Breaking: Fooocus just added a killer new upscaler feature.

chloek4 · 2026-07-21T19:37:32Z

Okay, I just updated to the latest Fooocus nightly and the new upscaler is a game-changer for my automation workflows. It's not just another model swap—they've integrated it directly into the "Upscale & Variation" section with a dedicated "Upscale (Step 2)" button and, crucially, **API-accessible parameters**. This is huge for reliability. Before, I'd generate an image in Fooocus, send it to another service via webhook for upscaling, and then handle the return trip. That's two external calls with potential points of failure. Now, I can potentially do it all in one Zapier/Make task if the API supports it. A few things I'm digging into: * **The new parameters** in `api_flask.py` look promising. I'm seeing `uov_upscaler_2` and `upscale_value_2`. The ability to trigger this second-step upscale programmatically is what I care about. * **Webhook potential:** If the API call can return both the initial image and the upscaled one in a single response, that simplifies my payload structures massively. * **Error handling:** I'm curious if a failure in this "Step 2" upscale fails the whole job or just returns the first-step image. That's critical for building a fault-tolerant pipeline. Has anyone tried calling this via the REST API yet? I'm about to run some tests with Postman. My main questions are: * What's the exact JSON structure for the new params? * Does it work with all the existing `uov_method` types (like VAE*)? * Any noticeable increase in timeouts we should account for in our zap steps? I'll post my API findings here. If this works as cleanly as it looks in the UI, it might just replace a whole Make scenario I have running. chloe

Page 2 / 2 Prev

Stable Diffusion Reviews

Last Post by Daniel M. 2 months ago

16 Posts

15 Users

0 Reactions

80 Views

RSS

Daniel M.

(@danielm)

Honorable Member

Joined: 2 months ago

Posts: 453

27/07/2026 12:18 pm

I don't think you can call this a reliability win until you test that error handling scenario. You've just moved your failure point, not removed it.

That single API response you're hoping for is also a single point of total failure. If the upscaler OOMs, you're not getting *any* image back. Your "simplified payload" becomes an empty 500 error page after burning minutes of GPU time.

The real question isn't if the parameters exist, it's whether the system does any validation on `upscale_value_2` *before* it commits to the initial generation. My money's on "no".

— skeptical but fair

ReplyQuote

Page 2 / 2 Prev