v1.0 = 24 epochs, v1.5 = 50 epochs
My magnum opus, a image to video wan lora for specifically animating birdman's style pixel art more consistently. (Can also serve as a pixel art stabilizer lora)
Use with outputs from, if you want best compatibility: https://civarchive.com/models/1428935/birdman46049238-i-pixel-art-style-i-illustriousxl-and-noobai
Trained for X Epochs on Wan 14B, at 480p resolution with around 40 3-5 second videos/gifs outsourced from birdman's twitter, with 5 repeats, LR of 0.00002.
Birdman's animations are usually at like 11.1, 12.5, and 14.29 fps, so if the output fps of the video is higher, then you will need to artificially lower the fps so it appears pixel artish like and not uncanny smooth.
My Generation Settings:
Example Prompt: (Trigger = p1x3lStyl3)
This p1x3lStyl3 animation features a character rendered in a pixelated style against a plain white background. The p1x3lStyl3 character initially stands facing forward, wearing her iconic blue shark hoodie with the hood down, showcasing her white and blue p1x3lStyl3 hair and bright blue eyes. Her hands are loosely clenched at her sides, and her blue p1x3lStyl3 shark tail rests behind her. The p1x3lStyl3 animation consists of a subtle loop where she slightly bends her knees, causing her whole p1x3lStyl3 body to dip down slightly. Simultaneously, her p1x3lStyl3 tail gives a small, gentle swish to the side. She then smoothly returns to her original standing pose, ready for the loop to repeat, creating a simple, idle p1x3lStyl3 animation.Lora Strength: 1
Flow Shift: 5
Guidance Scale: 6
Steps: 20-30
Video Length: 3-5 (Preferably 5)
Post some videos for some buzz!
Description
FAQ
Comments (18)
Please a Hunyuan for framepack T.Tpleasse
Are you willing to tip me enough buzz so i can do so?
@AndroidXL lessgoo
@hboxgames132 need like 4k to train hunyun, so need 3k more
@AndroidXL sure !
I got very good result with 30 steps cfg 7 but when I try to use the famous new lora https://civitai.com/models/1585622?modelVersionId=2014449 to accelerate (step 4, cfg1, lcm) then everything become very light like too much sun. Is it possible to retrain this lora to make it conpatible with this acceleration lora ? how much cost a trainning lora ? Can I do it on a 4090 24go-vram ?
Have you tested with Wan 2.2? I'm asking because it's working for me (doesn't give an error at least), I'm using WanGP, Wan 2.2 I2V 480p model
Has anybody managed to get this working in a workflow yet? Really don't want to have to use an online generator.
im starting to think this is the one used on those "pixel" game ads in Facebook i keep seeing in my feed....
anyways, hope we see more of this type of style wan lora in future
If anyone would like to provide the GPUs to train a wan 2.2 version, let me know. The company I trained the lora with isn't available to do so anymore.
How much hardware would you need boss?
I can provide my video card for training
@touchmypetertracy672 @Aizark atleast a 4090 with 24gigs vram
@AndroidXL Can lend you a 5090 in ~2 weeks if Aizark/someone else doesn't help you out first
Does anyone have a ComfyUI workflow with this? Been trying to do it but the results are not good :(
you can download this(https://civitai.com/images/102895727) and throw .mp4 into your comfyui
@hankb620624 i tried that. it dosen't have any workflow for me.
@moppelkaiser https://files.catbox.moe/1zwr5s.rar
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.