cinematic, film grain, RAW, movie still, realistic photography, dutch angle style, chromatic aberration
License Scope:
Creative License Scope
Online Image Generation
Merge
Allow Downloads
Commercial License Scope
Sale or Commercial Use of Generated Images
Resale of Models or Their Sale After Merging
Model Parameters:
Base Model:
SDXL 1.0
Epochs:
0
Iteration Steps:
0
Clip Skip:
0
Review:
0
Review
__Chinook__V1.0 MASSIVE Update !!
It's a model that I'm trying to optimise for cinematic rendering (1536*640) but of course you can use it in the classic SDXL sizes. It's the first version, so I've still got a lot of work to do on it, but the tone is set and frankly I'm really proud of it!
As it's a heavy mix, I finetuned it from low CFG (1.5-3) so that the noise would come up with more quality and detail. It also has a wide spectrum and needs to be tested assiduously!
1-5 -> 3: More creative, less photographic quality.
3->5 : Good compromise
6-> 8 : More photographic and a little blurry for the cinematic side.
For the sampler, I took the classic DPM++ 2M Karras, perhaps there's something better? I'll leave you to explore. For the Steps, between 10 and 100 and fixed hires, if you like!
Use of Adetailer sometimes necessary.
I use different VAEs for my renders, which is why I advise you to take the version WITHOUT a VAE and to add your own depending on your needs. The most incredible VAEs all come from and I encourage you to discover them all! The images in the gallery use them extensively.
For the prompt, in works equally well as a literal version (GPT/Claude text for example), or as words separated by commas.
So the basis of a prompt would be, for example :
Positive prompt: cinematic, film grain, RAW, movie still, realistic photography, [SUBJECT], [LIGHTING], dutch angle style, chromatic aberration
There are some STRONG trigger words: dutch angle style, chromatic aberration, movie style, analog,... which are very important for the cinematic side I wanted. I also worked on the notions of crowds and variety of faces.
This model isn't perfect! There are still some obvious finger and eye problems and as I've mixed a lot of Loras designed for 1024x1024 or portrait style, I have to struggle with the fact that it can centre the subject (but I've worked a lot on this aspect!). In short, there's still a lot of work to do, so please be patient!
__Chinook__V1.0 MASSIVE Update !!
It's a model that I'm trying to optimise for cinematic rendering (1536*640) but of course you can use it in the classic SDXL sizes. It's the first version, so I've still got a lot of work to do on it, but the tone is set and frankly I'm really proud of it!
As it's a heavy mix, I finetuned it from low CFG (1.5-3) so that the noise would come up with more quality and detail. It also has a wide spectrum and needs to be tested assiduously!
1-5 -> 3: More creative, less photographic quality.
3->5 : Good compromise
6-> 8 : More photographic and a little blurry for the cinematic side.
For the sampler, I took the classic DPM++ 2M Karras, perhaps there's something better? I'll leave you to explore. For the Steps, between 10 and 100 and fixed hires, if you like!
Use of Adetailer sometimes necessary.
I use different VAEs for my renders, which is why I advise you to take the version WITHOUT a VAE and to add your own depending on your needs. The most incredible VAEs all come from and I encourage you to discover them all! The images in the gallery use them extensively.
For the prompt, in works equally well as a literal version (GPT/Claude text for example), or as words separated by commas.
So the basis of a prompt would be, for example :
Positive prompt: cinematic, film grain, RAW, movie still, realistic photography, [SUBJECT], [LIGHTING], dutch angle style, chromatic aberration
negative prompt: drawing, illustration, 3d, sketch, cg
There are some STRONG trigger words: dutch angle style, chromatic aberration, movie style, analog,... which are very important for the cinematic side I wanted. I also worked on the notions of crowds and variety of faces.
This model isn't perfect! There are still some obvious finger and eye problems and as I've mixed a lot of Loras designed for 1024x1024 or portrait style, I have to struggle with the fact that it can centre the subject (but I've worked a lot on this aspect!). In short, there's still a lot of work to do, so please be patient!