 as well. |
|
Quoted: @bluemax_1 I’m currently running the exact same setup processor/ram/card, although 8gb vram on the 3060 (did they make a 6gb version?) You will be able to see the first (1 second) vid portion in around 5 min. I never finished downloading a five second clip, but carrying the same pace, 25-30 min. If you want to dive into the deep end, I can let you know what to expect from that setup on Wan 2.1 as well. I'm using an MSI laptop with a 6GB RTX3060. So it's not really a 'card'. |
|
The nice thing about a 5090 is generating a 5 second video in under 5 minutes. The 1 second preview is available in about 1 minute. I know FramePack is intended for low vram, but rendering speed scales nicely with 32GB and I haven’t seen any “more advanced” ComfyUI video generators with better output quality. |
|
Quoted: I'm trying a free one online, and it looks like it might work. It's been going for about 30min, and is still active. I'll update, if it works. https://www.pixverseai.org/en/framepack-ai Be careful. I wouldn’t upload any personal photos there.
|
|
Quoted: Be careful. I wouldn't upload any personal photos there. https://www.ar15.com/media/mediaFiles/563643/IMG_5654-3531774.jpg Yikes. It's still going, so I guess I'll just close it out. Thanks. |
|
Can you share what written instructions you are giving to create these? The last 3 or 4 I made suck. It's hard finding a word explanation that leads to your intended result. |
|
Quoted: That sucks. You can put the laptop in the refrigerator while it runs to help with thermals. Quoted: Quoted: Finally got it going on a 2070 Super. To say it's slow is an understatement... Trying to do a 5 second video from a picture of my mother from the 70s. EDIT: Well, After about 18 minutes the laptop, I believe, went to thermal shutdown. I'll have to see about another way to do this tomorrow. Nope... was a power supply issue. That sucks. You can put the laptop in the refrigerator while it runs to help with thermals. Lol, raise your hand if you've done this before. I actually considered buying a cheap mini fridge and cutting a hole in the face of it and putting plexiglass over it. I remember back in the days of spinning hard drives I was recovering one for a friend and I could pull it out of the freezer and have it work for just a little bit but it wasn't long enough to pull everything off. Ended up sandwiching the hard drive between two bags of Frozen peas that kept it cold long enough for the Armature to articulate and read the disc. I told him he owed me two bags of peas. Nothing more, nothing less. Oh those were the days. |
|
Oh wow! That picture used to be my desktop background 20 years ago. Turned out great |
|
Quoted: Can you share what written instructions you are giving to create these? The last 3 or 4 I made suck. It's hard finding a word explanation that leads to your intended result. I didn't save the prompt for this, but from memory, it was a very generic one- "4k uhd video of a stunningly attractive woman with perfect boobies, sitting on a bomb, wearing military uniform, bombs and vehicles in the background" The negative prompt- "????,??,??,??????,??,??,??,??,??,??,????,????,???,JPEG????,???,???,?????,???????,???????,???,???,???????,????,???????,?????,???,?????,???" Keep in mind that this is using the image-to-video process. A picture says a thousand words.... Also, know that my workflow is heavily modded, uses multiple different models for the same pic/video, and that there are way too many variables to express in this reply. The user interface itself can be quite important, as that greatly affects your level of aggravation during this whole process. I use Comfy UI, and the very best model currently for generating video is Wan2.1 Fourteen-billion parameter. Since this requires something like 60 GB of vram, I use a quantized version of it that requires *much* less. Kijai's quantized various Wan2.1 models Installation of ComfyUI ![]() How to install and use ComfyUI - Stable Diffusion. How to use ComfyUI (which lets you use *many* different models, FramePack, etc). This is an updated version of the video above which fleshes out how to operate it a bit more in-depth while still being understandable to the beginner. ![]() How to use ComfyUI for beginners. None of these are a simple one-click install that you are probably used to using since you bought your first computer. This software is still considered "experimental", and is updated constantly. There are many pre-requisites that have to be installed. You may need to manually configure your PATH settings. You may find out that the GPU that came with your pre-built computer doesn't have CUDA.... Best to determine that your equipment will meet the minimum specs *before* you invest many hours of aggravation. Be aware that merely installing the newest python/torch/etc can break your installation. If you succeed in getting it working, think twice before you mash the UPDATE button! It is quite likely to break things. In general, AI software is made with *specific* versions of python/torch/etc. |
|
