You've got to push AI into the weird places, because that's where the interesting stuff is
@fofrai
-
Replicate seeks early API access from xAI
By
–
Replicate would love to get early access to an API Who @xai is best to talk to about this?
-

Runway Gen-4 Best Model for Identity Preservation Video
By
–
A few folks have asked me which model is best at preserving identity. If I had to pick one, purely on character consistency, I'd go with Runway's Gen-4. https://
replicate.com/runwayml/gen4-
image
… Close up outputs are accurate and you can prompt many different scenes. Unlike kontext-pro you don't -
Training and Running Fast Flux Models with LoRAs
By
–
You can train and then run from here https://
replicate.com/replicate/fast
-flux-kontext-trainer/train
… Or do you mean loras you have already trained? -

Fine-tuned LoRA Model Transforms Faces Into Character Style
By
–
I trained a kontext lora on my 0_1 webp character. The aim was to see if I could make it change any person in an input image into 0_1. https://
replicate.com/fofr/kontext-0
_1-webp
… It kind of works. It's very hit or miss. Forcing an aspect ratio change makes it more likely to work. -

New Fast 2MP Image Generation Model Released
By
–
New and fast 2MP model incoming https://
replicate.com/prunaai/wan-im
age
… -

GPT-Image-1 now supports high fidelity input parameter
By
–
`input_fidelity=high` is now supported on the gpt-image-1 Replicate modelhttps://t.co/pGtAuukhPc
— fofr (@fofrAI) 18 juillet 2025
(This is a bring your own OpenAI token model) https://t.co/fH3xOs8rBV pic.twitter.com/ksNgbi0Irw`input_fidelity=high` is now supported on the gpt-image-1 Replicate model https://
replicate.com/openai/gpt-ima
ge-1
… (This is a bring your own OpenAI token model) -
Private Model Creation as Primary Training Destination
By
–
Create a private model, make that your training destination.
-

Fast LoRA Training and Hugging Face Export Guide
By
–
Kontext training: 1,000 steps, ~3m50, costs about $2.80 – get a fast always-on Replicate model to run your lora
– export directly to Hugging Face -

FLUX Kontext LoRA Training Now Available on Replicate
By
–
You can now train FLUX Kontext loras on Replicate. https://
replicate.com/replicate/fast
-flux-kontext-trainer/train
… – A 1,000 step training takes about 4 minutes
– Train with image pairs, a start and end image
– Write a short prompt instruction that takes you from the start to the end
– Use about 10 image pairs