Will z-image be great at comics?
So far, i saw people praising z-image turbo for comics capabilities, but i tries many many times and it is just a flop. I even reduced to 4 panel only. And my prompts real short. I also changed my top 4 sheduler+sampler combination and turbo version just not good enough.. sadly! Will z-image be more capable? i really hope so.
So far, i saw people praising z-image turbo for comics capabilities, but i tries many many times and it is just a flop. I even reduced to 4 panel only. And my prompts real short. I also changed my top 4 sheduler+sampler combination and turbo version just not good enough.. sadly! Will z-image be more capable? i really hope so.
https://zimageturbo.org/how-to-generate-comics-using-z-image
So far, i saw people praising z-image turbo for comics capabilities, but i tries many many times and it is just a flop. I even reduced to 4 panel only. And my prompts real short. I also changed my top 4 sheduler+sampler combination and turbo version just not good enough.. sadly! Will z-image be more capable? i really hope so.
https://zimageturbo.org/how-to-generate-comics-using-z-image
Im gone arguing with uneducated. You think i dont know this? You think I didnt used this as a template BEFORE posting this thread? Did you think anything at at all? Did you yourself made many comics? Show them! NO you did no such things. Did you tested anything? no you did not. Did you at least read that i wrote about me testing many many times? no you didnt read because you just gave one link thinking you are smartass.
@gemstonebro
wow, what a level of violence. If you did what you claim you did, then say it as a context in your question. How do you want people to guess?
Just share what you ve tried already and then ask your question.
There is already too much violence in this world to add some in those threads, please.
@gemstonebro wow, what a level of violence. If you did what you claim you did, then say it as a context in your question. How do you want people to guess?
Just share what you ve tried already and then ask your question.
There is already too much violence in this world to add some in those threads, please.
I just interpreted that way.. nobody just "gives link" without saying a word.. no normal person does such things. I got no such "logs" to give, but i did ~ 10 prompts each should had ~7panels and then i reduced prompt into 4 panels and it was still quite messy and not reliable enough. I was switching between euler simple and euler beta and ddim dgm uniform or something similar.. anyway, maybe im expecting too much out of turbo version. Did you had any success with comics?
@gemstonebro wow, what a level of violence. If you did what you claim you did, then say it as a context in your question. How do you want people to guess?
Just share what you ve tried already and then ask your question.
There is already too much violence in this world to add some in those threads, please.I just interpreted that way.. nobody just "gives link" without saying a word.. no normal person does such things. I got no such "logs" to give, but i did ~ 10 prompts each should had ~7panels and then i reduced prompt into 4 panels and it was still quite messy and not reliable enough. I was switching between euler simple and euler beta and ddim dgm uniform or something similar.. anyway, maybe im expecting too much out of turbo version. Did you had any success with comics?
You need to relax my dude I was trying to be helpful but with your attitude you need a different type of help.
So far, i saw people praising z-image turbo for comics capabilities, but i tries many many times and it is just a flop. I even reduced to 4 panel only. And my prompts real short. I also changed my top 4 sheduler+sampler combination and turbo version just not good enough.. sadly! Will z-image be more capable? i really hope so.
https://zimageturbo.org/how-to-generate-comics-using-z-image
Btw I did follow this interesting blog post (of interest to me at least), and for those interested I share some results below (8 different seed, for each of the 2 pages generated in the blog post).
The ability to follow the desired page layout is quite impressive, most of the details asked are followed. Now, many details not asked for are there too, so it would be good to have negative prompts as a possibility, e.g. "no chinese text" here. But promising results I would say, especially for 10secs per image o my ageing rtx3090 ....
@tetsfr thanks for the share. am currently building a tool that will enable people to create comics and share. can you help me with some doubts, am new to this? I built a full stack app for different features and comic gen is one of it. This post came up when I was looking if I should go about hosting my own workflow on run pod or just call gemini api for the jobs.
@tetsfr thanks for the share. am currently building a tool that will enable people to create comics and share. can you help me with some doubts, am new to this? I built a full stack app for different features and comic gen is one of it. This post came up when I was looking if I should go about hosting my own workflow on run pod or just call gemini api for the jobs.
Sorry but I am no comics specialist, I ve just done a quick check here really. The model is promising but pls note that if you look at the prompts from the blog carefully you ll realise not everything is generated consistently, so it woukd deserve so more experimentation on many comics use cases I think. Another limitation is consistency of characters, this model does not deal with this which I guess would be an issue for your use case. In my view it is still better to gen comics images one at a time with powerful control on the character consistency through, and then assemble those images into pages layouts at the end through more classic layout editing. I think there quite some work involved here.
Regarding the infrastructure, I ve never use runpod but looks like a decent option even if probably a bit pricey.