[NovelChef] Comics

[viscountsheep]

I busted my ass off for a whole day trying to create comics, turns out its way harder than it seems

it can't be that hard.. are you using ai for panels ?

 

[unknown254]

Yeah, i tried it a while back and just did one image at a time. It turns out you have to add postive/negative prompts and add special values to each. The hands were by far the toughest and if you fix one part it fucks up the another part randomly. Also creating a femboy is damn near impossible since it defaults to a male or female based on the features you specify. If I remember correctly it should be free for a certain number of prompts, maybe you should try and post some if your feeling ambitious like I did.

 

[viscountsheep]

Yeah, i tried it a while back and just did one image at a time. It turns out you have to add postive/negative prompts and add special values to each. The hands were by far the toughest and if you fix one part it fucks up the another part randomly. Also creating a femboy is damn near impossible since it defaults to a male or female based on the features you specify. If I remember correctly it should be free for a certain number of prompts, maybe you should try and post some if your feeling ambitious like I did.

maybe you should try the newer image generation models like illustrious or anima, it's easier i think. im guessing novelai uses some kind of sdxl model which us why the tags are necessary.

about trying it.. i mean yeah.. why not... it shouldn't be that hard to tell the ai to make no mistakes lol, i'll try my hand at it

 

[mihaluck]

I tried anima much, but how it works character looks different for each generation.

 

[unknown254]

Prompt engineering is a stressful task but hopefully you make some progress. Ik I can't lol. Will look into other ai tools though.

 

[danny-warrock]

Honestly the best way currently to make comics with recurring characters is with Flux 2 Klein 9b. with it you can combine images as a reference using text to dedicate Girl 1 and Girl 2. its...just slow. Currently testing out Krea2 and even though its references aren't so well done yet, using specific details and the same seed is making similar looking people. A Krea2 edit would work better with references but isnt possible right now (unless you get an extra conditioning node for comfyui). A benefit of Flux2 is its editing capabilities, so you can reuse the generated image and edit specifics on it (make her hair shorter, make her more boyish, make him wear a dress, etc, etc.).

 

[jhn]

Still no updates from moonly's protege

Last update of

CC -Apr 5

OP - Feb 27.



Bruh

 

[Sur_GalaChad]

Still no updates from moonly's protege

Last update of

CC -Apr 5

OP - Feb 27.



Bruh

Yikes haha and there are still subscribers?

 

[unknown254]

Bro secured the bag and dipped

 

[viscountsheep]

i ended up testing some stuff. best bet for consistency in anime illustration style is illustrious still.
amina offers way more freedom in description but it's very difficult to recreate an original character in a particular styling which i think is not a bad tradeoff.
krea2 looks seriously promising but the model is large and it's crashing my inference constantly.

good news, i think i have an idea on how to make things look coherent with some post processing and corrections in krita if need be.
bad news, my system fans are screaming all the time.
the window i can run these models are a bit short because im running cloud inferences since my gtx 1650 aint gonna cut it.

but YOU CAN get it done. id say it's moderately difficult still with ai generated images, it does save time in the sense that you already get an almost finished panel. but prompting is a bit tricky still for original characters and original style.

will try training some lora for custom characters when i have time later.

 

[mihaluck]

i ended up testing some stuff. best bet for consistency in anime illustration style is illustrious still.
amina offers way more freedom in description but it's very difficult to recreate an original character in a particular styling which i think is not a bad tradeoff.
krea2 looks seriously promising but the model is large and it's crashing my inference constantly.

good news, i think i have an idea on how to make things look coherent with some post processing and corrections in krita if need be.
bad news, my system fans are screaming all the time.
the window i can run these models are a bit short because im running cloud inferences since my gtx 1650 aint gonna cut it.

but YOU CAN get it done. id say it's moderately difficult still with ai generated images, it does save time in the sense that you already get an almost finished panel. but prompting is a bit tricky still for original characters and original style.

will try training some lora for custom characters when i have time later.

View attachment 2756626 View attachment 2756646 View attachment 2756669

you can try edit model for consistency

 

[viscountsheep]

you can try edit model for consistency

that's only available for comfyui right? im on forge
i'll have to set up comfy on cloud.. and it's a bit bloated. that'll eat up half the avaliable gpu time just setting it up,

im just trying to do a proof of concept here, so you guys with better hardware can make something better.

 

[jhn]

i ended up testing some stuff. best bet for consistency in anime illustration style is illustrious still.
amina offers way more freedom in description but it's very difficult to recreate an original character in a particular styling which i think is not a bad tradeoff.
krea2 looks seriously promising but the model is large and it's crashing my inference constantly.

good news, i think i have an idea on how to make things look coherent with some post processing and corrections in krita if need be.
bad news, my system fans are screaming all the time.
the window i can run these models are a bit short because im running cloud inferences since my gtx 1650 aint gonna cut it.

but YOU CAN get it done. id say it's moderately difficult still with ai generated images, it does save time in the sense that you already get an almost finished panel. but prompting is a bit tricky still for original characters and original style.

will try training some lora for custom characters when i have time later.

View attachment 2756626 View attachment 2756646 View attachment 2756669

Whats the pricing like

 

[viscountsheep]

Whats the pricing like

well.. free..

if you don't mind the time spent on waiting for the quota, im on colab free tier. you can get like an hour a day on it.
i'd recommend saving those quotas for 3-4 days. you can get upto 6 hours tops.

it typically takes 10 to 15 minutes setting things up and getting things ready each run. so trying to use it daily on the free plan is not ideal. you lose 15 minutes every run.

apart from that the gpu meter is on time not usage so idling for an hour and continuous generation through an hour are pretty much the same. so you might feel like you're in a rush and might not have time to think things through.

paid plans are more flexible the lowest being around 12 bucks a month

**cries in broke**

 

[mihaluck]

that's only available for comfyui right? im on forge
i'll have to set up comfy on cloud.. and it's a bit bloated. that'll eat up half the avaliable gpu time just setting it up,

im just trying to do a proof of concept here, so you guys with better hardware can make something better.

just install local comfy portable