Soul 2

Higgsfield's own image model, and the only image model in this catalog that its maker runs itself. It is tuned for photography of people and the things they wear: skin that keeps its texture, fabric that folds like fabric, light that falls the way a lens would record it. Where most of the catalog is a generalist that can be pushed toward a photograph, this starts there.

It costs half a credit at either of its two sizes, 720p or 1080p, which puts it at the bottom of the catalog beside FLUX Schnell and P-Image. It defaults to 1080p, because at that price the smaller size saves nothing worth having. Seven aspect ratios, from 9:16 to 16:9, with 2:3 and 3:2 for print-shaped frames.

What it does not do is take a reference. There is no tray: the prompt is the whole request, so it is a model for finding a look rather than for holding a subject across a series. When a face or a product has to stay the same from one picture to the next, a model with a reference channel is the one to use, and a Soul 2 frame you like can be that model's first reference.

Describe what you want in your own words and Kuro writes the prompt for you. If you would rather write it yourself, this is what Soul 2 responds to.

Soul 2 is text to image tuned for fashion and lifestyle photography: realistic skin, fabric and light from a prompt alone.

Write it like a photographer's brief — subject, wardrobe and what it is made of, setting, light, lens: "a woman in a linen suit on a Lisbon rooftop, late golden hour, 50mm, shallow depth of field".

The prompt is the whole input, so everything the picture must show has to be in it.

What it can't do

• No references at all. A specific person cannot be held from one picture to the next; a saved Subject contributes its written description only.
• 720p or 1080p.
• Seven aspect ratios: 1:1, 2:3, 3:2, 3:4, 4:3, 9:16 and 16:9.
• Editing one of its pictures moves the edit to a model that can read images.

Use something else when…

• A specific face, product or garment has to match a picture: Nano Banana 2.1.
• It needs legible lettering or graphic design: GPT Image 2.5 Flare.