I deleted the exact prompt but i think it has something to do with psychedelic zoo, fantasy forest, a lot of animal names and probably i used the loras PsychedelicFluids, Surrealismai and Psyai along with revAnimated
@@mushroots1679Could I pay you to tutor me? I've been using Deforum but having difficulty getting the results. You are right around the results I am seeking.
Sorry but it needs a lot of time and trying out to get these results and at the moment i'm more interested in generating more animations than tutorials but here are a few tips and rules i always follow: - higher resolutions = more consistency as flickering pixels kinda get "smaller" and almost invisible in the overall picture (this video was rendered at 1280x1440) - only render half of the screen. The models are trained on square pictures. A picture size that is almost a square gives you better results than wide screen - for wide screen, get the right amount of Y axis rotation according to your Z translation. The picture shouldn't move in or out at one side. Mirror the video on this side so the result looks like it's zooming into the middle. - strength schedule may not be what you think it is. It isn't directly how much your picture is gonna change, it's more how big in size those changes will be (or to wich step in the render it gets set back). Strength schedule at 0.98 for example will be changes almost at individual pixels level. 0.4 will change large clusters of pixels (this may not be accurate but it's how it appears to me). There are ways to increase how much your picture is changing without touching the strength schedule. For example: - use a lot of loras. I noticed by using 3-4 Loras, the picture can change a lot even at very high strength schedules. - using CN's in loopback mode. Some can increase the consistency a lot like the tile model but without any preprocessor. And others can increase the changes like clip_vision. (i used clip_vision for this one) - if you have troubles with vram or render time by using high resolutions and CN's, download the t2i adapters from tencent. They aren't as good as the original CN models but they use less vram and are faster. - i always use a depth CN. The 3D effect just looks better if you have one. - put everything into one big prompt instead of changing smaller prompts. Not everything you put into the prompt will appear in the first frame but it will appear when it fits somewhere and it can result in very interesting changes. - i highly recommend using rev-animated and clip skip 2
I've just watched all your videos. It's just amazing. Pliz never stop ;)
This one is my fav!!!!! What was the prompt? Any hints? Love the style!
I deleted the exact prompt but i think it has something to do with psychedelic zoo, fantasy forest, a lot of animal names and probably i used the loras PsychedelicFluids, Surrealismai and Psyai along with revAnimated
Sounds like a rabbit hole, I have not hear of pscyedelicfluids, surrealismai and psyai@@mushroots1679 🤔
@@mushroots1679Could I pay you to tutor me? I've been using Deforum but having difficulty getting the results. You are right around the results I am seeking.
Can we get a tutorial mate
Sorry but it needs a lot of time and trying out to get these results and at the moment i'm more interested in generating more animations than tutorials but here are a few tips and rules i always follow:
- higher resolutions = more consistency as flickering pixels kinda get "smaller" and almost invisible in the overall picture (this video was rendered at 1280x1440)
- only render half of the screen. The models are trained on square pictures. A picture size that is almost a square gives you better results than wide screen
- for wide screen, get the right amount of Y axis rotation according to your Z translation. The picture shouldn't move in or out at one side. Mirror the video on this side so the result looks like it's zooming into the middle.
- strength schedule may not be what you think it is. It isn't directly how much your picture is gonna change, it's more how big in size those changes will be (or to wich step in the render it gets set back). Strength schedule at 0.98 for example will be changes almost at individual pixels level. 0.4 will change large clusters of pixels (this may not be accurate but it's how it appears to me). There are ways to increase how much your picture is changing without touching the strength schedule. For example:
- use a lot of loras. I noticed by using 3-4 Loras, the picture can change a lot even at very high strength schedules.
- using CN's in loopback mode. Some can increase the consistency a lot like the tile model but without any preprocessor. And others can increase the changes like clip_vision. (i used clip_vision for this one)
- if you have troubles with vram or render time by using high resolutions and CN's, download the t2i adapters from tencent. They aren't as good as the original CN models but they use less vram and are faster.
- i always use a depth CN. The 3D effect just looks better if you have one.
- put everything into one big prompt instead of changing smaller prompts. Not everything you put into the prompt will appear in the first frame but it will appear when it fits somewhere and it can result in very interesting changes.
- i highly recommend using rev-animated and clip skip 2