Quoted: I open run.bat, and after a bit it opens up in Firefox tab. https://www.ar15.com/media/mediaFiles/51036/FP__1_-3528100.jpg I select a pic, type in some prompts, and click Generate. https://www.ar15.com/media/mediaFiles/51036/FP__2_-3528101.jpg It goes for a minute or two, with the boxes on the right highlighted with orange edges, then that goes away and the Generate button is ready again. I go to check the folder, and it has saved the jpg as a png. Quoted: I open run.bat, and after a bit it opens up in Firefox tab. https://www.ar15.com/media/mediaFiles/51036/FP__1_-3528100.jpg I select a pic, type in some prompts, and click Generate. https://www.ar15.com/media/mediaFiles/51036/FP__2_-3528101.jpg It goes for a minute or two, with the boxes on the right highlighted with orange edges, then that goes away and the Generate button is ready again. I go to check the folder, and it has saved the jpg as a png. Are you using Windows? How did you get it to run? Following the website instructions to first run update.bat, I get this error message even if I run as administrator. 'environment.bat' is not recognized as an internal or external command, operable program or batch file. 'git' is not recognized as an internal or external command, operable program or batch file. 'git' is not recognized as an internal or external command, operable program or batch file. Press any key to continue . . . I then tried to run run.bat and also got an error message: 'environment.bat' is not recognized as an internal or external command, operable program or batch file. The system cannot find the path specified. Press any key to continue . . . I opened the .bat files and did not recognize the git command one of them uses. Also do not understand why it says that envoronment.bat isn't a recognized command. Would someone please help? |
|
Quoted: Are you using Windows? How did you get it to run? Following the website instructions to first run update.bat, I get this error message even if I run as administrator. I then tried to run run.bat and also got an error message: I opened the .bat files and did not recognize the git command one of them uses. Also do not understand why it says that envoronment.bat isn't a recognized command. Would someone please help? Did you extract the folder you downloaded before executing “update.bat” and “run.bat”? Do you have an RTX 3000 series or newer? |
|
Quoted: Did you extract the folder you downloaded before executing “update.bat” and “run.bat”? Do you have an RTX 3000 series or newer? Quoted: Quoted: Are you using Windows? How did you get it to run? Following the website instructions to first run update.bat, I get this error message even if I run as administrator. I then tried to run run.bat and also got an error message: I opened the .bat files and did not recognize the git command one of them uses. Also do not understand why it says that envoronment.bat isn't a recognized command. Would someone please help? Did you extract the folder you downloaded before executing “update.bat” and “run.bat”? Do you have an RTX 3000 series or newer? Yep. Extracted the whole thing to the C: drive, which is a NVMe, so very quick. It created a folder framepack_cu126_torch26 with two subfolders: system and webui and three .bat files. CPU is a Ryzen 7, 16 GB RAM and GPU a NVidia 4060 with 8GB VRAM. |
|
Quoted: Do you also have Python installed? It won't work without that. I double-click the run.bat, a black box pops up, and then I just watch that until it finishes loading and opens a new tab in Firefox. Installed both. Thanks! That solved the Git issue. Now the problem is getting Windows' CMD window to recognize the environment.bat command. It keeps insisting it's not recognized. Weird thing is that if I run environment.bat by itself, no complaints. It's something about the other .bat files calling it. I did a few searches and some forums mentioned how CMD can be finicky with bat files encoded using UTF-8 instead of ASCII. I saved all .bat files as ANSI, since Notepad only has that option. Still nothing. |
|
Quoted: Installed both. Thanks! That solved the Git issue. Now the problem is getting Windows' CMD window to recognize the environment.bat command. It keeps insisting it's not recognized. Weird thing is that if I run environment.bat by itself, no complaints. It's something about the other .bat files calling it. I did a few searches and some forums mentioned how CMD can be finicky with bat files encoded using UTF-8 instead of ASCII. I saved all .bat files as ANSI, since Notepad only has that option. Still nothing. Quoted: Quoted: Do you also have Python installed? It won't work without that. I double-click the run.bat, a black box pops up, and then I just watch that until it finishes loading and opens a new tab in Firefox. Installed both. Thanks! That solved the Git issue. Now the problem is getting Windows' CMD window to recognize the environment.bat command. It keeps insisting it's not recognized. Weird thing is that if I run environment.bat by itself, no complaints. It's something about the other .bat files calling it. I did