Can't get a photo out of this model - only illustrations [ v0.2 turbo ]
All I'm getting is illustrations. I've tried the usual 'A photo of', etc, but I've had one photo in a dozen generations. Not entirely sure what I'm doing wrong.
All I'm getting is illustrations. I've tried the usual 'A photo of', etc, but I've had one photo in a dozen generations. Not entirely sure what I'm doing wrong.
That's so strange bro. I had no problems to generate real life things with this model. The tips I have for you is:
1 - Use its FP8 text encoder.
2 - Use a FP8/FP16 version of the main model (base or turbo).
3 - Use Euler simple as sampler and schedule.
4 - Set the cfg scale strength to "1".
5 - Keep the LoRa strength to "0.6-0.8" (if you're using the v0.1)
6 - Use the node ModelSamplingFlux (it improves the output).
7 - Always use natural prompting (for everything). Only uses tags for aesthetic styles, not for the entire composition.
I'm using the full model from this card, not the lora
All I'm getting is illustrations. I've tried the usual 'A photo of', etc, but I've had one photo in a dozen generations. Not entirely sure what I'm doing wrong.
I just tried and yes, same prompt that on default Krea2 I get photo, with kroma turbo you get animation. I used this lora to fix it for now: https://civitai.red/models/2823820/photodetail-slider-for-snofs-krea
its like slider at 1 strength changes animation into photo.
Lucky you, im doing somthing wrong, im only get eldrich horrors,
Lucky you, im doing somthing wrong, im only get eldrich horrors,
traditional combos works. euler ancestral +simple or normal feels like need 16 steps to cook more detail, res_2s+beta 8steps looks ok.
All I'm getting is illustrations. I've tried the usual 'A photo of', etc, but I've had one photo in a dozen generations. Not entirely sure what I'm doing wrong.
I just tried and yes, same prompt that on default Krea2 I get photo, with kroma turbo you get animation. I used this lora to fix it for now: https://civitai.red/models/2823820/photodetail-slider-for-snofs-krea
its like slider at 1 strength changes animation into photo.
Just getting illustrations as well on the turbo model. Thanks for posting this lora.
I've had much more luck today, using the Canon lora and starting prompts with 'c2n0n. A candid photo taken with a Canon DSLR camera.'