Workflow img2img QWEN 2.1 Image Edit Discussion Thread

R

ReelChirp

Lurker
Joined
Mar 19, 2022
Posts
29
Reaction score
0
For all your discussions concerning the qwen 2.1 model, available as a template in Comfy desktop. Brief synopsis of qwen: Unified text (you write in what you want to edit) and image editing model accepts up to 10 reference images. The example use case is outfit swapping. Create a dummy email and give it a try for free online on comfy's cloud service.



In this thread discuss:
Loras for qwen 2.1 image edit
Prompts
Results (no real people)

Ask questions and help each other out.

If there's already a dedicated qwen 2.1 discussion thread, share the link please. All I saw was a general prompt discussion thread and I'm sure there's overlap but each model behaves slightly differently from each other. A prompt for one image gen model is NOT written the same way a prompt is written for Qwen and the exact same prompt can behave very differently depending on the model used. In the same line of thinking, a Lora for one image generation model may or may not work for a qwen model. That's why I made this thread
 
Original/reference : Image 2: Result:

Subject defintion generated on grok. Upload the original image and ask grok to create a subject definition for qwen 2.1 based on reference picture. I've removed a few personal modifiers but I suggest you guys generate your own subject definition for best results.

Prompt (positive_prompt):
subject 1: a woman in with fair skin, medium-length soft wavy hair that is black at the roots fading into vibrant deep red ends, natural soft makeup with defined brows, long lashes, glossy nude-pink lips, and a calm slightly pouty expression, wearing multiple layered silver necklaces including a nameplate pendant and a long dagger pendant, small silver hoop earring, and several silver rings.

Preserve framing, background, angle, perspective, and lighting. Remove gray sweater that is partially covering shoulder. Replace outfit with black lace lingerie with matching collar shown in . sharp details, 4k, high quality.

The part in italics is probably the only parts you should copy paste. The rest of the prompt is specific to the original image so wouldn't work for your input.

technicals
cfg: 1 (was told to use 1 if not using any negative prompts)
Sampler: Euler
Scheduler: Simple
custom_size: disabled
refine_prompt+thinking: disabled
 
There's a big problem in that there isn't any decent NSFW Lora engine like Realism Engine or Snofs. Without that, it's pretty much useless.
 
You must log in or register to reply here.
Back
Top