Great result. I'm guessing it mostly changes the location to a convention because that's what that source photos had? I wonder how hard it would be to change that result with data descriptions.

Just wow. Many thanks for this!

What do you use to train it? I have so many flux kontext dev lora datasets I want to migrate over

you are good job.

Phenomenal results 👏

It manages cases of unrealistic perspectives, shapes, hairstyles, etc. surprisingly well where FluxKontext fails.
My only critique would be the LoRA's tendency to default to asian characters.

Thank you very much for creating this! 🙌

I'm not sure if it's an illusion, but all the characters have similar facial features, and they all look like the same person.

你好，请问你用什么训练器训练Qwen-Edit的lora模型？

How to save an image with a transition - animation "was-became"?

why does it generate random people in my background, how do I make sure that there is no one in the background? Is this configured in the "Text multiline" node?

Dunno if it's only me but it seems like latest Qwen-image-Edit 2509 update has broken your amazing LoRA T____T

This is absolutely brilliant! Thank you so, so much. I tried it out and I'm incredibly impressed and happy with the results. Seriously, thank you! 🙏

This LoRA is AWESOME!

request update for 2509 pwease!

I just added a review comparing results. I used it in Qwen Edit Plus (2509) and liked how well it worked, though it had some issues. Impressive considering I didn't input anime, but American comics instead. Tip: Combine this with the Samsung lora at 1 or .9 strength:

https://civitai.com/models/1551668/samsungcam-ultrareal

Or the Lenovo lora:

https://civitai.com/models/1662740/lenovo-ultrareal

and the results are boosted even more. It still has trouble converting drawn background objects, but if I describe the whole scene, and emphasize a lot that it's a photo, this lora does a great job. Without the lora, it just recreates the scene I describe vaguely, but doesn't line up with the original image. Here's an example prompt I had to use, to give an idea of how descriptive I had to get, and even with this, it still treats many background objects as 2D art, but with photograph coloring/shading, so it doesn't read as much as 2D art:

Positive:
transform into realistic photography, and remove the word bubbles. It will be a real photo of a woman standing in a city. The woman is wearing a leather jacket over a tied shirt exposing her midriff. She has her hands in her jacket pockets. She is a little angry, with a closed mouth, and is giving a stern look while glancing to the right. Her hair is in a ponytail and is blowing in the wind. She has a short buckskin skirt and metal belt. She's wearing shiny snakeskin boots, which are reflecting the light from the diner around their edges.

She is in front of a 40-year-old man who has a toothpick in his mouth, leaning back against something off-screen, and is smirking with a closed mouth while leering at her. Behind him is an 18-wheeler truck with a partially covered sign that says "HAUL ASS", but only the letters "HA AS" are visible. The truck is blue, with realistic textures from having driven long distances.

Also behind them, some people are reaching their arms out to greet each other happily. They are in front of a diner called "Truk Stop Restaurant and Grille", which has a well-lit interior, and a realistic bright sign. Above the diner is a highway bridge, with a street lamp, and trucks/cars driving over the bridge at night. The photo will be in the style of a professional photograph, capturing the detailed and gritty realism of the city. All the characters will be real people, and not drawings. It will look like a real professional photo by a famous photographer.

Negative (not necessary, but just in case):
line art, drawings, 2D, comic art, illustration, blurry, low quality, low resolution, bad anatomy, deformed, distorted, extra limbs, missing limbs, extra fingers, fused fingers, text, watermark, logo, artifacts, overexposed, underexposed, grainy

Does this Lora have an Asian bias? Or can it also do other races, too? Like Caucasian, etc. Asking because all of the examples show an Asian bias, which wouldn't be useful for me particularly. If so do you have particupar prompt advice for achieving it? Either way, cool Lora.

Hey, any kontext version for this?

Incredible work, but with a fatal flaw that needs fixing.
eyes are always looking at the camera
even if you give it already perfect 3d image to enhance it still resets eyes to the camera for no reason at all. even at 0.1 cfg value.
similar problem with mouth but at lest not that important

I second the request for a 2509 update. This is a great lora

Omg this is the absolute BEST transformation to realism LoRa I have tested so far!! Better then any other ones currently out there in my Opinion. Took me a bit to realize that CFG = 4 even with the Qwen lightning LoRa on and I found it works even a little bit better (shadows and details) when using the new “Unchained” NSFW LoRa with it as well. Keep CFG at 4 and 9 steps with basic euler and simple 👌 Great work on this LoRa!!

i know everyone loves their low fidelity fast gen time lightning but yo, make a non-lightning version!