Runchat Updates: August 16, 2025

🚄 LOW COST FLUX LORA TRAINING

• We've added a new endpoint for flux training that takes a bit longer (5 minutes) but is 1/4 of the price. So go train more LoRAs (and share them here...)
• This is experimental, let us know if you notice any significant difference in quality to the fast training model

🎭 MASKED CONTROL NETS IN FLUX

• You can now provide a mask as well as a control image to the Flux Control model
• This gives you much more control over formal variation than straight control nets

♊ BETTER NODE LAYOUTS

• Nodes that have multiple image inputs now show these inputs side by side to make it easier to compare them

🪲 MINOR BUG FIXES

• Edge control was broken on the SDXL IPAdapter model, this is now fixed. You can use edge + depth control with this model.
• Image size was broken on the Flux Control model, this is now fixed