SDR to HDR from ComfyUI, I've trained a LoRA over Qwen Edit 2011 based on the principle used in A research I've had the previlige to work on with Naomi Ken Korem for video HDR on LTX. Links for the LoRA and workflow (+ my fun grading node) below.

Github : LoRA and workflow : LogC3 decode Custome node (this node is also available already in LTXV nodes) :

@noamiKenKorem Do you think this could be done with Klein as well? 🤔

@noamiKenKorem I have a wip lora done in klein 4b, behaves a bit differently, but we do get range, I don't think ill have time for 9 iterations like i did with qe11 so I might share that version as is :D

@noamiKenKorem I wanted to try this for myself, so if you need a hand with the 9B version, I'm all in.

@noamiKenKorem Sure, you got compute and time to test? Let's talk in discord

@Nekodificador @noamiKenKorem How much compute do we need for Klein 9b? Does a 3090 24gb work? If so, I have an isolated machine I can leave running it for days.

@Nekodificador @noamiKenKorem I'm releasing a klein 4b and 9b version, we've done some initial tests with @Nekodificador it's an alpha version, but it should give you a decent range already.

@noamiKenKorem Very cool Mohamed!

@noamiKenKorem It's quite impressive. Could you briefly explain the training process and underlying principles?

@noamiKenKorem it's explained in detail in the paper here : In summary: take existing HDRs, encode them to LogC3, train the model on that, the model generates logC3, decode that back into HDR. That's it. The model doesn't even know it's learning HDR.

@AIAVJLeo @noamiKenKorem Would it be possible to just output logC3 images instead of HDRs?

@AIAVJLeo @noamiKenKorem It is what it does, I mean what it is trained for, you can just grab the image before the logc3 decode.

@noamiKenKorem Yooooo! This is AMAZING!!!

@noamiKenKorem Amazing work!

@noamiKenKorem that's 🔥 - check it out @dpreview

@ComfyUI @noamiKenKorem That’s fine but 2k max resolution 😭 maybe you know the way how to get ability generate 4-8k HDRs ?

@AIlover1995 @ComfyUI @noamiKenKorem Why would you be limited to 2k ? There are plenty of ways yo get much higher even if it didn't work natively. I've also been training Klein for this same task, which might be easier for doing 4k.

@ComfyUI @noamiKenKorem Yeah. Digging into it RN, thanks for your work!

@noamiKenKorem A powerful node for a powerful tool. Congrats!

@noamiKenKorem Very nice. I hope banding issue wont be a problem now. Although i used to fix it manually

@noamiKenKorem Cool, that looks nice! I like SDR to reverse tone mapping - easy and effective. But this is incredible, too.

@noamiKenKorem Very very cool! Is the SDR input image supposed to be sRGB or LogC colorspace?

@noamiKenKorem Input SDR is sRGB/rec.709

@noamiKenKorem Awesome, thanks!

@noamiKenKorem This is awesome—SDR→HDR via ComfyUI + a Qwen Edit 2011 LoRA sounds like a really clever pipeline, and the “fun grading node” made me smile 😄. If you end up comparing API plans, OpenStacks ( is handy.

@noamiKenKorem Spatial Raw

@ComfyUI @noamiKenKorem Awesome

@noamiKenKorem Hello, can you share this Lora?

@noamiKenKorem

@noamiKenKorem Amazing!

@noamiKenKorem Kicking ass like always! This is great!