Stable diffusion thread

[zheng]

trying temporal-kit + ebsynth.... knn so many steps

 

[coyote]

Edmwers' fantasy of living in matrix coming to life. AI just keep feeding edmwers chiobus' photo and edmwers can lie in a cryogenic chamber making electricity to power the AI for their lifetime.

 

[zheng]

Edmwers' fantasy of living in matrix coming to life. AI just keep feeding edmwers chiobus' photo and edmwers can lie in a cryogenic chamber making electricity to power the AI for their lifetime.

yes de yes de if u said so

 

[focus1974]

Edmwers' fantasy of living in matrix coming to life. AI just keep feeding edmwers chiobus' photo and edmwers can lie in a cryogenic chamber making electricity to power the AI for their lifetime.

not so fast.
First Universal Basic Wage have to be institutionalized.
then we lie flat.

 

[coyote]

not so fast.
First Universal Basic Wage have to be institutionalized.
then we lie flat.

Already became batteries, fantasy fulfilled, need income for what?

 

[focus1974]

Already became batteries, fantasy fulfilled, need income for what?

how to make you lie flat ?
first entice you to work in the matrix 8 to 5, and get paid UBI to use in real life.
then when majority is on it, disabled the permission to knock off from work and you will become batteries forever

 

[zheng]

Already became batteries, fantasy fulfilled, need income for what?

when chiu lie flat chiu is become that battery liao... is alway recharged and it never run flat

 

[doogyhatts]

 

[doogyhatts]

trying temporal-kit + ebsynth.... knn so many steps

Mainly used for converting realistic video to cartoon.
Last year's technique actually before AnimateDiff came out.

The author for Forge removed the temporal-net loopback functionality and so temporal-net is no longer usable.
So I moved onto AnimateDiff instead.

 

[zheng]

Mainly used for converting realistic video to cartoon.
Last year's technique actually before AnimateDiff came out.

The author for Forge removed the temporal-net loopback functionality and so temporal-net is no longer usable.
So I moved onto AnimateDiff instead.

installed temporalkit extension already its tab is missing

 

[doogyhatts]

@zheng
Can try to use SD3 in the discord, but it uses one of the earlier models and costs credits.
https://stability.ai/news/stable-artisan

 

[doogyhatts]

I could see why using the Hyper version of JuggernautXL is a poor choice for 8gb vram GPU.
It's upscaling output quality is rubbish.
No point trying to switch models as it will be a huge waste of time.
Might as well just use the normal version.

 

[zheng]

I could see why using the Hyper version of JuggernautXL is a poor choice for 8gb vram GPU.
It's upscaling output quality is rubbish.
No point trying to switch models as it will be a huge waste of time.
Might as well just use the normal version.

need to set low cfg scale

 

[doogyhatts]

need to set low cfg scale

Already using cfg 2.
Upscaled results still poor.
I will swap out the hyper for the normal one.

 

[doogyhatts]

After watching the tutorial on Panavision style videos.
Let's try Fooocus and compare it with Leonardo.ai.

Prompt:
Audrey Hepburn holding an iPhone, 1950's aesthetic, made in the 1950's, super Panavision 70 color film stock, beautiful cinematography, low contrast, balanced colors, 3 point studio lighting, high resolution, rich color palette, smooth tonality

Juggernaut-XL using Fooocus
Added SAI cinematic & analog-film styles.
Using dpm++ 2m karras, 35 steps, cfg 7.
Non-upscaled.

Leonardo.ai
Using LeonardoLightningXL model and cinematic style.

 

[zheng]

Leonardo.ai
Using LeonardoLightningXL model and cinematic style.

this one looks better

 

[zheng]

actually this version is more accurate to the Panavision 70 color film style

the bottom one is like a modern day digital cam with HDR tech

 

[doogyhatts]

I forgot to use the specific LoRA for Audrey Hepburn and DreamShaperXL model.
So here it is again:

Prompt:
close up photo of adr, holding iphone, beautiful and young Audrey Hepburn holding iPhone, 1950's aesthetic, made in the 1950's, super Panavision 70 color film stock, beautiful cinematography, low contrast, balanced colors, 3 point studio lighting, high resolution, rich color palette, smooth tonality

DreamShaper-XL1_alpha2 model
Added only SAI analog-film style.
Sampler dpm++ 2s ancestral, normal scheduler, 36 steps, cfg 8.
LoRAs: 1950s technicolor style XL, audrey hepburn XL, iphone mirror selfie XL

Spoiler

Comparing with JuggernautXL model using similar settings.
Sampler dpm++ 2m karras, 35 steps, cfg 7.

Spoiler

 

[doogyhatts]

Spend some time reducing unwanted prompts.
Removed the SAI analog film style.
Retained the 1950s technicolor LoRA as it has the most impact.

Besides scene description in the prompt, the remaining part is just the LoRA keywords and style.
1950's Technicolor style, beautiful cinematography

Analog film style is just a bunch of prompts.
Style: Analog film,"analog film photo {prompt} . faded film, desaturated, 35mm photo, grainy, vignette, vintage, Kodachrome, Lomography, stained, highly detailed, found footage"

Without 1950s technicolor LoRA. No additional styles other than defaults.

With 1950s technicolor LoRA. No additional styles other than defaults.

Without 1950s technicolor LoRA but enabled SAI analog film style.

 

[doogyhatts]

Finally an easy to understand tutorial on SUPIR upscaler.