a few searches and some forums mentioned how CMD can be finicky with bat files encoded using UTF-8 instead of ASCII. I saved all .bat files as ANSI, since Notepad only has that option. Still nothing. Still banging my head here. I ran environment.bat manually. No error messages this time. Then I ran run.bat and got several errors such as directory not found, etc. Python.exe not found, etc. I then edited the run.bat file to point to the python.exe file directly. Now I am getting this error. The system cannot find the path specified. "\python\python.exe" demo_gradio.py --server 127.0.0.1 --inbrowser C:\Users\rossi\AppData\Local\Microsoft\WindowsApps\python.exe: can't open file 'C:\\Windows\\system32\\demo_gradio.py': [Errno 2] No such file or directory Press any key to continue . . . It appears that there's some environment configuration issues. This is the command line and the variables but it seems to not recognize them. C:\Users\rossi\AppData\Local\Microsoft\WindowsApps\python.exe demo_gradio.py --server 127.0.0.1 --inbrowser I installed Python through the MS store. Maybe that's the problem? Should have downloaded and installed from Python.org website? |
|
Framepack F1 was recently released: https://github.com/lllyasviel/FramePack/discussions/459 |
|
installed framepack, git, python. opens in browser, i can select image and enter prompt, click start generation...then after some time it just resets and acts like I never hit start generation says text encoding, vae encoding, clip vision, start sampling, then resets. Geforce GTX 1660 Super |
|
last items are torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 28.87 GiB. GPU 0 has a total capacity of 6.00 GiB of which 2.09 GiB is free. Of the allocated memory 2.58 GiB is allocated by PyTorch, and 325.92 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables) Unloaded DynamicSwap_LlamaModel as complete. Unloaded CLIPTextModel as complete. Unloaded SiglipVisionModel as complete. Unloaded AutoencoderKLHunyuanVideo as complete. Unloaded DynamicSwap_HunyuanVideoTransformer3DModelPacked as complete. |
|
Identical to what I kept getting. What GPU do you have? I messed around with this a lot trying to make it work with a 2080. Kept getting out of memory errors even though the card is 8gb. It “started” working a few time in that I got the preview, but it could never complete. Found a 3060ti 12gb for cheap so I can play around with this. |
|
Quoted: does that mean slide the slider up? or buy a new card it says"Tried to allocate 28.87 GiB.", seems like a lot Quoted: Quoted: I’m surprised it runs at all on that card. The “GPU Inference Preserved Memory” says to increase if you encounter out of memory. does that mean slide the slider up? or buy a new card it says"Tried to allocate 28.87 GiB.", seems like a lot I’d try moving the slider. It works on 6 GB cards, but probably doesn’t recognize the 1660. |
|
Quoted: does that mean slide the slider up? or buy a new card it says"Tried to allocate 28.87 GiB.", seems like a lot Forgive me, because I don’t know the proper terminology so it may sound odd to the tech guys…. The program is designed for the 30/40 memory architecture. If it doesn’t see that, it falls back on older protocols that more than double the required memory. I don’t think it’ll Possible to make it work with your card. |
|
Quoted: Forgive me, because I don’t know the proper terminology so it may sound odd to the tech guys…. The program is designed for the 30/40 memory architecture. If it doesn’t see that, it falls back on older protocols that more than double the required memory. I don’t think it’ll Possible to make it work with your card. appreciate it! is there another similar program that will work on my computer? ive been paying for sea art ai and want to run local |
|
I got it working with my 2080, but it's slow as hell. I can't image what it would take on your 1660, nor do I know if it would work. https://github.com/freely-boss/FramePack-nv20 I just drag and dropped the files into the webui subfolder and let it replace what is needed. |
|
It's well-worth the effort to figure this stuff out. I managed to improve my generation speed about 30% last night by configuring my build to use Nvidia's Triton library. Attached File Spelling and punctuation isn't one of its strong points... |
|
Quoted: It's well-worth the effort to figure this stuff out. I managed to improve my generation speed about 30% last night by configuring my build to use Nvidia's Triton library. https://www.ar15.com/media/mediaFiles/3906/ComfyUI_00326__png-3534309.JPG Spelling and punctuation isn't one of its strong points... How did you do that? |
|
Trying to nail down another issue, but maybe it's the 20 series card and the workarounds I've had to use. You guys getting the "realistic" results, like skin texture, lines, ect, are you using prompts at all to generate that or is it just happening naturally? Nearly every one of my results is taking photos and giving them that smooth AI enhanced face. Arms, legs, shoulders and other exposed skin seem to stay relatively untouched. |




