It's quite remarkable that such a lightweight model can deliver excellent prompt adherence and impressive image quality, all at high speed. I've been able to generate quite coherent images near 2 Mpix.
This model is checking all the boxes at this stage. Furthermore, its small size will open up significant opportunities for finetuning and further improvements.
I am not able to get this working on Comfy cloud even though the workflow is there. No matter what text prompt I type, all I get is the default girl with the bicycle.
Why couldn’t you put them all in one update? And your Mistral Flux.2 small BF16 crashed my 3090. I wasn’t paying attention to the BF16, especially since you bundled it with an FP8 Flux.2
Took me two crashes to figure what was happening, then scoured huggingface for an FP16 Mistral.
This model randomly went viral, it's pretty amazing the quality given how fast it is.
It's quite remarkable that such a lightweight model can deliver excellent prompt adherence and impressive image quality, all at high speed. I've been able to generate quite coherent images near 2 Mpix.
This model is checking all the boxes at this stage. Furthermore, its small size will open up significant opportunities for finetuning and further improvements.
Latina-Image Turbo
Cool, but really slow on 3060 12GB...
I am not able to get this working on Comfy cloud even though the workflow is there. No matter what text prompt I type, all I get is the default girl with the bicycle.
Why couldn’t you put them all in one update? And your Mistral Flux.2 small BF16 crashed my 3090. I wasn’t paying attention to the BF16, especially since you bundled it with an FP8 Flux.2
Took me two crashes to figure what was happening, then scoured huggingface for an FP16 Mistral.
Works perfect on RTX 5000 with ComfyUI, Fast and photorealistic, waiting for the full model to tune it. Great work.
Doesn't work
ERROR AFTER ERROR - That's that.
This is actually pretty good. Instant generation to high-quality images.