https://t.co/0eeqjLp8Js

0x ROAS on X: "here's how to make hyper realistic images in 2026 (full breakdown)" / X

0x ROAS

@0xROAS

$

/$

here's how to make hyper realistic images in 2026 (full breakdown)

2 min read

if you're still generating bad character images in 2026, you're ngmi.

Higgsfield Soul 2.0 (not sponsored) is one of the main models i use for start frame or character generation only, especially for ai ugc ads.

it's one of the best models out there that i have tried and the difference is day and night with gpt image 2.6.

i will show you how to do this literally in 2 mins, without yapping, without making you subscribe to their yearly plan...

p.s. if you see an AI GURU doing a promoted post on X, you should know that they aren't making money with their methods lol.

Step 1: Find a reference image

as always we need a good reference image in order to generate a good image.

this is the most basic step you should know.

find the reference image on pinterest, tiktok, ads library or whatever...

Soul 2.0 doesnt take reference characteristics so don't worry about creator likeness.

Step 2: Create the prompt for the image

after we have the reference, we will go to claude/gpt or whatever you use...

and we will use my skill for generating prompts for realistic images (free)

Grab the Prompts Here

after you have the prompts, upload them to chatgpt..

and then say "prompt"

Step 3: Go to Open Higgsfield Soul 2.0

We will use soul 2 image-to-image model for generating our character.

i got this prompt from one of my generations.

A photorealistic vertical 9:16 iPhone UGC video frame of a beautiful woman approximately 28 to 33 years old, sitting directly in front of the camera inside a modern but ordinary apartment kitchen. She has long, thick, dark chocolate brown hair with a natural center part, styled in loose, soft waves falling over both shoulders, warm light olive skin, brown almond shaped eyes, naturally thick eyebrows, soft rosy cheeks, subtle pink lipstick, and realistic facial texture. She wears a white satin camisole with black polka dots and delicate black floral lace trimming around the neckline, small gold hoop earrings, a delicate gold necklace, and a thin gold bracelet. She holds the tiny inline microphone of a pair of white wired earphones close to her mouth with her right hand, casually using it as a microphone while talking. The white earphone cable hangs naturally down across her chest. Her left hand is extended toward the camera in the middle of an expressive conversational gesture, slightly blurred by natural hand movement. She is captured mid sentence with her lips slightly parted, looking directly into the camera with an engaged, confident, friendly expression, as though explaining something to her TikTok audience. The framing is a vertical 9:16 medium close shot, straight on at eye level, with her upper torso visible, her face centrally positioned, and generous visible space above her head. The background is a realistic contemporary apartment kitchen with matte white upper cabinets, a dark microwave and oven on the right, a pale countertop with ordinary kitchen objects, off white walls, rectangular ceiling ventilation grilles, and a slightly shadowed kitchen area extending behind her. Strong natural daylight enters from a window directly in front of her, illuminating her face evenly with slightly bright smartphone exposure, gentle highlights on her forehead and cheeks, and natural shadows in the background. Authentic front facing iPhone camera perspective, approximately 26mm equivalent focal length, realistic smartphone sharpness, deep depth of field, natural skin pores, believable hair strands, accurate hand anatomy, slight foreground motion blur, unretouched everyday texture, and natural white balance. The composition should resemble a genuine paused frame from an amateur TikTok talking video recorded at home. No cinematic lighting, no studio equipment, no artificial background blur, no beauty filters, no overly smooth plastic skin, no professional advertising aesthetic, no captions, no subtitles, no text overlays, no watermarks, and no visible recording device.

i uploaded the reference image and my prompt, and this is the result i got:

Step 4: Use GPT Image 2 for changes

We will only use higgsfield soul 2 for first frame or character reference only, all the other changes will be done with gpt image 2 or any other model that you want.

why do it this way and not just use gpt?

gpt image 2 is great for locking the character with image to image

with the soul and gpt combo it keeps the realism of the soul character but now places them in other rooms and things with the realism that gpt gives (soul is not that great for other environments and details but GREAT AT FACES

Yeah it's this easy.

Congrats, you just found the cheatcode to generate ultra realistic character.

If you want to learn the best AI workflows and get paid for your videos, make sure you join the AI ADS community.

1000+ ecom owners, big affiliate teams, job posts everyday, literally the best community in the AI ADS space.

You're literally missing out on the best skill of 2026 and 2027...

It will go through a crazy update with the latest modules I'm using to scale ads on Facebook.

everything’s high risk if you’re a p*ssy. 👇🏻 Join the #1 AI ADS Community in the world 🏆 More than $1B+ in ad spend using my workflows

More Articles

Join 60.7K followers

4:57 PM · Nov 14, 2026

·

4,576

Views

2

61

119