What type of camera or filter or anything do you use for your image? it looks weirdly pixelated but also crisp at the same time, the black and white gives it an overall artistic style to it! i realy like it. insanely actually.
The t5 models are configured slightly differently - schnell inputs have a max length of 256 tokens and dev have a max length of 512 - but everything else (weights, code) is the same. See these links: huggingface.co/black-forest-labs/FLUX.1-schnell/blob/main/text_encoder_2/config.json huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/text_encoder_2/config.json
@@replicatehq Strange ...did you set for flux dev a guidance to 0.5-1.0 for pictures ? I'm getting much better results with pictures like that for flux dev ( I'm using a comfy UI)
Thanks! Struggling through this myself, always welcome to see a breakdown how others gone about solving it
what mechanical keyboard are you using? it sounds nice.
EPOMAKER CIDOO V75 VIA - www.amazon.com/dp/B0C23GB1G6?psc=1&ref=product_details
What type of camera or filter or anything do you use for your image? it looks weirdly pixelated but also crisp at the same time, the black and white gives it an overall artistic style to it! i realy like it. insanely actually.
OBS and iPhone 13. High contrast. Zero saturation. El Gato green screen. 🤷♂
Very nice comparison. The answer is probably no, but would these work sith SD/SDXL LoRAs?
Highly unlikely that an SD/SDXL LoRA would work with FLUX because of architectural differences
@@lucataco I see. Thx for the answer.
What's the difference between the two, in terms of the actual model?
Dev is trained using guidance distillation, Schnell is smaller and was trained using latent adversarial diffusion distillation
Pencil drawings are very bad example for showing difference. I didnt understand how it will look in realistic.
in replicate flux schnell api any limit or we can use number of time?
Dev.1 is far better, people actually look real where as Schnell still looks cartoon like .
Sometimes Flux Schnell should stop putting lightning in every creation 😅😅
tanks and you can use the real esgan in comfyui locally even with slow cpu
with both did you use the same t5xx?
The t5 models are configured slightly differently - schnell inputs have a max length of 256 tokens and dev have a max length of 512 - but everything else (weights, code) is the same.
See these links:
huggingface.co/black-forest-labs/FLUX.1-schnell/blob/main/text_encoder_2/config.json
huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/text_encoder_2/config.json
@@replicatehq Strange ...did you set for flux dev a guidance to 0.5-1.0 for pictures ?
I'm getting much better results with pictures like that for flux dev ( I'm using a comfy UI)
720p on 2024 and everything is white, what the hell man!