FLUX KONTEXT + PULID (EXPERIMENTAL)

You can get "out of memory" error when you run it on both first sampler and second sampler,if you do just RE-RUN the workflow until it doesn't give OOM.It is designed to run it at 6GB vram.

---Device Error-----

Since this is an experimental,you can get this error with pulid, my comfyui version is : v0.3.43. For me it works.

The primitive which is downsampling factor is important,adjusts how consistent is the reference load image. So if you make it closer to 1 it makes it more identical to original, if you make it closer to 5 it makes more prompt following. My recommended value is between 3 to 4 or even 5.

Extra Notes Recommended to read :

I get 80% to 93% face consistency on some website results.

Do NOT upload image like too much ZOOM in to the face, upload image where the character is medium portrait, not too close to the camera and not too far. Like the reference images i uploaded on OPENART.

Highres Group which is "second sampler" is for making face more detailed and increase the consistency of it.I suggest you to wait and use it. BUT if you make like close portrait shot maybe first sampler will be enough.

Do not try to make full body shots, it is designed to make portrait type.

The Flux kontext is not for generating images, it is designed to "EDIT" the images. Making whole new image is a different task from my experience.

So with this kontext, i think the images has been fed "GPT like" images, not like FLUX realism.So there is a problem occurs, it is become unrealistic and plastic.To solve it we need a good lora but we dont have that (for now).

So i use Flux Dev loras on this workflow. Does it work? Yes but it makes image more blurry if you increase the strength. My recommend is to use it at max value of 1. BTW if you do not use lora the output image face will have less blur,BUT plastic looks come again.

As for this task i am using only one lora removes the plastic look better for me. I am giving you the link.

LORA (DO NOT FORGET KEYWORD ON PROMPT):

https://huggingface.co/prithivMLmods/Canopus-LoRA-Flux-UltraRealism-2.0/tree/main

The keyword for the lora : Ultra realistic

I am looking forward to hear your experiments on this WF.

If you liked this, i appreciate a sub on my YOUTUBE : https://www.youtube.com/@AITold/videos

Example Structures of the prompt that works well on this workflow :

1- An Ultra realistic, full-body shot of the stunningly beautiful woman with captivating hazel-green eyes and a constellation of natural freckles, captured in the candid, lonely style of amateur photography. She is sitting alone in a worn-out, red vinyl booth in a dingy 24-hour diner at night. The camera is positioned across the table, slightly crooked, as if taken by a companion. She is dressed in a simple, oversized gray hoodie and comfortable jeans. Her delicate, beige crocheted net shawl is slung over the back of the booth, looking out of place. With both hands, she cradles a thick ceramic mug of black coffee, her expression one of thoughtful melancholy as she stares out the large window at the rain-streaked streetlights. The table is cluttered with a crumpled napkin and a half-eaten piece of pie. The lighting is a terrible mix of the warm, yellow glow from a tabletop jukebox and the cool, blueish fluorescent light from the diner's main ceiling, creating multiple, clashing shadows. The photo is slightly grainy from the low light. The atmosphere is one of quiet, late-night contemplation.

2-An Ultra realistic, medium shot presented as a grainy, pixelated screenshot of a late-night video call. The stunningly beautiful woman with captivating hazel-green eyes and a constellation of natural freckles is sitting at her desk in a dimly lit room. She is looking directly into her webcam, and therefore at the viewer, with a tired but focused and engaged expression. The only illumination comes from the cool, blue-white light of her computer monitor, which casts a stark, digital glow on her face and creates deep shadows. She wears a comfortable, loose-fitting t-shirt, and her delicate, beige crocheted net shawl is draped over the back of her chair, visible as a soft shape in the poorly lit background. The background is a blurry, authentic mess of her room: the corner of a bookshelf, a poster on the wall, and general clutter. The image quality is intentionally low, with visible compression artifacts and digital noise, perfectly capturing the "amateur" webcam aesthetic.

3-An Ultra realistic portrait of the stunningly beautiful woman with captivating hazel-green eyes and a constellation of natural freckles, captured in the candid, awkward style of amateur photography inside a brightly lit, sterile 24-hour convenience store at night. The photo is taken from a low angle, as if by a friend who has just caught her off guard. The primary light source is the harsh, green-tinted overhead fluorescent lighting, which creates deep, unflattering shadows under her eyes and chin and makes her skin look pale. She is standing in an aisle, holding a bag of chips in one hand and a soda in the other, looking up at the camera with a slightly annoyed, "Are you seriously taking a picture right now?" expression. Her delicate, beige crocheted net shawl is worn loosely over a simple t-shirt, looking completely out of place against the backdrop of brightly colored, branded snack packaging. Her signature intricate, reddish-orange beaded choker and gold pendant clash with the mundane setting. The composition is off-center, and the colors are slightly washed out due to the terrible lighting. The atmosphere is one of jarring, mundane reality.

4-An Ultra realistic, medium shot of the stunningly beautiful woman with captivating hazel-green eyes and a constellation of natural freckles, captured in the candid, intrusive style of amateur photography. She is sitting on a park bench, and has just been interrupted while reading. She has looked up from her book, which is still in her lap, and is staring directly at the person holding the camera with an expression of mild, questioning annoyance. The bright, harsh midday sun filters through the leaves of a tree above, creating a distracting, dappled pattern of bright hotspots and deep shadows across her face and clothing. She wears a simple sundress, and her delicate, beige crocheted net shawl lies beside her on the bench. The camera's focus is slightly soft, and the composition is unbalanced, with too much empty space on one side. The photo captures a genuine, un-posed moment of being interrupted in a public space.

5-An Ultra realistic, medium portrait of the stunningly beautiful woman with captivating hazel-green eyes and a constellation of natural freckles, captured in the chaotic, snapshot style of amateur photography at a crowded outdoor street fair. She is looking directly at the camera, her expression a candid mix of surprise and a faint smile. She is wearing a simple summer dress. The bright, harsh midday sun is directly overhead, creating dark "raccoon" shadows under her eyes. The background is a noisy, distracting, out-of-focus bokeh of other people walking by, colorful vendor tents, and bright sunlight glinting off random objects. The composition is imperfect, with a stranger's shoulder encroaching on the edge of the frame. The photo feels like a genuine, fleeting moment stolen from a loud and busy day.

Knotenpaket · Registry